技术领域technical field
本申请涉及通信技术领域，尤其涉及一种音量调节方法、电子装置及计算机可读存储介质。The present application relates to the technical field of communications, and in particular to a volume adjustment method, an electronic device, and a computer-readable storage medium.
背景技术Background technique
用户通过电子装置收听音频信号时，为了防止环境噪声影响音频信号的收听效果，需要根据环境噪声的音量调整该电子装置输出的音频信号的音量。然而，现有的音量调节方法普遍根据环境噪声的音量调节的音频信号的音量，这种音量调节方法普遍存在无法根据用户的使用习惯调节待输出音频的音量等问题。When a user listens to an audio signal through an electronic device, in order to prevent environmental noise from affecting the listening effect of the audio signal, the volume of the audio signal output by the electronic device needs to be adjusted according to the volume of the environmental noise. However, the existing volume adjustment method generally adjusts the volume of the audio signal according to the volume of the ambient noise, and this volume adjustment method generally has problems such as being unable to adjust the volume of the audio to be output according to the user's usage habits.
发明内容Contents of the invention
本申请提供一种音量调节方法、电子装置及计算机可读存储介质，可以根据用户的使用习惯调节待输出音频的音量。The present application provides a volume adjustment method, an electronic device, and a computer-readable storage medium, which can adjust the volume of audio to be output according to the user's usage habits.
本申请实施例第一方面提供一种音量调节方法，包括：The first aspect of the embodiment of the present application provides a volume adjustment method, including:
获取多个音量样本数据，所述音量样本数据包括环境声音数据和音量设置数据；Acquiring a plurality of volume sample data, the volume sample data including ambient sound data and volume setting data;
将所述音量样本数据输入预测模型进行训练，得到目标音量，所述预测模型的训练结束条件为损失值小于预设的阈值；Inputting the volume sample data into a prediction model for training to obtain a target volume, the training end condition of the prediction model is that the loss value is less than a preset threshold;
将待输出音频的音量调节为所述目标音量。Adjust the volume of the audio to be output to the target volume.
本申请实施例第二方面提供一种电子装置，包括：The second aspect of the embodiment of the present application provides an electronic device, including:
获取模块，用于获取多个音量样本数据，所述音量样本数据包括环境声音数据和音量设置数据；An acquisition module, configured to acquire a plurality of volume sample data, the volume sample data including ambient sound data and volume setting data;
训练模块，用于将所述音量样本数据输入预测模型进行训练，得到目标音量，所述预测模型的训练结束条件为损失值小于预设的阈值；A training module, configured to input the volume sample data into a prediction model for training to obtain a target volume, and the training end condition of the prediction model is that the loss value is less than a preset threshold;
调节模块，用于将待输出音频的音量调节为所述目标音量。An adjustment module, configured to adjust the volume of the audio to be output to the target volume.
本申请实施例第三方面提供一种电子装置，包括存储器，处理器及存储在存储器上并可在处理器上运行的计算机程序，其特征在于，所述处理器执行所述计算机程序时，实现上述本申请实施例第一方面提供的音量调节方法。The third aspect of the embodiment of the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and operable on the processor, wherein when the processor executes the computer program, the The volume adjustment method provided in the first aspect of the above embodiment of the present application.
本申请实施例第四方面提供一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时，实现上述本发明实施例第一方面提供的音量调节方法。A fourth aspect of the embodiments of the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the volume adjustment method provided by the first aspect of the above-mentioned embodiments of the present invention is implemented.
上述本申请各实施例，根据音量样本数据对预测模型进行训练，得到目标音量，预测模型的训练结束条件为损失值小于预设的阈值，所以通过该预测模型训练后得到的目标音量与音量设置数据之间的差的欧几里得范数小于预设的阈值，将待输出音频的音量调节为目标音量，在该音量样本数据中包含用户音量设置数据，因此待输出音频的音量与用户音量设置数据相差较小，更符合用户的使用习惯，提高音量设置的个性化和智能化程度。In each of the above-mentioned embodiments of the present application, the prediction model is trained according to the volume sample data to obtain the target volume. The training end condition of the prediction model is that the loss value is less than the preset threshold value, so the target volume obtained after training the prediction model and the volume setting The Euclidean norm of the difference between the data is less than the preset threshold, the volume of the audio to be output is adjusted to the target volume, and the volume sample data contains the user volume setting data, so the volume of the audio to be output is consistent with the user volume The difference in the setting data is small, which is more in line with the user's usage habits, and the degree of personalization and intelligence of the volume setting is improved.
附图说明Description of drawings
图1为本申请一实施例提供的音量调节方法的流程示意图；FIG. 1 is a schematic flowchart of a volume adjustment method provided by an embodiment of the present application;
图2为本申请另一实施例提供的音量调节方法的流程图；FIG. 2 is a flowchart of a volume adjustment method provided in another embodiment of the present application;
图3为本申请另一实施例提供的音量调节方法的流程图；FIG. 3 is a flowchart of a volume adjustment method provided in another embodiment of the present application;
图4为一个神经网络模型的结构示意图；Fig. 4 is a structural representation of a neural network model;
图5为本申请一实施例提供的电子装置的结构示意图；FIG. 5 is a schematic structural diagram of an electronic device provided by an embodiment of the present application;
图6为本申请另一实施例提供的电子装置的结构示意图；FIG. 6 is a schematic structural diagram of an electronic device provided by another embodiment of the present application;
图7为本申请实施例提供的电子装置的硬件结构示意图。FIG. 7 is a schematic diagram of a hardware structure of an electronic device provided by an embodiment of the present application.
具体实施方式Detailed ways
为使得本申请的申请目的、特征、优点能够更加的明显和易懂，下面将结合本申请实施例提供的附图，对本申请实施例提供的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本申请一部分实施例，而非全部实施例。基于本申请提供的实施例，本领域技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。In order to make the purpose, features, and advantages of the application more obvious and understandable, the technical solutions provided by the embodiments of the application will be clearly and completely described below in conjunction with the accompanying drawings provided by the embodiments of the application. Obviously, the described The embodiments are only some of the embodiments of the present application, but not all of them. Based on the embodiments provided in this application, all other embodiments obtained by those skilled in the art without making creative efforts belong to the scope of protection of this application.
请参阅图1，图1为本申请一实施例提供的音量调节方法的流程示意图。该音量调节方法可应用于具有音频输出功能的电子装置，如：手机、智能耳机、平板电脑、手提电脑、MP3(MPEG Audio Layer-3音频动态压缩第三层)播放器、台式计算机等。如图所示，该音量调节方法包括：Please refer to FIG. 1 . FIG. 1 is a schematic flowchart of a volume adjustment method provided by an embodiment of the present application. The volume adjustment method can be applied to electronic devices with audio output functions, such as mobile phones, smart earphones, tablet computers, laptop computers, MP3 (MPEG Audio Layer-3 audio dynamic compression layer 3) players, desktop computers, and the like. As shown in the figure, the volume adjustment method includes:
101、获取音量样本数据，该音量样本数据包括环境声音数据和音量设置数据。101. Acquire volume sample data, where the volume sample data includes ambient sound data and volume setting data.
电子装置获取音量样本数据，该环境声音数据为用户所在环境中的声音数据，该音量设置数据为用户设置的不同环境下的待输出音频的音量数据。The electronic device acquires volume sample data, the environmental sound data is the sound data in the environment where the user is located, and the volume setting data is the volume data of audio to be output in different environments set by the user.
获取音量样本数据的方式，具体可以为每当检测到用户的音量设置操作时，获取该音量设置操作指向的音量设置数据。同时根据该音量设置数据，将待输出音频的音量调节为该音量设置数据对应的音量。然后，采集音量调节后预设时长内的环境声音数据。The manner of acquiring the volume sample data may specifically be to acquire the volume setting data pointed to by the volume setting operation whenever the user's volume setting operation is detected. At the same time, according to the volume setting data, the volume of the audio to be output is adjusted to the volume corresponding to the volume setting data. Then, the ambient sound data within a preset time period after volume adjustment is collected.
获取音量样本数据的方式，具体还可以为，电子装置中设置有用于监听系统中发生的各个事件的事件监听器，每当通过该事件监听器监听到音频播放事件时，实时采集音量设置数据和环境声音数据。The method of obtaining the volume sample data may also specifically be that the electronic device is provided with an event listener for monitoring various events occurring in the system, and whenever an audio playback event is monitored through the event listener, the volume setting data and ambient sound data.
102、将该音量样本数据输入预测模型进行训练，得到目标音量。102. Input the volume sample data into a prediction model for training to obtain a target volume.
具体的，将该环境声音数据输入该预测模型中，并对该预测模型进行训练，直至该预测模型满足预设的训练结束条件，得到目标音量。该训练结束条件可以为该预测模型的损失值小于预设的阈值，该损失值为该预测模型输出的音量与音量设置数据的差的欧几里得范数。Specifically, the environmental sound data is input into the predictive model, and the predictive model is trained until the predictive model satisfies a preset training end condition to obtain a target volume. The training end condition may be that the loss value of the prediction model is less than a preset threshold, and the loss value is the Euclidean norm of the difference between the volume output by the prediction model and the volume setting data.
该目标音量可以为预测模型训练完成后输出的音量。The target volume may be the output volume after the prediction model training is completed.
103、将待输出音频的音量调节为该目标音量。103. Adjust the volume of the audio to be output to the target volume.
需要说明的是，在预测模型满足预设的训练结束条件前，待输出音频的音量仍由音量设置数据决定。在预测模型满足预设的学习条件结束后，将待输出音频的音量调节为目标音量。又因为该目标音量与音量设置数据之间的差的欧几里得范数小于预设的阈值，故本实施例提供的音量调节方法，可以使待输出音频的音量符合用户的使用习惯。It should be noted that before the prediction model satisfies the preset training end condition, the volume of the audio to be output is still determined by the volume setting data. After the prediction model meets the preset learning conditions, the volume of the audio to be output is adjusted to the target volume. And because the Euclidean norm of the difference between the target volume and the volume setting data is smaller than the preset threshold, the volume adjustment method provided in this embodiment can make the volume of the audio to be output conform to the usage habits of the user.
进一步地，将待输出的音频的音量调节为该目标音量后，输出调节音量后的音频。Further, after the volume of the audio to be output is adjusted to the target volume, the volume-adjusted audio is output.
在本实施例中，根据音量样本数据对预测模型进行训练，得到目标音量，预测模型的训练结束条件为损失值小于预设的阈值，所以通过该预测模型训练后得到的目标音量与音量设置数据之间的差的欧几里得范数小于预设的阈值，将待输出音频的音量调节为目标音量，在该音量样本数据中包含用户音量设置数据，因此待输出音频的音量与用户音量设置数据相差较小，更符合用户的使用习惯，提高音量设置的个性化和智能化程度。In this embodiment, the prediction model is trained according to the volume sample data to obtain the target volume. The training end condition of the prediction model is that the loss value is less than the preset threshold, so the target volume and volume setting data obtained after training the prediction model The Euclidean norm of the difference between is less than the preset threshold value, adjust the volume of the audio to be output to the target volume, and the volume sample data contains the user volume setting data, so the volume of the audio to be output is consistent with the user volume setting The difference in data is small, which is more in line with the user's usage habits, and the degree of personalization and intelligence of the volume setting is improved.
请参阅图2，图2为本申请另一实施例提供的音量调节方法的流程图，该音量调节方法可应用于具有音频输出功能的电子装置，如：手机、智能耳机、平板电脑、手提电脑、MP3播放器、台式计算机等。如图2所示，该音量调节方法包括：Please refer to FIG. 2. FIG. 2 is a flow chart of a volume adjustment method provided by another embodiment of the present application. The volume adjustment method can be applied to electronic devices with audio output functions, such as mobile phones, smart earphones, tablet computers, and laptop computers. , MP3 players, desktop computers, etc. As shown in Figure 2, the volume adjustment method includes:
201、获取音量样本数据，该音量样本数据包括环境声音数据和音量设置数据。201. Acquire volume sample data, where the volume sample data includes ambient sound data and volume setting data.
电子设备获取音量样本数据，该音量样本数据包括环境声音数据和音量设置数据，该环境声音数据为用户所在环境中的声音数据，该音量设置数据为用户设置的不同环境下的待输出音频的音量数据。The electronic device acquires volume sample data, the volume sample data includes ambient sound data and volume setting data, the ambient sound data is the sound data in the environment where the user is located, and the volume setting data is the volume of the audio to be output in different environments set by the user data.
每当检测到用户的音量设置操作时，获取该音量设置操作指向的音量设置数据。同时根据该音量设置数据，将待输出音频的音量调节为该音量设置数据对应的音量。然后，采集音量调节后预设时长内的环境声音数据。Whenever a user's volume setting operation is detected, the volume setting data pointed to by the volume setting operation is obtained. At the same time, according to the volume setting data, the volume of the audio to be output is adjusted to the volume corresponding to the volume setting data. Then, the ambient sound data within a preset time period after volume adjustment is collected.
202、提取该环境声音数据中的高频环境声音数据，并将该高频环境声音数据输入显式预测模型中，该显式预测模型为以固定的关系式表示输入数据与输出数据之间的关系的预测模型。202. Extract high-frequency environmental sound data from the environmental sound data, and input the high-frequency environmental sound data into an explicit prediction model, where the explicit prediction model expresses the relationship between input data and output data with a fixed relational expression Predictive models of relationships.
具体的，该高频环境声音数据为频率大于预设频率阈值的环境声音数据。环境声音数据为环境音的响度的时间序列，对该时间序列进行傅立叶变换即可得到关于环境音的功率谱密度的频率序列，在该频率序列中提取频率大于预设频率阈值的功率谱密度谱线，并对该谱线进行逆傅里叶变换，即可得到高频环境声音数据。Specifically, the high-frequency ambient sound data is ambient sound data with a frequency greater than a preset frequency threshold. The environmental sound data is the time series of the loudness of the environmental sound, and the frequency series of the power spectral density of the environmental sound can be obtained by Fourier transforming the time series, and the power spectral density spectrum with a frequency greater than the preset frequency threshold is extracted from the frequency series line, and perform inverse Fourier transform on the spectral line to obtain high-frequency ambient sound data.
显式预测模型为以固定的关系式表示输入数据与输出数据之间的关系的预测模型，例如线性回归预测模型。An explicit prediction model is a prediction model that expresses the relationship between input data and output data with a fixed relational expression, such as a linear regression prediction model.
由于显式预测模型的输入数据与输出数据之间的关系可以由一个固定的关系式表示，训练该显式预测模型的过程的可解释性高且训练完成后该显式预测模型的输出结果的可靠性高。Since the relationship between the input data and the output data of the explicit prediction model can be expressed by a fixed relational expression, the process of training the explicit prediction model has high interpretability and the output of the explicit prediction model after training is complete. High reliability.
203、搜索该显式预测模型中固定的关系式的目标系数。203. Search for the target coefficient of the fixed relational expression in the explicit prediction model.
具体的，该目标系数使该显式预测模型的损失值取到最小值，该损失值为音量调节数据与音量设置数据的差的欧几里得范数，该音量调节数据为该显式预测模型输出的数据。Specifically, the target coefficient minimizes the loss value of the explicit prediction model, and the loss value is the Euclidean norm of the difference between the volume adjustment data and the volume setting data, and the volume adjustment data is the explicit prediction model The output data of the model.
可选的，搜索该固定的关系式的目标系数的方法可以是有导师的搜索，即根据该显式预测模型的损失值的大小改变该关系式的系数，例如，基于梯度下降法、粒子群算法或遗传算法搜索该目标系数。Optionally, the method of searching for the target coefficient of the fixed relational expression can be a search with a mentor, that is, changing the coefficient of the relational expression according to the loss value of the explicit prediction model, for example, based on the gradient descent method, particle swarm An algorithm or genetic algorithm searches for the target coefficients.
可选的，搜索该固定的关系式的目标系数的方法还可以是无导师的搜索，即该关系式的改变与该显式预测模型的损失值的大小无关，例如，基于穷举法或随机搜索法搜索该目标系数。Optionally, the method of searching for the target coefficient of the fixed relational expression can also be an unsupervised search, that is, the change of the relational expression has nothing to do with the size of the loss value of the explicit prediction model, for example, based on exhaustive method or random The search method searches for the target coefficient.
204、在每次搜索到该目标系数后，分析该显式预测模型的损失值是否小于预设的阈值，并当该显式预测模型的损失值小于预设的阈值时，确认该显式预测模型为训练完成显式预测模型，将该训练完成显式预测模型输出的音量确认为目标音量。204. After each search for the target coefficient, analyze whether the loss value of the explicit prediction model is less than the preset threshold, and confirm the explicit prediction when the loss value of the explicit prediction model is less than the preset threshold The model is an explicit prediction model after training, and the volume output by the explicit prediction model after training is confirmed as the target volume.
在实际应用中，为了防止偶然性样本对该显式预测模型的训练的影响，将预设数量的高频环境声音数据依次输入该显式预测模型，将每个高频环境声音数据输入该显式预测模型后，得到一个该显式预测模型的损失值，若这些损失值的平均值小于预设的阈值，则确认该显式预测模型为训练完成显式预测模型。In practical applications, in order to prevent accidental samples from affecting the training of the explicit prediction model, a preset number of high-frequency ambient sound data is sequentially input into the explicit prediction model, and each high-frequency ambient sound data is input into the explicit prediction model. After predicting the model, a loss value of the explicit prediction model is obtained, and if the average value of these loss values is less than a preset threshold, it is confirmed that the explicit prediction model is an explicit prediction model that has been trained.
可选的，为避免目标音量的改变过于频繁，还可分析该训练完成显式预测模型输出的音量所属的音量档位，获取该音量档位对应的参考音量作为目标音量。Optionally, in order to prevent the target volume from changing too frequently, it is also possible to analyze the volume level to which the volume output by the explicit prediction model after training belongs, and obtain the reference volume corresponding to the volume level as the target volume.
具体的，预设多个音量档位及各档位对应的参考音量，在一实际应用例中，假设理论目标音量的波动范围在0～100分贝，则可将该范围划分为五个档位：0～20分别为音量档位一档、20～40分贝为音量档位二档、40～60分贝为音量档位三档、60～80分贝为音量档位四档、80～100分贝为音量五档。Specifically, a plurality of volume levels and reference volumes corresponding to each level are preset. In a practical application example, assuming that the fluctuation range of the theoretical target volume is 0-100 decibels, the range can be divided into five levels. : 0-20 is the first volume level, 20-40 decibels is the second volume level, 40-60 decibels is the third volume level, 60-80 decibels is the fourth volume level, 80-100 decibels is the Volume five.
每个音量档位对应一个参考音量，将该训练完成显式预测模型输出的音量所属的音量档位对应的参考音量，作为目标音量。例如，音量档位一档对应的参考音量为10分贝，音量档位二档对应的参考音量为30分贝，音量档位三档对应的参考音量为50分贝，音量档位四档对应的参考音量为60分贝，音量档位五档对应的参考音量为70分贝。若该训练完成显式预测模型输出的音量为35分别，则属于音量档位二档，音量档位二档对应的参考音量50分贝即为目标音量。Each volume level corresponds to a reference volume, and the reference volume corresponding to the volume level to which the volume output by the explicit prediction model after training belongs is taken as the target volume. For example, the reference volume corresponding to the first volume gear is 10 decibels, the reference volume corresponding to the second volume gear is 30 decibels, the reference volume corresponding to the third volume gear is 50 decibels, and the reference volume corresponding to the fourth volume gear is is 60 decibels, and the reference volume corresponding to the fifth gear of the volume level is 70 decibels. If the training is completed and the output volume of the explicit prediction model is 35, it belongs to the second level of the volume level, and the reference volume corresponding to the second level of the volume level is 50 decibels, which is the target volume.
像这样，通过预设多个音量档位及各档位对应的参考音量，可以防止目标音量对高频环境音的改变过于敏感，避免目标音量的改变过于频繁。Like this, by presetting multiple volume levels and the reference volumes corresponding to each level, it is possible to prevent the target volume from being too sensitive to changes in high-frequency ambient sounds, and avoid changing the target volume too frequently.
205、将待输出音频的音量调节为该目标音量。205. Adjust the volume of the audio to be output to the target volume.
可选的，于本申请其他一实施例中，为进一步提高待输出音频的播放效果，达到主动隔音的效果，在将音量样本数据输入预测模型进行训练，得到目标音量，将待输出音频的音量调节为目标音量的同时，还可以提取环境声音数据中的低频环境声音数据，获取与该低频环境声音数据对应的低频隔音音频并输出。Optionally, in another embodiment of the present application, in order to further improve the playback effect of the audio to be output and achieve the effect of active sound insulation, the volume sample data is input into the prediction model for training to obtain the target volume, and the volume of the audio to be output While adjusting the volume to the target volume, the low-frequency environmental sound data in the environmental sound data can also be extracted, and the low-frequency sound-proof audio corresponding to the low-frequency environmental sound data can be obtained and output.
具体的，该低频环境声音数据为频率小于预设频率阈值的环境声音数据。环境声音数据为环境音的响度的时间序列，对该时间序列进行傅立叶变换即可得到关于环境音的功率谱密度的频率序列，在该频率序列中提取频率小于预设频率阈值的功率谱密度谱线，并对该谱线进行逆傅里叶变换，即可得到低频环境声音数据。Specifically, the low-frequency ambient sound data is ambient sound data with a frequency lower than a preset frequency threshold. The environmental sound data is the time series of the loudness of the environmental sound, and the frequency series of the power spectral density of the environmental sound can be obtained by Fourier transforming the time series, and the power spectral density spectrum whose frequency is less than the preset frequency threshold is extracted from the frequency series line, and perform inverse Fourier transform on the spectral line to obtain low-frequency ambient sound data.
对该低频环境声音数据进行实时计算，获取与该低频环境声音数据对应的低频隔音音频并输出。Real-time calculation is performed on the low-frequency environmental sound data, and low-frequency sound-proof audio corresponding to the low-frequency environmental sound data is obtained and output.
该低频隔音音频的音量的绝对值与低频环境声音数据的音量的绝对值相等，该低频隔音音频的正负号与低频环境声音数据的正负号相反，该低频隔音音频用于抵消低频环境音，达到主动隔音的效果。The absolute value of the volume of the low-frequency sound-isolating audio is equal to the absolute value of the volume of the low-frequency ambient sound data, the sign of the low-frequency sound-isolating audio is opposite to that of the low-frequency ambient sound data, and the low-frequency sound-isolating audio is used to offset the low-frequency ambient sound , to achieve the effect of active sound insulation.
需要说明的是，由于需要运算处理，在提取得到低频环境声音数据与输出低频隔音音频之间存在时间延迟，即低频环境声音数据与低频隔音音频之间存在相位差，相较于低频环境声音数据的周期，该相位差的大小可以忽略不计，故该相位差对低频隔音音频的主动隔音效果的影响非常小。It should be noted that due to the need for calculation processing, there is a time delay between the extracted low-frequency ambient sound data and the output of low-frequency sound-proof audio, that is, there is a phase difference between low-frequency ambient sound data and low-frequency sound-proof audio. Compared with low-frequency ambient sound data The phase difference is negligible, so the phase difference has very little influence on the active sound insulation effect of the low frequency sound insulation audio.
但对于高频环境声音数据，若在提取该高频环境声音数据后输出高频隔音音频，该高频隔音音频的音量的绝对值与高频环境声音数据的音量的绝对值相等，该高频隔音音频的正负号与高频环境声音数据的正负号相反，即利用该高频隔音音频抵消高频环境音，达到主动隔音的效果。But for high-frequency ambient sound data, if after extracting this high-frequency ambient sound data, output high-frequency sound-proof audio, the absolute value of the volume of this high-frequency sound-proof audio is equal to the absolute value of the volume of high-frequency ambient sound data, and the high-frequency The sign of the sound-proof audio is opposite to that of the high-frequency ambient sound data, that is, the high-frequency sound-proof audio is used to offset the high-frequency ambient sound to achieve the effect of active sound insulation.
由于需要处理计算，在提取该高频环境声音数据与输出高频隔音音频之间存在时间延时，即该高频环境声音数据与该高频隔音音频之间存在相位差，相较于高频环境声音数据的周期，该较为差不可忽略，该相位差对该高频隔音音频的隔音效果有很大的影响，故利用高频隔音音频主动隔音的效果不如利用低频隔音音频主动隔音的效果好。Due to the need for processing calculations, there is a time delay between extracting the high-frequency ambient sound data and outputting the high-frequency sound-proof audio, that is, there is a phase difference between the high-frequency ambient sound data and the high-frequency sound-proof audio, compared to the high-frequency The cycle of ambient sound data, the phase difference can not be ignored, the phase difference has a great impact on the sound insulation effect of the high-frequency sound-proof audio, so the active sound-proof effect of using high-frequency sound-proof audio is not as good as that of using low-frequency sound-proof audio. .
在实际应用中，利用低频隔音音频主动隔音，并根据高频环境声音数据调节待输出音频的音量，利用待输出音频覆盖该高频环境声音数据，能够防止低频环境声音数据对显式预测模型的训练的影响。In practical applications, using low-frequency sound-proof audio to actively isolate sound, and adjusting the volume of the audio to be output according to the high-frequency environmental sound data, and using the audio to be output to cover the high-frequency environmental sound data can prevent the low-frequency environmental sound data from affecting the explicit prediction model. The effect of training.
下面以将本实施例提供的音量调节方法应用于一个耳机上为例进行说明，并非对本实施例提供的音量调节方法进行任何形式的限制。The following uses an example of applying the volume adjustment method provided in this embodiment to an earphone for illustration, and does not limit the volume adjustment method provided in this embodiment in any form.
在火车车厢内环境嘈杂，既有乘客说话的声音又有火车车轮通过铁轨连接处产生的噪声，为了免受火车车厢内嘈杂环境的干扰，用户将耳机插入多媒体播放设备的而耳机孔中，利用该耳机播放音频，并增大将该耳机播放音频的音量。The environment in the train compartment is noisy, there are both the voice of passengers talking and the noise generated by the train wheels passing through the junction of the rails. The headset plays audio and increases the volume of the audio played on the headset.
该耳机的处理器检测到用户的音量设置操作，获取该音量设置操作的音量设置数据，同时根据该音量设置数据，将待输出音频的音量调节为该音量设置数据对应的音量。然后采集音量改变后预设时长内的环境声音数据。The processor of the earphone detects the user's volume setting operation, acquires the volume setting data of the volume setting operation, and simultaneously adjusts the volume of the audio to be output to the volume corresponding to the volume setting data according to the volume setting data. Then collect ambient sound data within a preset time period after the volume is changed.
然后处理器提取环境声音数据中的低频环境声音数据。具体的，该低频环境声音数据为频率小于1000赫兹的环境声音数据。环境声音数据为环境音的响度的时间序列，对该时间序列进行傅立叶变换即可得到关于环境音的功率谱密度的频率序列，在该频率序列中提取频率小于1000赫兹的功率谱密度的谱线，并对该谱线进行逆傅里叶变换，即可得到低频环境声音数据。The processor then extracts low frequency ambient sound data from the ambient sound data. Specifically, the low-frequency ambient sound data is ambient sound data with a frequency less than 1000 Hz. The environmental sound data is the time series of the loudness of the environmental sound, and the frequency series of the power spectral density of the environmental sound can be obtained by Fourier transforming the time series, and the spectral lines of the power spectral density with a frequency less than 1000 Hz are extracted from the frequency series , and the inverse Fourier transform of the spectral line can be obtained to obtain low-frequency environmental sound data.
该处理器在提取低频环境声音数据后，对该低频环境声音数据进行实时计算，获取与该低频环境声音数据对应的低频隔音音频并通过该耳机的听筒进行输出。该低频隔音音频的音量与低频环境声音数据的音量的绝对值相等，该低频隔音音频的与该低频环境声音数据的音量的正负号相反。该低频隔音音频用于抵消低频环境音，达到主动隔音的效果。After extracting the low-frequency environmental sound data, the processor performs real-time calculation on the low-frequency environmental sound data, acquires low-frequency sound-proof audio corresponding to the low-frequency environmental sound data, and outputs it through the earpiece of the earphone. The volume of the low-frequency sound-insulating audio is equal to the absolute value of the volume of the low-frequency ambient sound data, and the sign of the volume of the low-frequency sound-insulating audio is opposite to that of the low-frequency ambient sound data. The low-frequency sound-isolating audio is used to offset the low-frequency ambient sound to achieve the effect of active sound insulation.
在提取环境声音数据中的低频环境声音数据的同时，该处理器还提取该环境声音数据中的高频环境声音数据。具体的，该高频环境声音数据为频率大于1000赫兹的环境声音数据。环境声音数据为环境音的响度的时间序列，对该时间序列进行傅立叶变换即可得到关于环境音的功率谱密度的频率序列，在该频率序列中提取频率大于1000赫兹的功率谱密度的谱线，并对该谱线进行逆傅里叶变换，即可得到高频环境声音数据。While extracting low frequency ambient sound data from the ambient sound data, the processor also extracts high frequency ambient sound data from the ambient sound data. Specifically, the high-frequency ambient sound data is ambient sound data with a frequency greater than 1000 Hz. The environmental sound data is the time series of the loudness of the environmental sound, and the frequency series of the power spectral density of the environmental sound can be obtained by Fourier transforming the time series, and the spectral lines of the power spectral density with a frequency greater than 1000 Hz are extracted from the frequency series , and the inverse Fourier transform of the spectral line can be obtained to obtain high-frequency ambient sound data.
然后该处理器将该高频环境声音数据输入显式预测模型中。该显式预测模型中输入数据与输出数据之间的关系可以用一个固定的关系式表示，该关系式例如可以为：The processor then inputs the high frequency ambient sound data into an explicit predictive model. The relationship between input data and output data in the explicit prediction model can be represented by a fixed relational expression, which can be, for example:
a0+a1x+a2x2+a3x3＝y (1)a0 +a1 x+a2 x2 +a3 x3 =y (1)
式(1)中x为输入该显式预测模型的高频环境声音数据，y为该显式预测模型输出的音量，a0、a1、a2和a3分别为该固定的关系式的各项系数。In formula (1), x is the high-frequency environmental sound data input to the explicit prediction model, y is the volume output by the explicit prediction model, a0 , a1 , a2 and a3 are the fixed relational expressions various coefficients.
接着搜索该显式预测模型的固定的关系式的目标系数，该目标系数使该显式预测模型的损失值取到最小值，该显式预测模型的损失值为该显式预测模型与音量设置数据的差的欧几里得范数，该损失值的计算公式为：Then search for the target coefficient of the fixed relational expression of the explicit prediction model, the target coefficient makes the loss value of the explicit prediction model take the minimum value, the loss value of the explicit prediction model and the volume setting of the explicit prediction model The Euclidean norm of the difference of the data, the calculation formula of the loss value is:
ξ＝(y-y0)2 (2)ξ＝(yy0 )2 (2)
式(2)中ξ为损失值，y0为在火车车厢的嘈杂环境内用户设置该耳机待输出音频数据的音量数据。In formula (2), ξ is the loss value, and y0 is the volume data of the audio data to be output by the earphone set by the user in the noisy environment of the train carriage.
下面以有导师的搜索中的基于粒子群算法搜索目标系数为例具体说明搜索该关系式的目标系数的方法，并非对搜索该目标系数的方法进行任何形式的限制。In the following, the search for the target coefficient based on the particle swarm algorithm in the search with a tutor is taken as an example to specifically illustrate the method of searching the target coefficient of the relational expression, without any form of restriction on the method of searching the target coefficient.
将该关系式的各项系数a0、a1、a2和a3作为粒子群中的粒子，随机选取M个各粒子的初始位置，得到第一代粒子群A1，该第一代粒子群A1为一个M行四列的矩阵，每一行中各列中的数值对应关系式中的各项系数。M的大小根据实际需求确定，M越大粒子群的规模越大，搜索的结果更准确，但搜索耗费的时间也越长。The coefficients a0 , a1 , a2 and a3 of the relational expression are used as the particles in the particle swarm, and the initial positions of M particles are randomly selected to obtain the first-generation particle swarm A1 , the first-generation particle Group A1 is a matrix with M rows and four columns, and the values in each column in each row correspond to the coefficients in the relational expression. The size of M is determined according to actual needs. The larger M is, the larger the size of the particle swarm is, and the search results are more accurate, but the search takes longer.
再将第一代粒子群A1中的每一行分别带入公式(1)中得到M个显式预测模型，将高频环境声音数据输入分别输入该M个显式预测模型中，并根据公式(2)计算得到每个显式预测模型的损失值，将本代中该M个显式预测模型中具有最小的损失值的显式预测模型的关系式的各项系数作为为本代局部最优解Aj，将目前所有显式预测模型中具有最小的损失值的显式预测模型的关系式的各项系数作为全局最优解Aq，对于第一代粒子群，目前所有的显式预测模型的损失值即为本代中M个显式预测模型的损失值，故对于第一代粒子群全局最优解即为本代局部最优解Aj。Then, each line in the first-generation particle swarm A1 is brought into the formula (1) to obtain M explicit prediction models, and the high-frequency environmental sound data are respectively input into the M explicit prediction models, and according to the formula (2) Calculate the loss value of each explicit prediction model, and take the coefficients of the relational expression of the explicit prediction model with the smallest loss value among the M explicit prediction models in this generation as the local maximum of this generation For the optimal solution Aj , take the coefficients of the relational expression of the explicit prediction model with the smallest loss value among all the current explicit prediction models as the global optimal solution Aq , for the first generation of particle swarms, all the current explicit prediction models The loss value of the prediction model is the loss value of M explicit prediction models in this generation, so the global optimal solution of the first generation particle swarm is the local optimal solution Aj of this generation.
判断第一代粒子群是否满足结束条件，该结束条件为以全局最优解Aq为系数的显式预测模型的损失值小于预设的阈值。若第一代粒子群满足该结束条件，则输出目前的全局最优解Aq作为目标系数。Judging whether the first-generation particle swarm meets the end condition, the end condition is that the loss value of the explicit prediction model with the global optimal solution Aq as the coefficient is less than the preset threshold. If the first-generation particle swarm satisfies the end condition, the current global optimal solution Aq is output as the target coefficient.
若第一代粒子群不满足该结束条件，则根据公式：If the first-generation particle swarm does not meet the end condition, according to the formula:
Vi+1＝ωVi+c1(Ai－Aj)+c2(Ai－Aq) (3)Vi+1 =ωVi +c1 (Ai -Aj )+c2 (Ai -Aq ) (3)
更新粒子群的迁移速度，第一代粒子群的迁移速度V0为0。式(3)中，Vi为第i代粒子群的迁移速度。ω为惯性系数、c1为认知系数、c2为社会系数，ω、c1和c2均为预设大小的常数。The migration velocity of the particle swarm is updated, and the migration velocity V0 of the first generation particle swarm is 0. In formula (3), Vi is the migration velocity of the i-th generation particle group. ω is the inertial coefficient, c1 is the cognitive coefficient, c2 is the social coefficient, and ω, c1 and c2 are constants with preset sizes.
再然后根据公式：Then according to the formula:
Ai+1＝Ai+Vi (4)Ai+1 ＝Ai +Vi (4)
更新粒子群的位置得到第二代粒子群A2，式(4)中Ai为第i代粒子群的位置。Update the position of the particle swarm to obtain the second-generation particle swarm A2 , where Ai in formula (4) is the position of the i-th generation particle swarm.
将第二代粒子群A2中的每一行分别带入公式(1)中，得到M个显式预测模型，将高频环境声音数据输入分别输入该M个显式预测模型中，并根据公式(2)计算得到该M个显式预测模型的损失值，将本代中该M个显式预测模型中具有最小的损失值的显式预测模型的关系式的各项系数作为为本代局部最优解Aj，分析以本代局部最优解Aj显式预测模型的损失值是否小于以全局最优解Aq为系数的显式预测模型的损失值，若以本代局部最优解Aj显式预测模型的损失值小于以全局最优解Aq为系数的显式预测模型的损失值，则以本代局部最优解Aj覆盖全局最优解Aq，若以本代局部最优解Aj显式预测模型的损失值不小于以全局最优解Aq为系数的显式预测模型的损失值，则保持全局最优解Aq不变。Put each row in the second-generation particle swarm A2 into the formula (1) respectively to obtain M explicit prediction models, and input the high-frequency environmental sound data into the M explicit prediction models respectively, and according to the formula (2) Calculate the loss value of the M explicit prediction models, and take the coefficients of the relational expression of the explicit prediction model with the smallest loss value among the M explicit prediction models in this generation as the local The optimal solution Aj , analyze whether the loss value of the explicit prediction model with the local optimal solution Aj of this generation is smaller than the loss value of the explicit prediction model with the global optimal solution Aq as the coefficient, if the local optimal solution of the current generation If the loss value of the explicit prediction model for the solution Aj is smaller than the loss value of the explicit prediction model with the global optimal solution Aq as the coefficient, then the local optimal solution Aj of this generation covers the global optimal solution Aq . If the loss value of the explicit prediction model with the local optimal solution Aj is not less than the loss value of the explicit prediction model with the global optimal solution Aq as the coefficient, the global optimal solution Aq remains unchanged.
判断第二代粒子群是否满足结束条件，若第二代粒子群满足该结束条件，则输出全局最优解Aq作为目标系数。Judging whether the second generation particle swarm meets the end condition, if the second generation particle swarm meets the end condition, output the global optimal solution Aq as the target coefficient.
若第二代粒子群不满足该结束条件，则根据公式(3)更新粒子群的迁移速度，并根据公式(4)更新粒子群的位置得到新的一代粒子群。If the second-generation particle swarm does not meet the end condition, the migration velocity of the particle swarm is updated according to formula (3), and the position of the particle swarm is updated according to formula (4) to obtain a new generation of particle swarm.
迭代上述将第N代粒子群AN中的每一行分别带入公式(1)中，得到M个显式预测模型，将高频环境声音数据输入分别输入该M个显式预测模型中，并根据公式(2)计算得到该M个显式预测模型的损失值，将本代中该M个显式预测模型中具有最小的损失值的显式预测模型的关系式的各项系数作为为本代局部最优解Aj，分析以本代局部最优解Aj显式预测模型的损失值是否小于以全局最优解Aq为系数的显式预测模型的损失值，若以本代局部最优解Aj显式预测模型的损失值小于以全局最优解Aq为系数的显式预测模型的损失值，则以本代局部最优解Aj覆盖全局最优解Aq，若以本代局部最优解Aj显式预测模型的损失值不小于以全局最优解Aq为系数的显式预测模型的损失值，则保持全局最优解Aq不变。直至该第N代粒子群满足该结束条件，并输出全局最优解Aq作为目标系数。Iteratively bring each row in the Nth generation particle swarm AN into the formula (1) to obtain M explicit prediction models, and input the high-frequency environmental sound data into the M explicit prediction models respectively, and The loss values of the M explicit prediction models are calculated according to formula (2), and the coefficients of the relational expressions of the explicit prediction models with the smallest loss value among the M explicit prediction models in this generation are taken as the basis replace the local optimal solution Aj , and analyze whether the loss value of the explicit prediction model with the local optimal solution Aj of this generation is smaller than the loss value of the explicit prediction model with the global optimal solution Aq as the coefficient. The loss value of the explicit prediction model for the optimal solution Aj is smaller than the loss value of the explicit prediction model with the global optimal solution Aq as the coefficient, then the local optimal solution Aj of this generation covers the global optimal solution Aq , if The loss value of the explicit prediction model with the local optimal solution Aj of this generation is not less than the loss value of the explicit prediction model with the global optimal solution Aq as the coefficient, and the global optimal solution Aq remains unchanged. Until the Nth generation particle swarm satisfies the end condition, and output the global optimal solution Aq as the target coefficient.
为了避免由于粒子群的代数过多导致计算时间过长，还需要设置一个粒子群的代数阈值，该代数阈值例如为1000代，若在1000代粒子群内所有粒子群均不满足该结束条件，则在计算得到第1000代粒子群的本代局部最优解Aj和全局最优解Aq后无论该粒子群是否满足该结束条件，强制结束对目标系数的搜索，并输出第1000代的全局最优解Aq作为目标系数。In order to avoid too long calculation time due to too many algebras of particle swarms, it is also necessary to set an algebraic threshold of particle swarms. For example, the algebraic threshold is 1000 generations. Then, after calculating the local optimal solution Aj and the global optimal solution Aq of the 1000th generation particle swarm, no matter whether the particle swarm satisfies the end condition, the search for the target coefficient is forced to end, and the 1000th generation is output. The global optimal solution Aq is used as the target coefficient.
在每次搜索到目标系数后，该处理器分析该显式预测模型的损失值是否小于预设的阈值，并当该显式预测模型的损失值小于该预设的阈值时，确认该显式预测模型为训练完成显式预测模型，将该训练完成显式预测模型输出的音量确认为目标音量。After searching for the target coefficient each time, the processor analyzes whether the loss value of the explicit prediction model is less than a preset threshold, and when the loss value of the explicit prediction model is less than the preset threshold, confirms that the explicit The prediction model is an explicit prediction model after training, and the volume output by the explicit prediction model after training is confirmed as the target volume.
为了避免偶然性数据对该显式预测模型的训练的影响，将二十个高频环境声音数据输入该显式预测模型中，若该显式预测模型的损失值的平均值小于预设的阈值，则认为该显式预测模型为训练完成显式预测模型。In order to avoid the impact of accidental data on the training of the explicit prediction model, input twenty high-frequency environmental sound data into the explicit prediction model, if the average value of the loss value of the explicit prediction model is less than the preset threshold, Then the explicit prediction model is considered to be an explicit prediction model that has been trained.
可选的，为避免目标音量的改变过于频繁，还可以分析该训练完成显式预测模型输出的音量所属的音量档位，获取该音量档位对应的参考音量作为目标音量。Optionally, in order to prevent the target volume from changing too frequently, it is also possible to analyze the volume level to which the volume output by the explicit prediction model after training belongs, and obtain the reference volume corresponding to the volume level as the target volume.
最后该处理器将待输出音频的音量调节为该目标音量即可实现根据用户的习惯调节耳机播放音频的音量。Finally, the processor adjusts the volume of the audio to be output to the target volume, so that the volume of the audio played by the earphone can be adjusted according to the user's habit.
在本实施例中，第一方面，根据音量样本数据对预测模型进行训练，得到目标音量，预测模型的训练结束条件为损失值小于预设的阈值，所以通过该预测模型训练后得到的目标音量与音量设置数据之间的差的欧几里得范数小于预设的阈值，将待输出音频的音量调节为目标音量，在该音量样本数据中包含用户音量设置数据，因此待输出音频的音量与用户音量设置数据相差较小，更符合用户的使用习惯，提高音量设置的个性化和智能化程度。第二方面，由于根据高频环境声音数据对显式预测模型进行训练，故可以避免低频环境声音数据对显式预测模型的训练的影响。第三方面，由于预测模型为显式预测模型，该显式预测模型的输入数据和输出数据之间的关系可以由一个固定的关系式表示，故训练该显式预测模型的过程的可解释性高且训练完成后该显式预测模型的输出结果的可靠性高。第四方面，由于将训练完成显式预测模型输出的音量所属的音量档位，获取该音量档位对应的参考音量作为目标音量，故可以防止目标音量对高频环境音的改变过于敏感，避免目标音量的改变过于频繁，进而可以避免待输出音频的音量的改变过于频繁。In this embodiment, in the first aspect, the prediction model is trained according to the volume sample data to obtain the target volume. The training end condition of the prediction model is that the loss value is less than the preset threshold, so the target volume obtained after training the prediction model The Euclidean norm of the difference with the volume setting data is less than the preset threshold, and the volume of the audio to be output is adjusted to the target volume. The volume sample data contains the user volume setting data, so the volume of the audio to be output The difference from the volume setting data of the user is small, which is more in line with the user's usage habits, and the degree of personalization and intelligence of the volume setting is improved. In the second aspect, since the explicit prediction model is trained according to the high-frequency ambient sound data, the influence of the low-frequency ambient sound data on the training of the explicit prediction model can be avoided. In the third aspect, since the prediction model is an explicit prediction model, the relationship between the input data and output data of the explicit prediction model can be expressed by a fixed relation, so the interpretability of the process of training the explicit prediction model High and the reliability of the output result of the explicit prediction model is high after the training is completed. In the fourth aspect, since the volume gear of the volume output by the explicit prediction model after training is completed, the reference volume corresponding to the volume gear is obtained as the target volume, so it can prevent the target volume from being too sensitive to changes in high-frequency environmental sounds, and avoid The change of the target volume is too frequent, thereby preventing the volume of the audio to be output from changing too frequently.
请参阅图3，图3为本申请另一实施例提供的音量调节方法的流程图，该音量调节方法可应用于具有音频输出功能的电子装置，如：手机、智能耳机、平板电脑、手提电脑、MP3播放器、台式计算机等。如图3所示，该音量调节方法包括：Please refer to Figure 3, Figure 3 is a flowchart of a volume adjustment method provided by another embodiment of the present application, the volume adjustment method can be applied to electronic devices with audio output functions, such as: mobile phones, smart earphones, tablet computers, laptop computers , MP3 player, desktop computer, etc. As shown in Figure 3, the volume adjustment method includes:
301、获取音量样本数据，该音量样本数据包括环境声音数据和音量设置数据。301. Acquire volume sample data, where the volume sample data includes ambient sound data and volume setting data.
电子设备获取音量样本数据，具体的，该音量样本数据包括环境声音数据和音量设置数据，该环境声音数据为用户所在环境中的声音数据，该音量设置数据为用户设置的不同环境下的待输出音频的音量数据。The electronic device acquires volume sample data, specifically, the volume sample data includes ambient sound data and volume setting data, the ambient sound data is the sound data in the environment where the user is located, and the volume setting data is the audio data to be output in different environments set by the user. Audio volume data.
每当检测到用户的音量设置操作时，获取该音量设置操作指向的音量设置数据。同时根据该音量设置数据，将待输出音频的音量调节为该音量设置数据对应的音量。然后，采集音量改变后预设时长内的环境声音数据。Whenever a user's volume setting operation is detected, the volume setting data pointed to by the volume setting operation is obtained. At the same time, according to the volume setting data, the volume of the audio to be output is adjusted to the volume corresponding to the volume setting data. Then, the ambient sound data within a preset time period after the volume is changed is collected.
302、提取该环境声音数据中的高频环境声音数据，并将该高频环境声音数据输入隐式预测模型中，该隐式预测模型为以网络连接关系和连接权重表示输入数据与输出数据之间的关系的预测模型。302. Extract high-frequency environmental sound data from the environmental sound data, and input the high-frequency environmental sound data into an implicit prediction model. A predictive model of the relationship between them.
具体的，该高频环境声音数据为频率大于预设频率阈值的环境声音数据。环境声音数据为环境音的响度的时间序列，对该时间序列进行傅立叶变换即可得到关于环境音的功率谱密度的频率序列，在该频率序列中提取频率大于预设频率阈值的功率谱密度谱线，并对该谱线进行逆傅里叶变换，即可得到高频环境声音数据。Specifically, the high-frequency ambient sound data is ambient sound data with a frequency greater than a preset frequency threshold. The environmental sound data is the time series of the loudness of the environmental sound, and the frequency series of the power spectral density of the environmental sound can be obtained by Fourier transforming the time series, and the power spectral density spectrum with a frequency greater than the preset frequency threshold is extracted from the frequency series line, and perform inverse Fourier transform on the spectral line to obtain high-frequency ambient sound data.
该隐式预测模型为以网络连接关系与连接权重表示输入数据与输出数据之间的关系的预测模型，例如循环神经网络模型、卷积神经网络模型。The implicit prediction model is a prediction model that expresses the relationship between input data and output data by network connection relationship and connection weight, such as a recurrent neural network model and a convolutional neural network model.
需要说明的是，隐式预测模型的输入数据与输出数据之间的关系不由一个确定的表达式确定，而是通过网络连接关系与连接权重确定，通过改变网络的连接关系和连接权重可以表示不同的输入数据与输出数据之间的关系，相较于通过一个确定表达式确定输入数据与输出数据之间的关系的隐式预测模型，隐式预测模型表示的输入数据与输出数据之间的关系更加灵活，故相较于隐式预测模型，隐式预测模型的训练速度快且训练完成后损失值小，该损失值为音量调节数据与音量设置数据的差的欧几里得范数，该音量调节数据为该隐式预测模型输出的数据。It should be noted that the relationship between the input data and output data of the implicit prediction model is not determined by a definite expression, but is determined by the network connection relationship and connection weight. By changing the connection relationship and connection weight of the network, different Compared with the implicit prediction model that determines the relationship between the input data and the output data through a definite expression, the relationship between the input data and the output data represented by the implicit prediction model It is more flexible, so compared with the implicit prediction model, the training speed of the implicit prediction model is fast and the loss value is small after the training is completed. The loss value is the Euclidean norm of the difference between the volume adjustment data and the volume setting data. The volume adjustment data is data output by the implicit prediction model.
303、搜索该隐式预测模型的目标连接关系和目标连接权重。303. Search for the target connection relationship and target connection weight of the implicit prediction model.
具体的，该隐式预测模型的目标连接关系和目标连接权重使该隐式预测模型的损失值取到最小值。Specifically, the target connection relationship and the target connection weight of the implicit prediction model minimize the loss value of the implicit prediction model.
可选的，搜索该隐式预测模型的目标连接关系和目标连接权重的方法可以是有导师的搜索，即根据该隐式预测模型的损失值的大小改变该隐式训练模型的连接关系和连接权重，例如，基于梯度下降法、粒子群算法或混度搜索法搜索该隐式预测模型的目标连接关系和该目标连接权重。Optionally, the method of searching for the target connection relationship and target connection weight of the implicit prediction model may be a search with a mentor, that is, changing the connection relationship and connection weight of the implicit training model according to the loss value of the implicit prediction model For weighting, for example, the target connection relationship and the target connection weight of the implicit prediction model are searched based on a gradient descent method, a particle swarm optimization algorithm or a confusion search method.
可选的，搜索该隐式预测模型的目标连接关系和目标连接权重的方法还可以是无导师的搜索，即该隐式预测模型的连接关系和连接权重的改变与该隐式预测模型的损失值的大小无关，例如，基于穷举法或随机搜索法搜索该隐式预测模型的。Optionally, the method of searching for the target connection relationship and target connection weight of the implicit prediction model can also be a search without a supervisor, that is, the change of the connection relationship and connection weight of the implicit prediction model and the loss of the implicit prediction model The size of the value is irrelevant, for example, based on exhaustive or random search of the implicit predictive model.
304、在每次搜索到该目标连接关系和该目标权重后，分析该隐式预测模型的损失值是否小于预设的阈值，并当该隐式预测模型的损失值小于预设的阈值时，确认该隐式预测模型为训练完成隐形预测模型，将该训练完成隐式预测模型输出的音量确认为目标音量。304. After searching for the target connection relationship and the target weight each time, analyze whether the loss value of the implicit prediction model is smaller than a preset threshold, and when the loss value of the implicit prediction model is smaller than the preset threshold, It is confirmed that the implicit prediction model is the trained implicit prediction model, and the volume output by the trained implicit prediction model is confirmed as the target volume.
在实际应用中，为了防止偶然性样本对该隐式预测模型的训练的影响，将预设数量的高频环境声音数据依次输入该隐式预测模型后，将每个高频环境声音数据输入该隐式预测模型后，得到一个该隐式预测模型的损失值，若这些损失值的平均值小于预设的阈值，则确认该隐式预测模型为训练完成隐式预测模型。In practical applications, in order to prevent accidental samples from affecting the training of the implicit prediction model, a preset number of high-frequency ambient sound data is sequentially input into the implicit prediction model, and each high-frequency ambient sound data is input into the implicit prediction model. After the formula prediction model is obtained, a loss value of the implicit prediction model is obtained, and if the average value of these loss values is less than the preset threshold, it is confirmed that the implicit prediction model is a trained implicit prediction model.
可选的，为避免目标音量的改变过于频繁，还可分析该训练完成隐式预测模型输出的音量所属的音量档位，获取该音量档位对应的参考音量作为目标音量。Optionally, in order to prevent the target volume from changing too frequently, it is also possible to analyze the volume level to which the volume output by the trained implicit prediction model belongs, and obtain the reference volume corresponding to the volume level as the target volume.
具体的，预设多个音量档位及各档位对应的参考音量，在一实际应用例中，假设理论目标音量的波动范围在0～100分贝，则可将该范围划分为五个档位：0～20分别为音量档位一档、20～40分贝为音量档位二档、40～60分贝为音量档位三档、60～80分贝为音量档位四档、80～100分贝为音量五档。Specifically, a plurality of volume levels and reference volumes corresponding to each level are preset. In a practical application example, assuming that the fluctuation range of the theoretical target volume is 0-100 decibels, the range can be divided into five levels. : 0-20 is the first volume level, 20-40 decibels is the second volume level, 40-60 decibels is the third volume level, 60-80 decibels is the fourth volume level, 80-100 decibels is the Volume five.
每个音量档位对应一个参考音量，将该训练完成隐式预测模型输出的音量所属的音量档位对应的参考音量，作为目标音量。例如，音量档位一档对应的参考音量为10分贝，音量档位二档对应的参考音量为30分贝，音量档位三档对应的参考音量为50分贝，音量档位四档对应的参考音量为60分贝，音量档位五档对应的参考音量为70分贝。若该训练完成隐式预测模型输出的音量为35分别，则属于音量档位二档，音量档位二档对应的参考音量50分贝即为目标音量。Each volume level corresponds to a reference volume, and the reference volume corresponding to the volume level to which the volume output by the implicit prediction model after training belongs is taken as the target volume. For example, the reference volume corresponding to the first volume gear is 10 decibels, the reference volume corresponding to the second volume gear is 30 decibels, the reference volume corresponding to the third volume gear is 50 decibels, and the reference volume corresponding to the fourth volume gear is is 60 decibels, and the reference volume corresponding to the fifth gear of the volume level is 70 decibels. If the volume output by the implicit prediction model after the training is 35, it belongs to the second volume gear, and the reference volume corresponding to the second volume gear is 50 decibels, which is the target volume.
像这样，通过预设多个音量档位及各档位对应的参考音量，可以防止目标音量对高频环境音的改变过于敏感，避免目标音量的改变过于频繁。Like this, by presetting multiple volume levels and the reference volumes corresponding to each level, it is possible to prevent the target volume from being too sensitive to changes in high-frequency ambient sounds, and avoid changing the target volume too frequently.
305、将待输出音频的音量调节为该目标音量。305. Adjust the volume of the audio to be output to the target volume.
可选的，于本申请其他一实施例中，为进一步提高待输出音频的播放效果，达到主动隔音的效果，在将音量样本数据输入预测模型进行训练，得到目标音量，将待输出音频的音量调节为目标音量的同时，还可以提取环境声音数据中的低频环境声音数据，获取与该低频环境声音数据对应的低频隔音音频并输出。Optionally, in another embodiment of the present application, in order to further improve the playback effect of the audio to be output and achieve the effect of active sound insulation, the volume sample data is input into the prediction model for training to obtain the target volume, and the volume of the audio to be output While adjusting the volume to the target volume, the low-frequency environmental sound data in the environmental sound data can also be extracted, and the low-frequency sound-proof audio corresponding to the low-frequency environmental sound data can be obtained and output.
具体的，该低频环境声音数据为频率小于预设频率阈值的环境声音数据。环境声音数据为环境音的响度的时间序列，对该时间序列进行傅立叶变换即可得到关于环境音的功率谱密度的频率序列，在该频率序列中提取频率小于预设频率阈值的功率谱密度谱线，并对该谱线进行逆傅里叶变换，即可得到低频环境声音数据。Specifically, the low-frequency ambient sound data is ambient sound data with a frequency lower than a preset frequency threshold. The environmental sound data is the time series of the loudness of the environmental sound, and the frequency series of the power spectral density of the environmental sound can be obtained by Fourier transforming the time series, and the power spectral density spectrum whose frequency is less than the preset frequency threshold is extracted from the frequency series line, and perform inverse Fourier transform on the spectral line to obtain low-frequency ambient sound data.
对该低频环境声音数据进行实时计算，获取与该低频环境声音数据对应的低频隔音音频并输出。Real-time calculation is performed on the low-frequency environmental sound data, and low-frequency sound-proof audio corresponding to the low-frequency environmental sound data is obtained and output.
该低频隔音音频的音量的绝对值与低频环境声音数据的音量的绝对值相等，该低频隔音音频的正负号与低频环境声音数据的正负号相反，该低频隔音音频用于抵消低频环境音，达到主动隔音的效果。The absolute value of the volume of the low-frequency sound-isolating audio is equal to the absolute value of the volume of the low-frequency ambient sound data, the sign of the low-frequency sound-isolating audio is opposite to that of the low-frequency ambient sound data, and the low-frequency sound-isolating audio is used to offset the low-frequency ambient sound , to achieve the effect of active sound insulation.
需要说明的是，由于需要运算处理，在提取得到低频环境声音数据与输出低频隔音音频之间存在时间延迟，即低频环境声音数据与低频隔音音频之间存在相位差，相较于低频环境声音数据的周期，该相位差的大小可以忽略不计，故该相位差对低频隔音音频的主动隔音效果的影响非常小。It should be noted that due to the need for calculation processing, there is a time delay between the extracted low-frequency ambient sound data and the output of low-frequency sound-proof audio, that is, there is a phase difference between low-frequency ambient sound data and low-frequency sound-proof audio. Compared with low-frequency ambient sound data The phase difference is negligible, so the phase difference has very little influence on the active sound insulation effect of the low frequency sound insulation audio.
但对于高频环境声音数据，若在提取该高频环境声音数据后输出高频隔音音频，该高频隔音音频的音量的绝对值与高频环境声音数据的音量的绝对值相等，该高频隔音音频的正负号与高频环境声音数据的正负号相反，即利用该高频隔音音频抵消高频环境音，达到主动隔音的效果。But for high-frequency ambient sound data, if after extracting this high-frequency ambient sound data, output high-frequency sound-proof audio, the absolute value of the volume of this high-frequency sound-proof audio is equal to the absolute value of the volume of high-frequency ambient sound data, and the high-frequency The sign of the sound-proof audio is opposite to that of the high-frequency ambient sound data, that is, the high-frequency sound-proof audio is used to offset the high-frequency ambient sound to achieve the effect of active sound insulation.
由于需要处理计算，在提取该高频环境声音数据与输出高频隔音音频之间存在时间延时，即该高频环境声音数据与该高频隔音音频之间存在相位差，相较于高频环境声音数据的周期，该较为差不可忽略，该相位差对该高频隔音音频的隔音效果有很大的影响，故利用高频隔音音频主动隔音的效果不如利用低频隔音音频主动隔音的效果好。Due to the need for processing calculations, there is a time delay between extracting the high-frequency ambient sound data and outputting the high-frequency sound-proof audio, that is, there is a phase difference between the high-frequency ambient sound data and the high-frequency sound-proof audio, compared to the high-frequency The cycle of ambient sound data, the phase difference can not be ignored, the phase difference has a great impact on the sound insulation effect of the high-frequency sound-proof audio, so the active sound-proof effect of using high-frequency sound-proof audio is not as good as that of using low-frequency sound-proof audio. .
在实际应用中，利用低频隔音音频主动隔音，并根据高频环境声音数据调节待输出音频的音量，利用待输出音频覆盖该高频环境声音数据，能够防止低频环境声音数据对显式预测模型的训练的影响。In practical applications, using low-frequency sound-proof audio to actively isolate sound, and adjusting the volume of the audio to be output according to the high-frequency environmental sound data, and using the audio to be output to cover the high-frequency environmental sound data can prevent the low-frequency environmental sound data from affecting the explicit prediction model. The effect of training.
下面以将本实施例提供的音量调节方法应用于一个耳机上为例进行说明，并非对本实施例提供的音量调节方法进行任何形式的限制。The following uses an example of applying the volume adjustment method provided in this embodiment to an earphone for illustration, and does not limit the volume adjustment method provided in this embodiment in any form.
在火车车厢内环境嘈杂，既有乘客说话的声音又有火车车轮通过铁轨连接处产生的噪声，为了免受火车车厢内嘈杂环境的干扰，用户将耳机插入多媒体播放设备的而耳机孔中，利用该耳机播放音频，并增大将该耳机播放音频的音量。The environment in the train compartment is noisy, there are both the voice of passengers talking and the noise generated by the train wheels passing through the junction of the rails. The headset plays audio and increases the volume of the audio played on the headset.
该耳机的处理器检测到用户的音量设置操作，获取该音量设置操作的音量设置数据，同时根据该音量设置数据，将待输出音频的音量调节为该音量设置数据对应的音量。然后采集音量改变后预设时长内的环境声音数据。The processor of the earphone detects the user's volume setting operation, acquires the volume setting data of the volume setting operation, and simultaneously adjusts the volume of the audio to be output to the volume corresponding to the volume setting data according to the volume setting data. Then collect ambient sound data within a preset time period after the volume is changed.
然后处理器提取环境声音数据中的低频环境声音数据。具体的，该低频环境声音数据为频率小于1000赫兹的环境声音数据。环境声音数据为环境音的响度的时间序列，对该时间序列进行傅立叶变换即可得到关于环境音的功率谱密度的频率序列，在该频率序列中提取频率小于1000赫兹的功率谱密度的谱线，并对该谱线进行逆傅里叶变换，即可得到低频环境声音数据。The processor then extracts low frequency ambient sound data from the ambient sound data. Specifically, the low-frequency ambient sound data is ambient sound data with a frequency less than 1000 Hz. The environmental sound data is the time series of the loudness of the environmental sound, and the frequency series of the power spectral density of the environmental sound can be obtained by Fourier transforming the time series, and the spectral lines of the power spectral density with a frequency less than 1000 Hz are extracted from the frequency series , and the inverse Fourier transform of the spectral line can be obtained to obtain low-frequency environmental sound data.
该处理器在提取低频环境声音数据后，对该低频环境声音数据进行实时计算，获取与该低频环境声音数据对应的低频隔音音频并通过该耳机的听筒进行输出。该低频隔音音频的音量与低频环境声音数据的音量的绝对值相等，该低频隔音音频的与该低频环境声音数据的音量的正负号相反。该低频隔音音频用于抵消低频环境音，达到主动隔音的效果。After extracting the low-frequency environmental sound data, the processor performs real-time calculation on the low-frequency environmental sound data, acquires low-frequency sound-proof audio corresponding to the low-frequency environmental sound data, and outputs it through the earpiece of the earphone. The volume of the low-frequency sound-insulating audio is equal to the absolute value of the volume of the low-frequency ambient sound data, and the sign of the volume of the low-frequency sound-insulating audio is opposite to that of the low-frequency ambient sound data. The low-frequency sound-isolating audio is used to offset the low-frequency ambient sound to achieve the effect of active sound insulation.
在提取环境声音数据中的低频环境声音数据的同时，该处理器还提取该环境声音数据中的高频环境声音数据。具体的，该高频环境声音数据为频率大于1000赫兹的环境声音数据。环境声音数据为环境音的响度的时间序列，对该时间序列进行傅立叶变换即可得到关于环境音的功率谱密度的频率序列，在该频率序列中提取频率大于1000赫兹的功率谱密度的谱线，并对该谱线进行逆傅里叶变换，即可得到高频环境声音数据。While extracting low frequency ambient sound data from the ambient sound data, the processor also extracts high frequency ambient sound data from the ambient sound data. Specifically, the high-frequency ambient sound data is ambient sound data with a frequency greater than 1000 Hz. The environmental sound data is the time series of the loudness of the environmental sound, and the frequency series of the power spectral density of the environmental sound can be obtained by Fourier transforming the time series, and the spectral lines of the power spectral density with a frequency greater than 1000 Hz are extracted from the frequency series , and the inverse Fourier transform of the spectral line can be obtained to obtain high-frequency ambient sound data.
然后该处理器将该高频环境声音数据输入隐式预测模型中，该隐式预测模型为神经网络模型，请参阅图4，图4为一个神经网络模型的结构示意图，如图4所示，该神经网络包括一个输入层、一个隐含层和一个输出层，该输入层包括一个神经节点x1，该隐含层包括三个神经节点x2、x3、x4，该输出层包括一个神经节点x5，该神经网络为全互联神经网络。需要说明的是图4所示的神经网络仅为一个神经网络模型的示例，并非对本实施例提供的隐式预测模型进行任何形式的限制。连接神经节点x1和神经节点x2的边的连接权重为M1，连接神经节点x1和神经节点x2的边的连接权重为M2，连接神经节点x1和神经节点x3的边的连接权重为M3，连接神经节点x2和神经节点x5的边的连接权重为M4，连接神经节点x3和神经节点x5的边的连接权重为M5，连接神经节点x4和神经节点x5的边的连接权重为M6。将输入层称为隐含层的上一层，将隐含层称为输出层的上一层。将高频环境声音数据以x1的形式输入该神经网络模型中，则根据公式：Then the processor inputs the high-frequency ambient sound data into the implicit prediction model, the implicit prediction model is a neural network model, please refer to Figure 4, Figure 4 is a schematic structural diagram of a neural network model, as shown in Figure 4, The neural network includes an input layer, a hidden layer and an output layer, the input layer includes a neural node x1 , the hidden layer includes three neural nodes x2 , x3 , x4 , the output layer includes a Neural node x5 , the neural network is a fully interconnected neural network. It should be noted that the neural network shown in FIG. 4 is only an example of a neural network model, and does not limit the implicit prediction model provided by this embodiment in any form. The connection weight of the edge connecting neural node x1 and neural node x2 is M1 , the connection weight of the edge connecting neural node x1 and neural node x2 is M2 , and the edge connecting neural node x1 and neural node x3 The weight of the connection is M3 , the connection weight of the edge connecting the neural node x2 and the neural node x5 is M4 , the connection weight of the edge connecting the neural node x3 and the neural node x5 is M5 , and the connection weight of the edge connecting the neural node x4 The connection weight of the edge with neural node x5 is M6 . The input layer is called the layer above the hidden layer, and the hidden layer is called the layer above the output layer. Input the high-frequency ambient sound data into the neural network model in the form of x1 , then according to the formula:
计算该神经网络模型中其他节点的数值，式(5)中，xj为一个神经节点，xi为与xj连接的上一层节点，Mi为连接神经节点xj与神经节点xi的边的连接权重，f(xi)神经节点xi的阈值函数，该阈值函数为：Calculate the values of other nodes in the neural network model. In formula (5), xj is a neural node, xi is the upper layer node connected to xj , and Mi is the connection between neural node xj and neural node xi The connection weight of the edge of f(xi ) the threshold function of neural node xi , the threshold function is:
该阈值函数用于根据神经节点的大小改变节点的连接关系。The threshold function is used to change the connection relationship of nodes according to the size of neural nodes.
将高频环境声音数据以x1的形式输入该神经网络模型中，则将式(6)带入式(5)中即可得到该神经网络模型输出的音量x5：Input the high-frequency ambient sound data into the neural network model in the form of x1 , then put formula (6) into formula (5) to get the volume x5 output by the neural network model:
x5＝M4f(M1f(x1))+M5f(M2f(x1))+M6f(M3f(x1)) (7)x5 ＝M4 f(M1 f(x1 ))+M5 f(M2 f(x1 ))+M6 f(M3 f(x1 )) (7)
若在该火车车厢的嘈杂环境内，用户设置该耳机待输出音频的音量为x0，即设置数据为x0，则该神经网络模型的损失值为：If in the noisy environment of the train carriage, the user sets the volume of the audio to be output by the earphone to x0 , that is, sets the data to x0 , then the loss value of the neural network model is:
ε＝(x5-x0)2 (8)ε＝(x5 -x0 )2 (8)
下面以混沌搜索法搜索该神经网络模型的目标连接关系和目标连接权重为例对搜索隐式预测模型的目标连接关系和目标连接权重为例进行说明，并非对搜索隐式预测模型的目标连接关系和目标连接权重的方式进行任何形式的限制。The following uses the chaotic search method to search for the target connection relationship and target connection weight of the neural network model as an example to illustrate the search for the target connection relationship and target connection weight of the implicit prediction model, not to search for the target connection relationship of the implicit prediction model Any restrictions on the way the weights are connected to the target.
首先随机选取一组连接权重，记为[M]，[M]＝[M1,M2,M3,M4,M5]，以预设步长改变该连接权重，产生五组第一代搜索连接权重：First, a group of connection weights is randomly selected, recorded as [M], [M]=[M1 , M2 , M3 , M4 , M5 ], and the connection weights are changed with a preset step size to generate five groups of first Generation Search Connection Weights:
式(9)中ΔM为该预设步长，该预设步长满足：In formula (9), ΔM is the preset step size, and the preset step size satisfies:
然后将这五组第一代搜索连接权重输入一个混沌系统中，产生五组第一代混沌搜索连接权重，该混沌系统例如可以为logistic映射系统：Then these five groups of first-generation search connection weights are input into a chaotic system to generate five groups of first-generation chaotic search connection weights. The chaotic system can be, for example, a logistic mapping system:
[N]n+1＝μ[N]n{1-[N]n} (11)[N]n+1 = μ[N]n {1-[N]n } (11)
式(11)中[N]i为归一混沌搜索连接权重，由于logistic映射的定义域为(0,1)，故需要利用公式：[N]i in formula (11) is the weight of the normalized chaotic search connection. Since the domain of definition of the logistic map is (0,1), it is necessary to use the formula:
将搜索连接权归一化。Normalize search join weights.
当μ为3.8时，logistic映射系统为混沌系统，将归一化的搜索连接权重输入logistic映射系统中得到归一混沌搜索连接权重，再将归一混沌搜索连接权重根据公式：When μ is 3.8, the logistic mapping system is a chaotic system, and the normalized search connection weight is input into the logistic mapping system to obtain the normalized chaotic search connection weight, and then the normalized chaotic search connection weight is calculated according to the formula:
[C]i＝[N]i[max(M1,M2,M3,M4,M5)+ΔM] (13)[C]i =[N]i [max(M1 ,M2 ,M3 ,M4 ,M5 )+ΔM] (13)
将归一混沌搜索连接权重还原为混沌搜索连接权重，得到五组第一代混沌搜索连接权重。式(13)中[C]i为混沌搜索连接权重。由于混沌系统具有对初始条件的敏感性和有界性，故产生的混沌搜索连接权重可以遍历所有连接权重，即在所有连接权重中搜索目标连接权重。The normalized chaos search connection weights are restored to the chaos search connection weights, and five sets of first-generation chaos search connection weights are obtained. [C]i in formula (13) is the connection weight of chaos search. Since the chaotic system is sensitive and bounded to the initial conditions, the generated chaotic search connection weights can traverse all connection weights, that is, search the target connection weights in all connection weights.
将得到的五组第一代混沌搜索连接权重带入隐式预测模型中，得到五个隐式预测模型，再将高频环境音依次数据输入这五个隐式预测模型中，得到这五个隐式预测模型的损失值，将这五个隐式预测模型中具有最小的损失值的隐式预测模型的连接权重作为第一代最优解。Bring the obtained five sets of first-generation chaotic search connection weights into the implicit prediction model to obtain five implicit prediction models, and then input the high-frequency ambient sound data into the five implicit prediction models in sequence to obtain the five implicit prediction models. The loss value of the implicit prediction model, the connection weight of the implicit prediction model with the smallest loss value among the five implicit prediction models is taken as the first generation optimal solution.
将该第一代最优解带入式(9)中，并根据式(11)至式(13)得到第二代混沌搜索连接权重，将得到的第二代五组混沌搜索连接权重带入隐式预测模型中，得到五个隐式预测模型，再将高频环境音依次数据输入这五个隐式预测模型中，得到这五个隐式预测模型的损失值，将这五个隐式预测模型中具有最小的损失值的隐式预测模型的连接权重作为第二代最优解。Bring the first-generation optimal solution into Equation (9), and obtain the second-generation chaotic search connection weights according to Equations (11) to (13), and bring the obtained second-generation five groups of chaotic search connection weights into In the implicit prediction model, five implicit prediction models are obtained, and then the high-frequency ambient sound data is input into the five implicit prediction models in sequence, and the loss values of the five implicit prediction models are obtained, and the five implicit prediction models are The connection weight of the implicit prediction model with the smallest loss value in the prediction model is used as the second-generation optimal solution.
迭代将该第N代最优解带入式(9)中，并根据式(11)至式(13)得到第N+1代混沌搜索连接权重，将得到的第N+1代五组混沌搜索连接权重带入隐式预测模型中，得到五个隐式预测模型，再将高频环境音依次数据输入这五个隐式预测模型中，得到这五个隐式预测模型的损失值，将这五个隐式预测模型中具有最小的损失值的隐式预测模型的连接权重作为第N+1代最优解。直至以第N+1代最优解为连接权重的隐式预测模型的损失值不小于以第N代最优解为连接权重的隐式预测模型的损失值，则该第N+1代最优解即为目标连接权重，以该N+1代最优解为连接权重的隐式预测模型的连接关系即为目标连接关系。Iteratively bring the optimal solution of the Nth generation into Equation (9), and obtain the N+1th generation chaotic search connection weights according to Equation (11) to Equation (13), the obtained N+1th generation five groups of chaos The search connection weights are brought into the implicit prediction model to obtain five implicit prediction models, and then the high-frequency ambient sound data is input into the five implicit prediction models in sequence to obtain the loss values of the five implicit prediction models. The connection weight of the implicit prediction model with the smallest loss value among the five implicit prediction models is taken as the optimal solution of the N+1th generation. Until the loss value of the implicit prediction model with the optimal solution of the N+1th generation as the connection weight is not less than the loss value of the implicit prediction model with the optimal solution of the Nth generation as the connection weight, then the N+1st generation’s most The optimal solution is the target connection weight, and the connection relationship of the implicit prediction model with the N+1 generation optimal solution as the connection weight is the target connection relationship.
为了避免偶然性数据对该隐式预测模型的训练的影响，将二十个高频环境声音数据输入该隐式预测模型中，若该隐式预测模型的损失值的平均值小于预设的阈值，则认为该隐式预测模型为训练完成隐式预测模型。In order to avoid the impact of accidental data on the training of the implicit prediction model, input twenty high-frequency environmental sound data into the implicit prediction model, if the average value of the loss value of the implicit prediction model is less than the preset threshold, Then it is considered that the implicit prediction model is a training completed implicit prediction model.
可选的，为避免目标音量的改变过于频繁，还可以分析该训练完成隐式预测模型输出的音量所属的音量档位，获取该音量档位对应的参考音量作为目标音量。Optionally, in order to prevent the target volume from changing too frequently, it is also possible to analyze the volume level to which the volume output by the implicit prediction model after training belongs, and obtain the reference volume corresponding to the volume level as the target volume.
最后该处理器将待输出音频的音量调节为该目标音量即可实现根据用户的习惯调节耳机播放音频的音量。Finally, the processor adjusts the volume of the audio to be output to the target volume, so that the volume of the audio played by the earphone can be adjusted according to the user's habit.
在本实施例中，第一方面，根据音量样本数据对预测模型进行训练，得到目标音量，预测模型的训练结束条件为损失值小于预设的阈值，所以通过该预测模型训练后得到的目标音量与音量设置数据之间的差的欧几里得范数小于预设的阈值，将待输出音频的音量调节为目标音量，在该音量样本数据中包含用户音量设置数据，因此待输出音频的音量与用户音量设置数据相差较小，更符合用户的使用习惯，提高音量设置的个性化和智能化程度。第二方面，由于根据高频环境声音数据对隐式预测模型进行训练，故可以避免低频环境声音数据对隐式预测模型的训练的影响；第三方面，由于预测模型为隐式预测模型，该隐式预测模型的输入数据和输出数据之间的关系可以由通过网络连接关系与连接权重表示，隐式预测模型表示的输入数据与输出数据之间的关系更加灵活，故隐式预测模型的训练速度快且训练完成后损失值小。第四方面，由于将训练完成显式预测模型输出的音量所属的音量档位，获取该音量档位对应的参考音量作为目标音量，故可以防止目标音量对高频环境音的改变过于敏感，避免目标音量的改变过于频繁，进而可以避免待输出音频的音量的改变过于频繁。In this embodiment, in the first aspect, the prediction model is trained according to the volume sample data to obtain the target volume. The training end condition of the prediction model is that the loss value is less than the preset threshold, so the target volume obtained after training the prediction model The Euclidean norm of the difference with the volume setting data is less than the preset threshold, and the volume of the audio to be output is adjusted to the target volume. The volume sample data contains the user volume setting data, so the volume of the audio to be output The difference from the volume setting data of the user is small, which is more in line with the user's usage habits, and the degree of personalization and intelligence of the volume setting is improved. In the second aspect, since the implicit prediction model is trained according to the high-frequency ambient sound data, the influence of the low-frequency ambient sound data on the training of the implicit prediction model can be avoided; in the third aspect, since the prediction model is an implicit prediction model, the The relationship between the input data and the output data of the implicit prediction model can be represented by the network connection relationship and the connection weight. The relationship between the input data and the output data represented by the implicit prediction model is more flexible, so the training of the implicit prediction model The speed is fast and the loss value is small after the training is completed. In the fourth aspect, since the volume gear of the volume output by the explicit prediction model after training is completed, the reference volume corresponding to the volume gear is obtained as the target volume, so it can prevent the target volume from being too sensitive to changes in high-frequency environmental sounds, and avoid The change of the target volume is too frequent, thereby preventing the volume of the audio to be output from changing too frequently.
请参阅图5，图5为本申请一实施例提供的电子装置的结构示意图，该电子装置包括：Please refer to FIG. 5. FIG. 5 is a schematic structural diagram of an electronic device provided by an embodiment of the present application. The electronic device includes:
获取模块410，用于获取音量样本数据，该音量样本数据包括环境声音数据和音量设置数据。The acquiring module 410 is configured to acquire volume sample data, where the volume sample data includes ambient sound data and volume setting data.
训练模块420，用于将该音量样本数据输出预测模型进行训练，得到目标音量，该预测模型的训练结束条件为损失值小于预设的阈值。The training module 420 is configured to output the volume sample data to a prediction model for training to obtain the target volume. The training end condition of the prediction model is that the loss value is less than a preset threshold.
调节模块430，用于将待输出音频的音量调节为该目标音量。An adjustment module 430, configured to adjust the volume of the audio to be output to the target volume.
在本实施例中，由于根据音量样本数据对预测模型进行训练，得到目标音量，并将待输出音频的音量调节为目标音量，又因为该目标音量与音量设置数据之间的差的欧几里得范数小于预设的阈值，故本实施例提供的音量调节方法，可以使待输出音频的音量更符合用户的使用习惯，提高音量设置的个性化和智能化程度。In this embodiment, because the prediction model is trained according to the volume sample data, the target volume is obtained, and the volume of the audio to be output is adjusted to the target volume, and because the Euclidean value of the difference between the target volume and the volume setting data The resulting norm is smaller than the preset threshold, so the volume adjustment method provided by this embodiment can make the volume of the audio to be output more in line with the user's usage habits, and improve the degree of personalization and intelligence of the volume setting.
请参阅图6，图6为本申请另一实施例提供的电子装置的结构示意图，如图6所示，与前述图5所示的电子装置不同的是，在本实施例中：Please refer to FIG. 6. FIG. 6 is a schematic structural diagram of an electronic device provided by another embodiment of the present application. As shown in FIG. 6, different from the electronic device shown in FIG. 5, in this embodiment:
进一步地，该电子装置还包括：Further, the electronic device also includes:
低频提取模块440，用于提取环境声音数据中的低频环境声音数据，该低频环境声音数据为频率小于预设频率阈值的环境声音数据。The low-frequency extraction module 440 is configured to extract low-frequency environmental sound data from the environmental sound data, where the low-frequency environmental sound data is environmental sound data with a frequency lower than a preset frequency threshold.
输出模块450，用于获取与该低频环境声音数据对应的低频隔音音频并输出，该低频隔音音频的音量的绝对值与低频环境声音数据的音量的绝对值相等，该低频隔音音频的正负号与该低频环境声音数据的正负号相反。The output module 450 is used to obtain and output the low-frequency sound-proof audio corresponding to the low-frequency environmental sound data. The absolute value of the volume of the low-frequency sound-proof audio is equal to the absolute value of the volume of the low-frequency environmental sound data. The opposite sign of the low-frequency ambient sound data.
进一步地，预测模型为显式预测模型，该显式预测模型为以固定的关系式表示输入数据与输出数据之间的关系的预测模型，则训练模块420包括：Further, the prediction model is an explicit prediction model, and the explicit prediction model is a prediction model that expresses the relationship between input data and output data with a fixed relational expression, then the training module 420 includes:
提取模块421，用于提取环境声音数据中的高频环境声音数据，并将该高频环境声音数据输入该显式预测模型中，该高频环境声音数据为频率大于预设频率阈值的环境声音数据。An extraction module 421, configured to extract high-frequency environmental sound data from the environmental sound data, and input the high-frequency environmental sound data into the explicit prediction model, where the high-frequency environmental sound data is environmental sound with a frequency greater than a preset frequency threshold data.
搜索模块422，用于搜索显式预测模型中固定的关系式的目标系数，该目标指数使该显式预测模型的损失值取到最小，该损失值为该显式预测模型输出的音量与音量设置数据的差的欧几里得范数。The search module 422 is used to search for the target coefficient of the fixed relational expression in the explicit prediction model, and the target index minimizes the loss value of the explicit prediction model, and the loss value is the output volume and volume of the explicit prediction model Sets the Euclidean norm of the difference of the data.
分析模块423，用于在每次搜索代目标系数后，分析显式预测模型的损失值是否小于预设的阈值，并当该显式预测模型的损失值小于预设的阈值时，确认该显式预测模型为训练完成显式预测模型，将该训练完成显式预测模型输出的音量确认为目标音量。The analysis module 423 is used to analyze whether the loss value of the explicit prediction model is less than the preset threshold value after each search for the generation target coefficient, and confirm that the explicit prediction model loss value is less than the preset threshold value. The explicit prediction model is an explicit prediction model after training, and the volume output by the explicit prediction model after training is confirmed as the target volume.
或者预测模型为隐式预测模型，该隐式预测模型为以网络连接关系和连接权重表示输入数据与输出数据之间的关系的预测模型，则Or the prediction model is an implicit prediction model, and the implicit prediction model is a prediction model that expresses the relationship between input data and output data by network connection relationship and connection weight, then
提取模块421，还用于提取环境声音数据中的高频环境声音数据，并将该高频环境声音数据输入该隐式预测模型中，该高频环境声音数据为频率大于预设频率阈值的环境声音数据。The extraction module 421 is also used to extract high-frequency environmental sound data from the environmental sound data, and input the high-frequency environmental sound data into the implicit prediction model. The high-frequency environmental sound data is an environment whose frequency is greater than a preset frequency threshold sound data.
搜索模块422，还用于搜索该隐式预测模型的目标连接关系和目标连接权重，该目标连接关系和该目标连接权重使该隐式预测模型的损失值取到最小值，该损失值为该隐式预测模型输出的音量与音量设置数据的差的欧几里得范数。The search module 422 is also used to search for the target connection relationship and the target connection weight of the implicit prediction model, the target connection relationship and the target connection weight make the loss value of the implicit prediction model take the minimum value, and the loss value is the The Euclidean norm of the difference between the volume output by the implicit predictive model and the volume setting data.
分析模块423，还用于在每次搜索到目标连接关系和目标连接权重后，分析该隐式预测模型的损失值是否小于预设的阈值，并当该隐式预测模型的损失值小于预设的阈值时，确认该隐式预测模型为训练完成隐式预测模型，将该训练完成隐式预测模型输出的音量确认为目标音量调节数据。The analysis module 423 is also used to analyze whether the loss value of the implicit prediction model is less than a preset threshold after each search for the target connection relationship and the target connection weight, and when the loss value of the implicit prediction model is less than the preset threshold When the threshold is , it is confirmed that the implicit prediction model is the trained implicit prediction model, and the volume output by the trained implicit prediction model is confirmed as the target volume adjustment data.
或者，分析模块423，还用于分析预测模型输出的音量所属的音量档位，获取该音量档位对应的参考音量作为目标音量。Alternatively, the analysis module 423 is further configured to analyze the volume level to which the volume output by the prediction model belongs, and obtain a reference volume corresponding to the volume level as the target volume.
在本实施例中，第一方面，根据音量样本数据对预测模型进行训练，得到目标音量，预测模型的训练结束条件为损失值小于预设的阈值，所以通过该预测模型训练后得到的目标音量与音量设置数据之间的差的欧几里得范数小于预设的阈值，将待输出音频的音量调节为目标音量，在该音量样本数据中包含用户音量设置数据，因此待输出音频的音量与用户音量设置数据相差较小，更符合用户的使用习惯，提高音量设置的个性化和智能化程度；第二方面，由于根据高频环境声音数据对隐式预测模型进行训练，故可以避免低频环境声音数据对隐式预测模型的训练的影响。第三方面，由于在预设的目标音量调节数据的大小的范围内将该目标音量调节数据划分为至少两个音量档位，并根据音量档位输出的音量数据确定目标音量，故可以防止目标音量对高频环境音的改变过于敏感，避免目标音量的改变过于频繁，进而可以避免待输出音频的音量的改变过于频繁。In this embodiment, in the first aspect, the prediction model is trained according to the volume sample data to obtain the target volume. The training end condition of the prediction model is that the loss value is less than the preset threshold, so the target volume obtained after training the prediction model The Euclidean norm of the difference with the volume setting data is less than the preset threshold, and the volume of the audio to be output is adjusted to the target volume. The volume sample data contains the user volume setting data, so the volume of the audio to be output The difference from the user's volume setting data is small, which is more in line with the user's usage habits, and the degree of personalization and intelligence of the volume setting is improved; secondly, since the implicit prediction model is trained according to high-frequency environmental sound data, low-frequency noise can be avoided. The effect of ambient sound data on the training of implicit predictive models. In the third aspect, since the target volume adjustment data is divided into at least two volume levels within the size range of the preset target volume adjustment data, and the target volume is determined according to the volume data output by the volume levels, it is possible to prevent the target The volume is too sensitive to the change of high-frequency ambient sound, avoiding too frequent changes of the target volume, and thus avoiding too frequent changes of the volume of the audio to be output.
请参阅图7，图7为本申请实施例提供的电子装置的硬件结构示意图，该电子装置包括存储器501、处理器502及存储在存储器501上并可在处理器502上运行的计算机程序，处理器502执行该计算机程序时，实现图1至图3所示的音量调节方法。Please refer to FIG. 7. FIG. 7 is a schematic diagram of the hardware structure of the electronic device provided by the embodiment of the present application. The electronic device includes a memory 501, a processor 502, and a computer program stored in the memory 501 and operable on the processor 502. When the computer program is executed by the computer 502, the volume adjustment methods shown in FIGS. 1 to 3 are realized.
进一步地，该电子装置还包括：Further, the electronic device also includes:
至少一个输入设备503；至少一个输出设备504。at least one input device 503; at least one output device 504.
上述处理器501、存储器502、输入设备503和输出设备504通过总线505连接。The aforementioned processor 501 , memory 502 , input device 503 and output device 504 are connected through a bus 505 .
其中输入设备503例如可以是麦克风、噪声传感器等可以将音频信号转化为电信号的设备。输出设备503例如可以是耳机、功放喇叭等可以将电信号转化为音频信号的设备。The input device 503 may be, for example, a microphone, a noise sensor, and other devices capable of converting audio signals into electrical signals. The output device 503 may be, for example, a device capable of converting an electrical signal into an audio signal, such as an earphone or a power amplifier speaker.
进一步地，本申请实施例还提供了一种计算机可读存储介质，该计算机可读存储介质可以是设置于上述各实施例中的移动终端中，该计算机可读存储介质可以是前述图7所示实施例中的存储器。该计算机可读存储介质上存储有计算机程序，该程序被处理器执行时实现前述图1至图3所示实施例中描述的网络切换方法。进一步地，该计算机可存储介质还可以是U盘、移动硬盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、磁碟或者光盘等各种可以存储程序代码的介质。Further, the embodiment of the present application also provides a computer-readable storage medium, which can be set in the mobile terminal in each of the above-mentioned embodiments, and the computer-readable storage medium can be the memory in the example embodiment. A computer program is stored on the computer-readable storage medium, and when the program is executed by a processor, the network switching method described in the foregoing embodiments shown in FIG. 1 to FIG. 3 is implemented. Further, the computer storage medium can also be a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, etc., which can store medium for program code.
在本申请所提供的多个实施例中，应该理解到，所揭露的电子装置和方法，可以通过其它的方式实现。例如，以上所描述的实施例仅仅是示意性的，例如，所述模块的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个模块或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信链接可以是通过一些接口，模块的间接耦合或通信链接，可以是电性，机械或其它的形式。In the multiple embodiments provided in this application, it should be understood that the disclosed electronic device and method can be implemented in other ways. For example, the above-described embodiments are only illustrative. For example, the division of the modules is only a logical function division. In actual implementation, there may be other division methods. For example, multiple modules or components can be combined or can be Integrate into another system, or some features may be ignored, or not implemented. In another point, the mutual coupling or direct coupling or communication link shown or discussed may be through some interfaces, and the indirect coupling or communication link between modules may be in electrical, mechanical or other forms.
所述作为分离部件说明的模块可以是或者也可以不是物理上分开的，作为模块显示的部件可以是或者也可以不是物理模块，即可以位于一个地方，或者也可以分布到多个网络模块上。可以根据实际的需要选择其中的部分或者全部模块来实现本实施例方案的目的。The modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place, or may be distributed to multiple network modules. Part or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.
另外，在本申请各个实施例中的各功能模块可以集成在一个处理模块中，也可以是各个模块单独物理存在，也可以两个或两个以上模块集成在一个模块中。上述集成的模块既可以采用硬件的形式实现，也可以采用软件功能模块的形式实现。In addition, each functional module in each embodiment of the present application may be integrated into one processing module, each module may exist separately physically, or two or more modules may be integrated into one module. The above-mentioned integrated modules can be implemented in the form of hardware or in the form of software function modules.
需要说明的是，对于前述的各方法实施例，为了简便描述，故将其都表述为一系列的动作组合，但是本领域技术人员应该知悉，本发明并不受所描述的动作顺序的限制，因为依据本发明，某些步骤可以采用其它顺序或者同时进行。其次，本领域技术人员也应该知悉，说明书中所描述的实施例均属于优选实施例，所涉及的动作和模块并不一定都是本发明所必须的。It should be noted that, for the sake of simplicity of description, the aforementioned method embodiments are expressed as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described action sequence. Because of the present invention, certain steps may be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification belong to preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.
在上述实施例中，对各个实施例的描述都各有侧重，某个实施例中没有详述的部分，可以参见其它实施例的相关描述。In the foregoing embodiments, the descriptions of each embodiment have their own emphases, and for parts not described in detail in a certain embodiment, reference may be made to relevant descriptions of other embodiments.
以上为对本申请所提供的音量调节方法、电子装置及计算机可读存储介质的描述，对于本领域的技术人员，依据本申请实施例的思想，在具体实施方式及应用范围上均会有改变之处，综上，本说明书内容不应理解为对本申请的限制。The above is the description of the volume adjustment method, electronic device and computer-readable storage medium provided by this application. For those skilled in the art, according to the idea of the embodiment of this application, there will be changes in the specific implementation and application scope. In summary, the contents of this specification should not be construed as limiting the application.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810339437.6ACN108536420A (en) | 2018-04-16 | 2018-04-16 | Volume adjusting method, electronic device and computer readable storage medium |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810339437.6ACN108536420A (en) | 2018-04-16 | 2018-04-16 | Volume adjusting method, electronic device and computer readable storage medium |
| Publication Number | Publication Date |
|---|---|
| CN108536420Atrue CN108536420A (en) | 2018-09-14 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810339437.6APendingCN108536420A (en) | 2018-04-16 | 2018-04-16 | Volume adjusting method, electronic device and computer readable storage medium |
| Country | Link |
|---|---|
| CN (1) | CN108536420A (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109819375A (en)* | 2019-01-11 | 2019-05-28 | 平安科技（深圳）有限公司 | Adjust method and apparatus, storage medium, the electronic equipment of volume |
| CN110012386A (en)* | 2019-03-29 | 2019-07-12 | 维沃移动通信有限公司 | A kind of terminal volume adjustment method and terminal |
| CN110187860A (en)* | 2019-04-24 | 2019-08-30 | 北京声智科技有限公司 | Volume Fuzzy Regulating Method, device, electronic equipment and storage medium |
| CN110866877A (en)* | 2019-11-12 | 2020-03-06 | Oppo广东移动通信有限公司 | Color correction method, device and terminal device based on constrained particle swarm optimization |
| CN111986696A (en)* | 2020-08-27 | 2020-11-24 | 湖南融视文化创意有限公司 | Method for efficiently processing song volume balance |
| CN112037771A (en)* | 2020-08-28 | 2020-12-04 | 中移（杭州）信息技术有限公司 | Method, device, electronic device and storage medium for volume adjustment |
| CN114741047A (en)* | 2022-03-31 | 2022-07-12 | 中国第一汽车股份有限公司 | Volume adjusting method and volume adjusting system |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH11261355A (en)* | 1998-03-13 | 1999-09-24 | Nec Shizuoka Ltd | Automatic adjusting function system for loudspeaker sound volume and automatic adjusting method for loudspeaker sound volume |
| CN201733426U (en)* | 2009-08-15 | 2011-02-02 | 刘瑜 | Intelligent sound volume controller of automobile sound system |
| CN103345929A (en)* | 2013-06-25 | 2013-10-09 | 三星电子（中国）研发中心 | Method and device for adjusting volume |
| CN103686009A (en)* | 2012-09-19 | 2014-03-26 | 海尔集团公司 | Method and device for intelligent perception on TV |
| CN104507003A (en)* | 2014-11-28 | 2015-04-08 | 广东好帮手电子科技股份有限公司 | A method and a system for adjusting a volume intelligently according to a noise in a vehicle |
| CN106303783A (en)* | 2016-08-12 | 2017-01-04 | 北京金锐德路科技有限公司 | Noise-reduction method and device |
| CN106453945A (en)* | 2016-11-10 | 2017-02-22 | 上海传英信息技术有限公司 | Automatic external acoustical quality adjusting system for mobile phone |
| CN106648527A (en)* | 2016-11-08 | 2017-05-10 | 乐视控股（北京）有限公司 | Volume control method, device and playing equipment |
| CN106953972A (en)* | 2017-03-15 | 2017-07-14 | 上海青橙实业有限公司 | Mobile terminal and the method and system for adjusting audio file broadcast sound volume |
| CN106970774A (en)* | 2017-03-29 | 2017-07-21 | 广州阿里巴巴文学信息技术有限公司 | A kind of volume adjustment device and method, a kind of terminal |
| CN107682561A (en)* | 2017-11-10 | 2018-02-09 | 广东欧珀移动通信有限公司 | volume adjusting method, device, terminal and storage medium |
| CN107766030A (en)* | 2017-11-13 | 2018-03-06 | 百度在线网络技术（北京）有限公司 | Volume adjusting method, device, equipment and computer-readable medium |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH11261355A (en)* | 1998-03-13 | 1999-09-24 | Nec Shizuoka Ltd | Automatic adjusting function system for loudspeaker sound volume and automatic adjusting method for loudspeaker sound volume |
| CN201733426U (en)* | 2009-08-15 | 2011-02-02 | 刘瑜 | Intelligent sound volume controller of automobile sound system |
| CN103686009A (en)* | 2012-09-19 | 2014-03-26 | 海尔集团公司 | Method and device for intelligent perception on TV |
| CN103345929A (en)* | 2013-06-25 | 2013-10-09 | 三星电子（中国）研发中心 | Method and device for adjusting volume |
| CN104507003A (en)* | 2014-11-28 | 2015-04-08 | 广东好帮手电子科技股份有限公司 | A method and a system for adjusting a volume intelligently according to a noise in a vehicle |
| CN106303783A (en)* | 2016-08-12 | 2017-01-04 | 北京金锐德路科技有限公司 | Noise-reduction method and device |
| CN106648527A (en)* | 2016-11-08 | 2017-05-10 | 乐视控股（北京）有限公司 | Volume control method, device and playing equipment |
| CN106453945A (en)* | 2016-11-10 | 2017-02-22 | 上海传英信息技术有限公司 | Automatic external acoustical quality adjusting system for mobile phone |
| CN106953972A (en)* | 2017-03-15 | 2017-07-14 | 上海青橙实业有限公司 | Mobile terminal and the method and system for adjusting audio file broadcast sound volume |
| CN106970774A (en)* | 2017-03-29 | 2017-07-21 | 广州阿里巴巴文学信息技术有限公司 | A kind of volume adjustment device and method, a kind of terminal |
| CN107682561A (en)* | 2017-11-10 | 2018-02-09 | 广东欧珀移动通信有限公司 | volume adjusting method, device, terminal and storage medium |
| CN107766030A (en)* | 2017-11-13 | 2018-03-06 | 百度在线网络技术（北京）有限公司 | Volume adjusting method, device, equipment and computer-readable medium |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109819375A (en)* | 2019-01-11 | 2019-05-28 | 平安科技（深圳）有限公司 | Adjust method and apparatus, storage medium, the electronic equipment of volume |
| CN110012386A (en)* | 2019-03-29 | 2019-07-12 | 维沃移动通信有限公司 | A kind of terminal volume adjustment method and terminal |
| CN110187860A (en)* | 2019-04-24 | 2019-08-30 | 北京声智科技有限公司 | Volume Fuzzy Regulating Method, device, electronic equipment and storage medium |
| CN110866877A (en)* | 2019-11-12 | 2020-03-06 | Oppo广东移动通信有限公司 | Color correction method, device and terminal device based on constrained particle swarm optimization |
| CN111986696A (en)* | 2020-08-27 | 2020-11-24 | 湖南融视文化创意有限公司 | Method for efficiently processing song volume balance |
| CN111986696B (en)* | 2020-08-27 | 2023-07-07 | 湖南融视文化创意有限公司 | Method for efficiently processing song volume balance |
| CN112037771A (en)* | 2020-08-28 | 2020-12-04 | 中移（杭州）信息技术有限公司 | Method, device, electronic device and storage medium for volume adjustment |
| CN112037771B (en)* | 2020-08-28 | 2024-03-12 | 中移（杭州）信息技术有限公司 | Method and device for adjusting volume, electronic equipment and storage medium |
| CN114741047A (en)* | 2022-03-31 | 2022-07-12 | 中国第一汽车股份有限公司 | Volume adjusting method and volume adjusting system |
| Publication | Publication Date | Title |
|---|---|---|
| CN108536420A (en) | Volume adjusting method, electronic device and computer readable storage medium | |
| El-Moneim et al. | Text-independent speaker recognition using LSTM-RNN and speech enhancement | |
| CN112912897B (en) | Sound classification system | |
| JP2022531574A (en) | Speech recognition methods and devices, neural network training methods and devices, and computer programs | |
| CN105788592A (en) | Audio classification method and apparatus thereof | |
| CN111312273A (en) | Reverberation elimination method, apparatus, computer device and storage medium | |
| CN111357051B (en) | Speech emotion recognition method, intelligent device and computer readable storage medium | |
| WO2022012206A1 (en) | Audio signal processing method, device, equipment, and storage medium | |
| CN111402922B (en) | Audio signal classification method, device, equipment and storage medium based on small samples | |
| US20170194021A1 (en) | Crowdsourced database for sound identification | |
| CN106407960A (en) | Multi-feature-based classification method and system for music genres | |
| CN113823301A (en) | Training method and device of voice enhancement model and voice enhancement method and device | |
| CN105448302A (en) | Environment adaptive type voice reverberation elimination method and system | |
| WO2019233361A1 (en) | Method and device for adjusting volume of music | |
| WO2024027246A1 (en) | Sound signal processing method and apparatus, and electronic device and storage medium | |
| CN114582325A (en) | Audio detection method, device, computer equipment, storage medium | |
| Kim et al. | Acoustic Event Detection in Multichannel Audio Using Gated Recurrent Neural Networks with High‐Resolution Spectral Features | |
| CN114678038B (en) | Audio noise detection method, computer device and computer program product | |
| WO2025061091A1 (en) | Training method and apparatus for wake-on-voice model, executing method for wake-on-voice model, and device and storage medium | |
| CN115510299A (en) | Data classification method, model compression method, device, equipment and program product | |
| CN117976003A (en) | Cross-modal emotion analysis method, training method, device and equipment | |
| CN105989838B (en) | Speech recognition method and device | |
| CN117373468A (en) | Far-field voice enhancement processing method, far-field voice enhancement processing device, computer equipment and storage medium | |
| CN113838466A (en) | Voice recognition method, device, equipment and storage medium | |
| CN113808606A (en) | Voice signal processing method and device |
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication | Application publication date:20180914 |