CN100508031C - A Method for Recognizing and Eliminating the Echo Produced by Far-End Speech in SCDMA Mobile Phone - Google Patents

Abstract

A method for identifying and eliminating echo generated from far end voice in CDMA mobile phone includes obtaining a frame sampling signal at near end to judge whether there is voice at far end or not according last 5-15 frames at far end, detecting voice echo of said frame if it is and calculating low frequency energy of said frame signal to judge out whether only echo is existed or not then carrying out smooth on echo or replace echo by comfort noise signal if echo is existed and not varying current signal if echo is not existed, or otherwise not requiring to carry out detection on voice of said frame at near end and directly coding and transmitting voice.

Description

A kind of method of in the SCDMA mobile phone, discerning and eliminating the echo of far-end speech generation

Technical field

The present invention relates in the wireless communications application system, utilize spectral characteristic identification and eliminate a kind of method of mobile phone far-end speech the echo of near-end generation.

Background technology

In mobile phone, be derived from the sound wave of receiver (earphone), reflect into the microphone (microphone) or the speaker-phone of mobile phone by the solid objects that is arranged on the voice path, thereby form acoustic echo, cause that speech quality reduces.

At present, the method of eliminating acoustic echo mainly is based on dual end communication (Double Talk) and detects, double talk detection is the state of detection of handset current talking, be only exist far-end speech (having only receiver sound), only exist near-end speech (having only microphone sound), or there be (dual end communication) simultaneously in the far-end near-end speech.The main method that is used for double talk detection be according to Geigel algorithm-system to the decay of echo signal energy greater than 6dB-design realization.But in many practical applications, for example in the SCDMA mobile phone, echo channel does not satisfy the Geigel algorithm, can not use the Geigel algorithm simply and carry out double talk detection.

Summary of the invention

The purpose of this invention is to provide and a kind ofly realize simply in the SCDMA mobile phone, it is less to take resource, can carry out both-end effectively and detect, and the method for identification far-end echo is to eliminate the acoustic echo that produces in the SCDMA wireless communication system.

The present invention mainly utilizes the echo signal low frequency energy characteristic lower than efficient voice, far-end echo and near-end efficient voice that mobile phone produces is distinguished, to reach the purpose that suppresses acoustic echo.

The present invention includes following steps:

(1) obtains a frame sampling signal at near-end, energy according to far-end 5～15 frames before this frame judges whether far-end exists voice, if have, then need this frame of near-end is carried out voice detection of echoes (VED), if do not have, then do not need this frame of near-end is carried out the voice detection of echoes, forward step (4) to;

(2) carry out the frame of voice detection of echoes for needs, calculate the low frequency energy of this frame signal, if its low frequency energy is less than a thresholding T2 then think and only have echo; Otherwise think to exist efficient voice or echo and efficient voice to exist simultaneously, forward step (4) to;

(3) for judging the situation only there is echo, the result who judges being carried out smoothly, still be echo for the judged result after level and smooth, takes comfort noise signal replacement echo to import speech coder, otherwise does not change the signal of present frame;

(4) comfort noise to voice or replacement echo carries out coding transmission.

The SCDMA terminal V32 mobile phone that has adopted the inventive method is tested, and people's ear can feel more significantly that acoustic echo weakens to some extent, and the effect of sound is improved.

Description of drawings

Fig. 1 is identification and eliminate the processing flow chart of the method for the echo that far-end speech produces in the SCDMA mobile phone of the present invention;

Fig. 2 a～2f shows respectively for the far-end speech source signal, the near end echo signal, near-end efficient voice signal carries out the resultant curve map of frequency-response analysis, wherein Fig. 2 a, Fig. 2 c, Fig. 2 e are 4096 signal waveform curve, and Fig. 2 b, Fig. 2 d, Fig. 2 f are 4096 amplitude-versus-frequency curve;

Fig. 3 a～Fig. 3 c is respectively under the situation that near end signal in microphone place input is quiet, echo and efficient voice, the curve map of the low frequency energy that calculates:

Fig. 4 is a detail flowchart of realizing echo cancellation process in SCDMA V32 mobile phone.

Embodiment

Fig. 1 shows identification and eliminate the base conditioning flow process of the method for the echo that far-end speech produces in the SCDMA mobile phone of the present invention.

Shown in Fig. 2 a～Fig. 2 f, for the far-end speech source signal of on the SCDMA mobile phone, gathering, the near end echo signal, near-end efficient voice signal is done the analysis of frequency characteristic, Fig. 2 a wherein, Fig. 2 c, Fig. 2 e is 4096 a signal waveform curve, and Fig. 2 b, Fig. 2 d, Fig. 2 f is 4096 a amplitude-versus-frequency curve, the energy that can see the efficient voice signal mainly concentrates on low frequency, and the energy of the echo that the process echo channel produces mainly concentrates on high frequency, the full range band energy of echo not necessarily satisfies the Geigle algorithm, but can utilize the size of low frequency energy that echo and efficient voice are distinguished, thereby reach the purpose of eliminating echo.

Fig. 3 a～Fig. 3 c is respectively under the situation that near end signal in SCDMA mobile microphone place input is quiet, echo and efficient voice, the curve of the low frequency energy that use DSP algorithm computation obtains in mobile phone, wherein horizontal ordinate is represented corresponding frame number, the low frequency energy of efficient voice will be higher than quiet and echo as can be seen, is low frequency energy effective as the condition of distinguishing echo and efficient voice.

As shown in Figure 4, use the method for the present invention test that experimentizes in SCDMA terminal V32 mobile phone, concrete steps are as follows:

1, obtains a frame signal from near-end (mobile phone microphone) sampling, whether have more than or equal to a certain thresholding T1 (this numerical value is empirical value) according to the energy of far-end (in the receiver) 5～15 frames before it (thinking that the echo channel time-delay is between 50ms～150ms) and to judge whether far-end exists voice, if there are voice in the judgement far-end then need this frame of near-end is carried out the voice detection of echoes, if judge that there are not voice in far-end, then there is not echo certainly in near-end, do not need this frame of near-end is carried out the voice detection of echoes,forward step 4 to;

2, carry out the frame of voice detection of echoes for needs, calculate the low frequency energy E1 of this frame signal, if E1 is less than thresholding T2 (this numerical value is empirical value), think that then only there is echo in near-end, the voice detection of echoes result of this frame is changed to " 0 (echo) ", otherwise think that near-end exists efficient voice or echo and efficient voice to exist simultaneously, the voice detection of echoes result of this frame is changed to " 1 (voice) ", changestep 4 over to;

3, for judging the situation that only has echo, the result who judges is carried out smoothly, if the voice detection of echoes result of a plurality of frames before the present frame has outnumbered T3 (this numerical value empirical value) for " 1 " continuously, think that then efficient voice begins, when even the energy of this frame no longer satisfies greater than T2, if satisfy following two conditions simultaneously:

A) time of occurrence is in after the efficient voice in 10 frames;

B) low frequency energy is greater than another thresholding T4 (this numerical value is empirical value);

Then think to exist efficient voice or echo and efficient voice to exist simultaneously, otherwise think and only have echo;

For the judged result after level and smooth still be echo, takes comfort noise signal replacement echo, otherwise does not change the signal of present frame;

4, the comfort noise to voice or replacement echo carries out coding transmission.

The described method of the application of the invention can be carried out double talk detection effectively, identifies far-end echo, thereby eliminates the acoustic echo that produces in the SCDMA wireless communication system.The SCDMA terminal V32 mobile phone that has adopted the inventive method is tested, and people's ear can feel more significantly that acoustic echo weakens to some extent, and the effect of sound is improved.

Claims (4)

Translated fromChinese

1.一种在SCDMA手机中识别并消除远端语音产生的回声的方法，其包括以下步骤：1. a method for identifying and eliminating the echo produced by far-end voice in SCDMA mobile phone, it may further comprise the steps:(1)在近端获取一帧采样信号，根据该帧之前远端多个帧的能量判断远端是否存在语音，如果存在，则需要对近端该帧进行语音回声检测，如果不存在，则不需要对近端该帧进行语音回声检测，转到步骤(4)；(1) Obtain a frame of sampling signal at the near end, and judge whether there is voice at the far end according to the energy of multiple frames at the far end before the frame. If it exists, it is necessary to perform voice echo detection on the near end frame. Do not need to carry out speech echo detection to this near-end frame, go to step (4);(2)对于需要进行语音回声检测的帧，计算该帧信号的低频能量，如果其低频能量小于一门限T2则认为只存在回声；否则认为存在有效语音或者回声和有效语音同时存在，转到步骤(4)；(2) For the frame that needs to carry out voice echo detection, calculate the low-frequency energy of the frame signal, if its low-frequency energy is less than a threshold T2 then think that there is only echo; otherwise think that there is effective voice or echo and effective voice exist simultaneously, go to step (4);(3)对于判断只存在回声的情况，对判断的结果进行平滑，对于平滑之后的判断结果仍然为回声的，采取舒适噪声信号代替回声，否则不改变当前帧的信号；(3) For judging that there is only echo, the result of judging is smoothed, and if the judging result after smoothing is still echo, the comfort noise signal is used instead of echo, otherwise the signal of the current frame is not changed;(4)对语音或者代替回声的舒适噪声进行编码传输。(4) Coding and transmitting speech or comfort noise instead of echo.2.如权利要求1所述的一种方法，其特征在于，步骤(1)中，判断远端是否存在语音的依据是：如果近端该帧之前的5～15帧远端信号帧能量有大于等于某一门限T1的，则判断远端存在语音。2. A kind of method as claimed in claim 1, it is characterized in that, in step (1), the basis for judging whether there is voice at the far end is: if the 5～15 frames of far-end signal frame energy before this frame of near-end have If it is greater than or equal to a certain threshold T1, it is judged that there is voice at the far end.3.如权利要求1或2所述的一种方法，其特征在于，步骤(2)中，当所述近端信号帧的低频能量小于所述门限T2时，将该信号的语音回声检测结果置为0，否则，将其语音回声检测结果置为1。3. A kind of method as claimed in claim 1 or 2, it is characterized in that, in step (2), when the low-frequency energy of described near-end signal frame is less than described threshold T2, the voice echo detection result of this signal Set to 0, otherwise, set the voice echo detection result to 1.4.如权利要求3所述的一种方法，其特征在于，在步骤(3)中所述平滑过程包括以下步骤：如果当前帧之前的多个帧的语音回声检测结果连续为“1”的数目超过了一定值T3，则认为有效语音已经开始，即使该帧的能量不再满足大于步骤(2)中所述门限T2时，如果其同时满足以下两个条件：4. a kind of method as claimed in claim 3 is characterized in that, in step (3), described smooth process comprises the following steps: if the voice echo detection result of a plurality of frames before current frame is " 1 " continuously Number surpasses certain value T3, then thinks that effective speech has begun, even if the energy of this frame no longer satisfies when greater than the threshold T2 described in step (2), if it satisfies the following two conditions simultaneously:a)出现时间处于有效语音之后10帧以内；a) The appearance time is within 10 frames after the effective speech;b)其低频能量大于另一门限T4；b) its low-frequency energy is greater than another threshold T4;则认为存在有效语音或者回声和有效语音同时存在，否则认为只存在回声。Then it is considered that there is effective speech or echo and effective speech exist at the same time, otherwise it is considered that there is only echo.

Images