Summary of the invention
The deficiency existing for solving prior art, the invention discloses the interactive voice control method based on Intelligent Mobile Robot, it is means that this control method be take the application and development of speech recognition and phonetic synthesis, be beneficial to realization reliably efficiently robot voice control and interactive sessions.
For achieving the above object, concrete scheme of the present invention is as follows:
Interactive voice control method based on Intelligent Mobile Robot, comprises the following steps:
Step 1: utilize the speech transducer being arranged on crusing robot to gather voice signal, the audio collection card that speech transducer carries with crusing robot is connected, and audio collection card is sent into the control system of crusing robot using the voice signal obtaining as speech recognition input source;
Step 2: the speech recognition system of crusing robot is carried out end-point detection pre-service, feature extraction to the voice signal collecting, then by identification engine, judge whether voice content is legal phonetic order, legal phonetic order comprises puts question to instruction and steering order two classes; If legal phonetic order, further judges the type of phonetic order, if the phonetic order after speech recognition is legal and be identified as while puing question to instruction, turn to step 3, otherwise, turn to step 4;
Step 3: search out the reply instruction corresponding with described enquirement instruction from reply command file, then crusing robot is replied with the form of voice broadcast;
Step 4: if the phonetic order after speech recognition is legal and while being identified as steering order, complete assigned operation to crusing robot generation corresponding control signal.
In described step 1, the control system of crusing robot detects the voice signal that audio collection card obtains, if voice signal intensity and duration reach setting threshold, shows to have efficient voice input, starts to carry out speech recognition.
In described step 2, the content of phonetic entry is judged in speech recognition, and recognition result adopts the form of instruction text to export.
In described step 2, put question to instruction and steering order to be stored in one by one in same interactive instruction file, put question to instruction in front, steering order rear, put question to instruction to be provided with the corresponding instruction of replying for every, allly reply instruction and be stored in one by one and reply in command file, in initialization procedure, identification engine can identify interactive instruction file and reply the position of command file, and can read interactive instruction file and the content of replying each instruction in command file.
In described step 2, voice content is analyzed to identification, if judge, voice content puts question to instruction or control command as in interactive instruction file a certain, thinks and successfully obtains recognition result.
Before enquirement instruction in described step 2 and the identification of steering order, write syntax rule file, the word that wherein need to identify in syntax rule document definition and phrase, SDK speech recognition engine loads the voice that this syntax rule is identified user.
In described step 3, crusing robot is replied with the form of voice broadcast, and the robot form of replying is the voice of prerecording, or first by phonetic synthesis, is realized the conversion of Text To Speech, realized voice output again.
The described syntax rule file of writing, the syntax rule of SAPI adopts XML form, when speech recognition, to the judgement of every instruction, is the mode that adopts whole comparison or segmentation comparison.
Beneficial effect of the present invention:
1, the present invention utilizes the phonetic order that the correct identifying operation personnel of speech recognition technology send, again according to the type of recognition result judgement phonetic order, by arranging, put question to instruction and steering order to realize the Interactive control to robot, assurance crusing robot, after recognizing efficient voice dialogue and voice operating instruction, can be made in time and reply or complete assigned operation according to recognition result.
2, the present invention has designed complete interactive voice development process according to application demand; Based on the speech application interface SAPI of Microsoft, completed a set of man machine language's intersection control routine on this basis.On-the-spot test result shows, the property indices of system meets design requirement.
3, the present invention is conceived to the functional requirement that Intelligent Mobile Robot interactive voice is controlled, the application and development of speech recognition and phonetic synthesis of take is means, realized reliably robot voice control efficiently and interactive sessions, in power industry, had broad application prospects.
Embodiment:
Below in conjunction with accompanying drawing, the present invention is described in detail:
In conjunction with shown in Fig. 1 to Fig. 3, the interactive voice control method based on Intelligent Mobile Robot, comprises the steps:
A, voice collecting
The audio collection card that speech transducer is carried with crusing robot is connected, and utilize the speech transducer that crusing robot carries to gather voice signal, and the voice signal that audio collection card is obtained is as speech recognition input source; In voice collecting process, detect the voice signal that audio collection card obtains, if reaching certain threshold value, voice signal intensity and duration show to have efficient voice input, start to carry out speech recognition;
B, speech recognition, obtain voice content by identification engine, then by search instruction file, judges whether it is valid instruction; Valid instruction is further judged to its type.
By the voice signal collecting is carried out to analyzing and processing, feature extraction matching identification goes out voice content, then by identification engine, judge whether voice content is legal phonetic order, if legal phonetic order, further judge the type of phonetic order, and then control crusing robot and make response, legal phonetic order comprises puts question to instruction and steering order two classes, put question to instruction to comprise the instructions such as " what is your name ", " what you can do ", steering order comprises the instructions such as " starting charging ", " stopping charging ", " opening ultrasonic "; The content of phonetic entry is judged in speech recognition, and recognition result adopts the form of instruction text to export;
Above-mentioned enquirement instruction and steering order are stored in same interactive instruction file one by one, put question to instruction in front, steering order rear, put question to instruction to be provided with the corresponding instruction of replying for every, allly reply instruction and be stored in one by one and reply in command file, in initialization procedure, identification engine can identify interactive instruction file and reply the position of command file, and can read interactive instruction file and the content of replying each instruction in command file; Voice content is analyzed to when identification, if judges, voice content, as a certain enquirement instruction in interactive instruction file or control command, thinks and successfully obtains recognition result;
C, dialogue are replied
If the phonetic order after speech recognition is legal and be identified as while puing question to instruction, from reply command file, search out the reply instruction corresponding with described enquirement instruction, then crusing robot is replied with the form of voice broadcast; The robot form of replying is the voice of prerecording, or first by phonetic synthesis, is realized the conversion of Text To Speech, realized voice output again;
D, executable operations
If the phonetic order after speech recognition is legal and while being identified as steering order, complete assigned operation to crusing robot generation corresponding control signal.
In interactive voice control method, speech recognition and be to realize the key that inspection machine person speech interaction is controlled for talking with the phonetic synthesis of replying.Coupling system demand and voice technology state of development, the mode that the present invention is based on existing voice kit has realized whole interactive voice control system.
Speech recognition technology is exactly to allow crusing robot voice signal be changed into the technology of corresponding text or order by identification and understanding process, and technology contents mainly comprises Feature Extraction Technology, pattern match criterion and three aspects of model training technology.The SAPI language engine of Microsoft and the InterReco speech recognition system that domestic University of Science and Technology news fly to provide are provided conventional voice developing instrument at present.The present invention adopts the SAPI of Microsoft to carry out speech recognition exploitation.SAPI is a set of application programming interface about speech processes that Microsoft Speech SDK provides, and has comprised the basic function of realizing word-speech recognition and voice operation program, has simplified the difficulty of voice programming, has reduced the workload of voice programming.
As shown in Figure 3, speech engine is communicated by letter by device drives interface DDI layer SAPI Runtime Library, and application program is mutual by application programming interfaces api layer and SAPI, by using these API to carry out the exploitation of speech recognition and phonetic synthesis aspect.
The detailed exploitation of speech recognition, SAPI builds speech recognition system can adopt two kinds of recognition modes: voice command control model and voice dictation pattern.The speech recognition system that adopts voice command control model to build, be applicable to the speech recognition of little vocabulary, isolated word, unspecified person, but need to create syntax rule, identification range can only be confined to designed phrase or words in syntax rule, bad adaptability but to limit content recognition rate high.The speech recognition system of voice dictation mode construction is suitable for the identification of large vocabulary, continuous speech, simultaneously without building complicated syntax rule, and the strong adaptability of speech recognition, but discrimination is relatively lower.
The mutual instruction of controlling of robot voice is limited and relatively-stationary, therefore adopts the order that discrimination is higher to control recognition mode.According to the development process of SAPI, the basic configuration and the initialization of identification engine that complete speech recognition can be carried out speech recognition.
The key issue of voice command recognition mode is phonetic rules, is that the standardization of the command library to identifying is described.The syntax rule of SAPI adopts XML form.Before carrying out identification work, need to write a syntax rule file, wherein defined the word and the phrase that need identification, SDK speech recognition engine loads the voice that this syntax rule is identified user.
In XML, each entity or element are comprised of beginning label < attribute-name > and end mark </ attribute-name >, and the statement of folder is exactly the grammatical content that this entity or element comprise therebetween.The content of the syntax can be common language, or the daughter element of grammatical element.Formal definition for legal grammatical content in XML standard is the form that adopts many aggregate expressions.Utilize these definition, grammer that just can explication file and the rule in the syntax.In the syntax, what in <P> and </P>, insert is one group of word to be identified or phrase, and what in <L></LGreatT.Gre aT.GT, insert is to take word or the phrase that <P></PGreatT.Gre aT.GT is possible match user voice that can options.The part grammar rule that robot voice is controlled employing alternately as follows:
<GRAMMAR?LANGID＝“804”>
-<DEFINE>
<ID?NAME＝“CMD”VAL＝“10”/>
</DEFINE>
-<RULE?NAME＝“COMMAND”ID＝“CMD”TOPLEVEL＝“ACTIVE”>
-<L>
What space </P> <P>+ you are
<P>+ you from where </P>
<P>+ you all can What for </P>
<P>+ starts high resistance regional patrol task </P>
<P>+ stops task </P>
Wherein, GRAMMAR LANGID=" 804 ", having indicated identifying object is Chinese character, in identifying, will call Chinese acoustic model; Is all recognition command between <P> and </P>.XML grammar file can need to synchronously upgrade and load XML file according to recognition command edit when recognition command changes, and speech recognition engine could be identified new instruction.Therefore edit mode is not easy to more new syntax, has realized on stream the sequencing renewal of grammar file according to XML rule, adopts MSXML2::IXMLDOMDocumentPtr and MSXML2::IXMLDOMElementPtr to edit XML file.
The detailed exploitation of phonetic synthesis, the massage voice reading that phonetic synthesis can be converted into any Word message standard smoothness in real time out.By computer speech, synthesize and can at any time any text-converted be become to have the voice of high naturalness, realize and allow machine " as people, lift up one's voice ".This is the demand of robot voice alternate acknowledge just.Adopt the SAPI of Microsoft to realize step and the speech recognition of phonetic synthesis roughly similar, comprise equally volume, word speed and engine initialization.One of them key issue is to select pronunciation storehouse, and Ji Rang robot with which type of intonation speaks.The SAPI of Microsoft has comprised Chinese and English pronunciation storehouse, but Chinese pronunciations type is few and effect is poor in SAPI5.1 and following version; Can strengthen phonetic synthesis effect by third-party sound bank is installed, as Neospeech sound bank.
According to above-mentioned flow process, complete the exploitation of robot voice intersection control routine, through on-the-spot test correct recognition rata, surpassed 90%.Phonetic recognization rate to be further improved, voice pre-service and user's training can be added.Especially, in syntax rule design, carried out experimental study herein.
Adopt whole comparison grammatical form, be to adopt whole comparison to the judgement of every instruction when speech recognition, final recognition result be with above-mentioned list in the most approaching option.Experimental results show that under the syntax rule of this kind of form, correct recognition rata is higher, but false alarm rate is higher, some interference tones may be identified as to certain instruction easily, particularly some short instructions, as " recording ", " opening " etc.Analyzing the reason that produces this problem, should be to contain and the similar phonetic element of these short instructions during some disturbs.In order to guarantee discrimination, the comparison that comparison under the syntax rule of this whole instruction type is not strict, because recognition system will be tolerated input deviation to a certain degree, as being read as to " three-east ", " Shandong " also should be able to identify, user's pronunciation may vary after all, if can be as newsreader issuing command problem just simple.
Therefore, under above-mentioned syntax rule, some short instructions are often known by mistake.For improving this problem, designed more strict syntax rule, as follows:
-<GRAMMAR?LANGID＝“804”>
-<RULE NAME=" what is your name " TOPLEVEL=" ACTIVE " >
You are </P> <P>+
What </P> of <P>+
<P>+ name </P>
</RULE>
<RULE NAME=" you from where " TOPLEVEL=" ACTIVE " >
You carry out </P> <P>+
<P>+ is </P> from where
</RULE>
<RULE NAME=TOPLEVEL=" ACTIVE " > that " furthers "
<P>+ draws </P>
The nearly </P> of <P>+
</RULE>
The syntax rule of this segmentation form has realized the identification of stricter coupling, as identified " furthering ", need to will " draw " and " closely " all compared out simultaneously.Above two kinds of multi-form identification grammers have his own strong points, and therefore how reasonable design syntax rule is to improve an approach of speech recognition effect.
For strengthening the voice interactive function of Intelligent Mobile Robot, according to application demand, designed complete interactive voice development process; Based on the speech application interface SAPI of Microsoft, completed a set of man machine language's intersection control routine on this basis.On-the-spot test result shows, the property indices of system meets design requirement.Certainly the development level that is limited to speech recognition technology, the speech recognition capabilities of robot is compared with people and there is a big difference.The antijamming capability of this voice interactive system need to improve, particularly in the reliable recognition ability having under the background of making an uproar.