JP2004061609A - Method and device for speech recognition - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To perform tutor-attended speaker adaptive learning of a speech recognition device underground. <P>SOLUTION: In a speech recognizing method of comparing an input speech pattern with registered speech patterns of a plurality of words and recognizes speech-inputted words according to the most similar registered speech pattern, the recognition result of the speech recognition device and the input speech pattern to an external registered speech pattern improving device, which generates improvement data on the registered speech pattern registered in a speech recognizing engine of the speech recognition device according to the sent information and transmits the data. Then the speech recognition device updates the registered speech pattern data registered in the speech recognizing engine with the sent improvement data. <P>COPYRIGHT: (C)2004,JPO

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a speech recognition method and a speech recognition device, and more particularly to a speech recognition method and a speech recognition device using speaker adaptation technology.
[0002]
[Prior art]
In order to improve the voice recognition performance for a specific speaker, the voice recognition device uses a speaker adaptation technique for learning the voice of the specific speaker. Such speaker adaptation techniques can be broadly classified into two types: supervised and unsupervised. Here, the teacher refers to a phoneme notation sequence representing the utterance content of the input utterance. The supervised adaptation is an adaptation method in a case where a phoneme notation sequence for an input utterance is known. In the adaptation, it is necessary to instruct an unknown speaker in advance on a vocabulary vocabulary.
[0003]
On the other hand, unsupervised adaptation is an adaptation method in the case where the phoneme notation sequence for an input utterance is unknown, and does not limit the utterance content of the input utterance to the unknown speaker, that is, the utterance content to the unknown speaker. This method is easy for the user to use because it is not necessary to give an instruction, and the adaptation can be performed using the input speech that is actually using the speech recognition without being aware of the unknown speaker.
In general, unsupervised adaptation has lower recognition performance after adaptation than supervised adaptation, so supervised adaptation is often used at present.
[0004]
Hereinafter, a conventional speech recognition apparatus using a supervised adaptation technique will be described with reference to FIG.
The utterance S of the speaker input to the voice recognition device 1 is input to the input pattern creation unit 2 and undergoes processes such as AD conversion and voice analysis to obtain a feature vector for each unit called a frame having a certain time length. Converted to a series. This time series of feature vectors is referred to as an input pattern here. The feature vector is obtained by extracting the feature amount of the speech spectrum at that time, and is usually 10 to 100 dimensions.
[0005]
On the other hand, the standard pattern storage unit 3 stores a hidden Markov model (HMM: Hidden Markov Model). HMM is one of the information source models of speech, and its parameters can be learned using speech of a speaker. The HMM is usually prepared for each predetermined recognition unit. Here, a phoneme is taken as an example of the recognition unit. Therefore, the phoneme HMM is stored in the standard pattern storage unit 3 as a standard pattern as shown in FIG. As the phoneme HMM, for example, an unspecified speaker HMM that has been learned in advance using the utterances of many speakers is used.
It is assumed that 1000 words are to be recognized, that is, a case where a correct answer of one word is obtained from recognition candidates of 1000 words. At the time of word recognition, the vocabulary pattern creation unit 5 creates an HMM (word MFF) of a recognition candidate word by connecting the HMMs of the phonemes constituting the word. In the case of 1000 word recognition, a word HMM for 1000 words is created. That is, as shown in FIG. 4, the vocabulary pattern creation unit 5 includes a word storage unit 5a that stores words of 1000 words and a word HMM creation unit 5b that creates a word HMM for each word by connecting phoneme HMMs. I have.
[0006]
The recognizing unit 4 recognizes the input pattern using the word HMM of 1000 words created by the vocabulary pattern creating unit 5. The phoneme HMM is a model of a speech information source, and a statistical idea is introduced in the description of a standard pattern in order to cope with various fluctuations of a speech pattern. A phoneme HMM usually consists of 1 to 10 states and state transitions between them. Usually, a start state and an end state are defined, and a symbol is output from each state and a state transition is performed for each unit time. The voice pattern of a phoneme is represented as a time series of symbols output during a state transition from a start state to an end state. A symbol appearance probability is defined for each state, and a transition probability is defined for each transition between the states. The transition probability parameter is a parameter for expressing temporal fluctuation of the voice pattern. The appearance probability parameter expresses the fluctuation of the timbre of the voice pattern. By setting the probability of the initial state to a certain value and multiplying the appearance pattern and the transition probability for each state transition by the input pattern, it is possible to calculate the occurrence probability when the input voice is assumed to be generated from a certain HMM. it can.
[0007]
In the speech recognition by the HMM, a word HMM is prepared for 1000 recognition candidate words, and when a speech is input, the occurrence probability is calculated for the word HMM of each recognition candidate word, and the maximum word HMM is generated. Is determined as the source, and the recognition result is set to the recognition candidate word corresponding to the word HMM. The recognition result word is sent to the recognition result output unit 6. The recognition result output unit 6 performs processing such as outputting the recognition result on a screen or transmitting a control command corresponding to the recognition result to another device. In the above description, the HMM is stored in the standard pattern storage unit 3 according to the phoneme, but the word HMM can be stored according to the word as shown in FIG. In such a case, the vocabulary pattern creating section 5 becomes unnecessary.
[0008]
Next, a supervised speaker adaptation technique for the speech recognition device 1 will be described. In the supervised speaker adaptation, a word to be uttered is instructed to the user in advance, and the parameters of the phoneme HMM are updated using the word notation and the input speech. This is called supervised adaptation in the sense that the correct word for the utterance is known in advance.
[0009]
First, as in the case of recognition, the input pattern creating unit 7 creates an input pattern from the input voice. In the case of supervised adaptation, since the correct word is known in advance, the adaptation dictionary creating unit 8 creates an adaptation dictionary from the input correct word notation (input speech notation). Next, the vocabulary pattern creation unit 9a of the supervised adaptation unit 9 inputs the vocabulary pattern using the phoneme sequence of the adaptation dictionary and the adaptation initial HMM for each phoneme stored in the adaptation initial standard pattern storage unit 9b in advance. Create a word HMM corresponding to the pattern. Then, the adaptation unit 9c performs likelihood calculation between the input pattern and the word HMM for adaptation, performs an adaptation process on one or a plurality of input patterns, and calculates an average vector after the adaptation. An HMM is obtained after the adaptation, and the adapted HMM is input to the standard pattern storage unit 3 and stored instead of the standard HMM.
The above-described supervised adaptation technique is well known, and is described in detail in, for example, Japanese Patent Application Laid-Open No. 7-230295.
[0010]
[Problems to be solved by the invention]
However, the supervised adaptation method has a drawback in that a word instructed by the apparatus must be uttered as training separately from the generation of speech in the original speech recognition, and the burden is large. In other words, it is necessary to work for a purpose different from the original purpose (improving the recognition performance of voice recognition) different from the original purpose equipped with the voice recognition device for improving the human interface (to facilitate the operation of the device), which is complicated. In addition, there is a problem of imposing a burden on the user.
In view of the above, an object of the present invention is to provide a speech recognition method and a speech recognition apparatus capable of automatically providing the same performance as that of the supervised adaptation method without making the user aware.
It is another object of the present invention to provide a speech recognition method and a speech recognition apparatus which can have the same performance as the supervised adaptation method with a simple configuration.
[0011]
[Means for Solving the Problems]
The first aspect of the present invention is a voice recognition method for comparing a registered voice pattern of a plurality of words with an input voice pattern and recognizing a word that has been voice-input based on the most similar registered voice pattern. Sending the recognition result of the speech recognition device and the input speech pattern to an external registered speech pattern improving device; and (2) in the registered speech pattern improving device, the speech recognition engine of the speech recognition device based on the transmitted information. Generating and transmitting registered voice pattern improvement data registered in (3), in the voice recognition device, updating registered voice pattern data registered in the voice recognition engine with the transmitted improved data. have.
[0012]
A second aspect of the present invention is a voice recognition device for recognizing a word input by voice, (1) a voice pattern registration unit for registering a voice pattern corresponding to a word, and (2) a voice pattern registration unit for a plurality of words. A voice recognition engine that compares the input voice pattern and recognizes a word that has been voice-input based on the most similar registered voice pattern; (3) a storage unit that stores the recognition result and the input voice pattern as a set; 4) a data transmission unit for periodically transmitting the recognition result and the input voice pattern stored in the storage unit to an external registered voice pattern improvement device; and, (5) a voice transmission unit based on the information transmitted from the voice recognition device. A data receiving unit for receiving the registered voice pattern improvement data from the registered voice pattern improvement device for generating data for improving the registered voice pattern of the voice recognition engine of the recognition device; Has an improved unit for improving the registered voice pattern data registered in the voice recognition engine by improving data sent it said.
ADVANTAGE OF THE INVENTION According to this invention, since it becomes possible to perform the supervised speaker adaptive learning of a speech recognition apparatus in an underground, a user does not need to supply a teacher signal to a speech recognition apparatus, and the learning itself is supervised. Since the learning is performed, there is no fear that the learning proceeds in an erroneous direction, and highly accurate learning can be performed, and the recognition rate can be improved.
[0013]
BEST MODE FOR CARRYING OUT THE INVENTION
(A) Configuration of Speech Recognition System FIG. 1 is a configuration diagram of a speech recognition system including a speech recognition device according to the present invention. A speech recognition device 11 attached to a device controlled by a speech input improves recognition performance. As described above, the improvement center 12 for updating the parameters for specifying the standard pattern stored in the standard pattern storage unit of the voice recognition device 11 is provided. Communication between the voice recognition device 11 and the improvement center 12 can be freely performed by an arbitrary communication method, for example, a wireless communication using a LAN or a mobile phone.
[0014]
In the speech recognition apparatus 11, the speech input microphone 21 detects and outputs speech input by the speaker, the AD converter 22 converts the speech signal into digital, and the speech recognition engine 23 operates as shown in FIG. It has the same configuration as the recognition device 1) and recognizes input speech. The voice data file / recognition result storage unit 24 stores the recognition result of the input voice and stores the input voice pattern data in a WAV format. The data transmission unit 25 transmits one or more recognition results stored in the storage unit 24 and the input voice pattern data to the improvement center in pairs. The data receiving unit 26 receives the improved data (parameters for specifying the standard pattern) from the improvement center 12, and the installer 27 updates the standard pattern data of the speech recognition engine 23 with the received improved data.
[0015]
In the improvement center 12, the data receiving unit 31 receives a plurality of sets of recognition result / input pattern data sent from the speech recognition device 11, and the teacher signal generation unit 32 sends the data from the speech recognition area engine 23 of the speech recognition device 11. It has a much higher performance speech recognition engine, performs speech recognition based on each set of input pattern data sent, and outputs the recognition result as input speech notation (teacher signal) of the input pattern. The improvement unit 33 performs the same adaptation processing as the supervised adaptation unit 9 in FIG. 2 using the input phonetic notation) and the input pattern data sent from the speech recognition device 11, and thereby adapts the HMM of the phoneme. Ask for. Thereafter, a different phoneme is obtained by comparing the recognition result sent from the voice recognition device with the teacher signal (input phonetic notation), and an HMM adapted to the different phoneme is output. The improved data storage unit 34 stores the improved data (the HMM adapted for each phoneme) output from the improving unit 33, and the data transmitting unit 35 transmits the improved data to the speech recognition device.
[0016]
(B) Overall operation of the speech recognition system {circle around (1)} When the user uses the speech recognition apparatus 11, the speech content (speech pattern) at that time and the recognition result recognized by the speech recognition engine 23 are respectively recognized as a speech data file / recognition. The result is stored in the result storage unit 24. The utterance content is held in, for example, a WAV file format, and an ID code is added. The recognition result is recorded in a text format, for example, corresponding to the ID code.
{Circle over (2)} When a certain period (which may be set by the user or set by the manufacturer) has elapsed, the data transmission unit 25 transmits the audio data file recorded in the audio data file / recognition result storage unit 24 to the audio data file. The recognition result is transmitted to the improvement center 12 as a pair.
[0017]
{Circle around (3)} In the improvement center 12, the data receiving unit 31 stores the information received from the voice recognition device 11 in a built-in memory and sequentially pairs the utterance content and the recognition result to the teacher signal generation unit 32 and the improvement unit 33. input. The teacher signal generation unit 32 performs voice recognition based on the input voice pattern data, outputs the recognition result as an input voice notation of the input voice pattern, and the improvement unit 33 converts the input voice notation and the input voice pattern data. The HMM of the phoneme that has been adapted by performing the adaptation process is obtained. Thereafter, the improvement unit 33 compares the recognition result sent from the speech recognition device 11 with the teacher signal (input speech notation) to obtain a different phoneme, and outputs an HMM adapted to the different phoneme. The improved data storage unit 34 stores the improved data (the HMM to which each phoneme has been adapted) output from the improving unit 33, and the data transmitting unit 35 transmits the improved data to the speech recognition device 11.
[0018]
{Circle around (4)} Upon receiving the improved data, the installer 27 installs the improved data (the HMM adapted for each phoneme) in the standard pattern storage unit in the speech recognition engine 23 and stores the data instead of the standard HMM. .
In the above description, the standard HMM is a phoneme HMM, but may be a word HMM.
[0019]
【The invention's effect】
As described above, according to the present invention, it is possible to perform the supervised speaker adaptive learning of the speech recognition device in the underground, so that the user (speaker) does not need to supply a teacher signal to the speech recognition device.
Further, according to the present invention, since the learning itself is supervised learning, there is no possibility that the learning proceeds in the wrong direction, and highly accurate learning can be performed. The recognition performance that can be enjoyed by the user can be improved.
Further, according to the present invention, the user's voice is sent to a center equipped with a high-performance voice recognition engine to generate improved data, and the improved data is installed in the voice recognition engine of the user's voice recognition device. A speech recognition device with an improved recognition rate for unspecified speakers can be provided. In addition, a high-performance speech recognition engine is not required for the user's voice recognition, so that a low-cost and compact device configuration can be realized.
[Brief description of the drawings]
FIG. 1 is a configuration diagram of a speech recognition system including a speech recognition device of the present invention.
FIG. 2 is a conventional speech recognition apparatus using a supervised adaptation technique.
FIG. 3 is an example of storing a phoneme HMM as a standard pattern.
FIG. 4 is a configuration diagram of a vocabulary pattern creation unit.
FIG. 5 is an example of storing a word HMM as a standard pattern.
[Explanation of symbols]
11 Speech Recognition Device 12 Improvement Center 21 Microphone 22 for Speech Input 22 A / D Converter 23 Speech Recognition Engine 24 Speech Data File / Recognition Result Storage Unit 27 Installer 31 Data Reception Unit 32 Teacher Signal Generation Unit 33 Improvement Unit 34 Improved Data Storage Unit

Claims (3)

A voice recognition method for comparing a registered voice pattern of a plurality of words with an input voice pattern and recognizing a word that has been voice-input based on the most similar registered voice pattern,
The recognition result of the voice recognition device and the input voice pattern are sent to an external registered voice pattern improvement device,
In the registered voice pattern improvement device, based on the transmitted information, generates and transmits registered voice pattern improvement data registered in the voice recognition engine of the voice recognition device,
In the voice recognition device, updating the registered voice pattern data registered in the voice recognition engine with the transmitted improved data,
A speech recognition method characterized in that: The registered voice pattern improved data generation step includes:
Speech recognition is performed based on the input speech pattern, the recognition result is regarded as an input speech notation in supervised speaker adaptive learning, and supervised speaker adaptive learning is performed using the input speech notation to generate registered speech pattern improvement data. ,
2. The speech recognition method according to claim 1, wherein: In a voice recognition device for recognizing a word input by voice,
A voice pattern registration unit for registering a voice pattern corresponding to a word,
A voice recognition engine that compares a registered voice pattern of a plurality of words with an input voice pattern and recognizes a word that has been voice-input based on the most similar registered voice pattern;
A storage unit for storing the recognition result and the input voice pattern as a set,
A data transmission unit that periodically sends the recognition result and the input voice pattern stored in the storage unit to an external registered voice pattern improvement device,
Data for receiving the registered voice pattern improvement data from the registered voice pattern improvement device that generates data for improving the registered voice pattern of the voice recognition engine of the voice recognition device based on the information sent from the voice recognition device. Receiver,
An improving unit for improving registered voice pattern data registered in the voice recognition engine by the transmitted improved data,
A speech recognition device comprising:

Images