CN116129879A - Voice recognition method, device, equipment and storage medium - Google Patents

Abstract

The invention provides a voice recognition method, a device, equipment and a storage medium, wherein the method comprises the following steps: receiving voice information to be recognized; performing voice recognition on the voice information to be recognized by using a voice recognition model to obtain posterior data corresponding to the voice information to be recognized of each frame; the voice recognition model is obtained based on short word vocabulary training, and the short word vocabulary is obtained by carrying out semantic cutting on each long word in the long word vocabulary; decoding the posterior data by utilizing the word list of the short words to sequentially obtain a plurality of recognition short words corresponding to the voice information to be recognized; and determining a voice recognition result of the voice information to be recognized according to the long word list and the plurality of recognition short words. The invention can reject the word of the mood or other irrelevant words in the voice information to be recognized, thereby guaranteeing the accuracy of the voice recognition result and the success rate of the voice recognition.

Description

Speech recognition method, device, equipment and storage medium

technical field

The present invention relates to the technical field of artificial intelligence, in particular to a voice recognition method, device, equipment and storage medium.

Background technique

With the advancement of science and technology, intelligent voice technology is also developing continuously, such as: voice recognition technology, and now voice recognition technology is gradually becoming popular, and various smart devices are now basically equipped with voice recognition technology. Due to limited resources, low-power devices such as smart wearable devices basically prepare a certain number of recognition words in advance for support, and the number of recognition words that can be supported is limited.

Most of the current offline speech recognition on smart devices is based on the product vocabulary. According to the pre-set vocabulary of the product, the speech recognition model is trained, and then the audio is sent to the trained speech recognition model to output the posterior probability, and the recognition result is obtained through the posterior probability. However, in the process of model training, it is very likely that it is impossible to predict whether the speaker has some oral habits or speaking habits during the speaking process. If there is a deviation in the content of the vocabulary, it may not be recognized, which will affect the result of speech recognition.

Therefore, there is an urgent need in the art for a speech recognition solution that can improve the accuracy of speech recognition results.

Contents of the invention

In view of this, the embodiments of the present invention provide a voice recognition method, device, device and storage medium, so as to eliminate or improve one or more defects existing in the prior art.

One aspect of the present invention provides a method of speech recognition, the method comprising the following steps:

Receive voice information to be recognized;

Use the speech recognition model to perform speech recognition on the speech information to be recognized, and obtain the posterior data corresponding to the speech information to be recognized in each frame; wherein, the speech recognition model is obtained based on short word vocabulary training, and the short words The vocabulary is obtained by semantic segmentation of each long word in the long word vocabulary;

Decoding the posterior data by using the short word vocabulary, sequentially obtaining a plurality of recognized short words corresponding to the speech information to be recognized;

A speech recognition result of the to-be-recognized speech information is determined according to the long word vocabulary and the plurality of recognized short words.

In some embodiments of the present invention, the training method of the speech recognition model includes:

Obtain the long word vocabulary of the original speech recognition model;

Long words in the long word vocabulary are semantically cut, each long word in the long word vocabulary is cut into a plurality of short words, and the short word vocabulary is obtained;

Collect short word speech samples based on the short word vocabulary, use the short word speech samples to perform model training, and obtain the speech recognition model.

In some embodiments of the present invention, the method also includes:

According to the source of each short word in the short word vocabulary, each short word in the short word vocabulary is set up a mapping relationship with each long word in the long word vocabulary;

The determining the speech recognition result of the to-be-recognized speech information according to the long word vocabulary and the plurality of recognized short words includes:

According to the recognition order corresponding to the multiple recognition short words, combine the multiple recognition short words to obtain the combined words;

According to the mapping relationship of each recognized short word in the combined word in the long word vocabulary, the combined word is matched with the long word vocabulary to obtain the speech recognition result.

In some embodiments of the present invention, the posterior data is the probability of a phoneme corresponding to the speech information to be recognized in each frame;

The decoding of the posterior data by using the short word vocabulary, and sequentially obtaining a plurality of recognized short words corresponding to the speech information to be recognized includes:

According to the ranking of the probability of the phonemes corresponding to the speech information to be recognized in each frame in the posteriori data, obtain a screened phoneme set corresponding to the speech information to be recognized in each frame within a specified ranking;

The short word vocabulary is used to sequentially perform matching decoding on the screened phoneme sets corresponding to the speech information to be recognized in each frame, and sequentially obtain a plurality of recognized short words corresponding to the speech information to be recognized.

In some embodiments of the present invention, the decoding of the posterior data by using the short word vocabulary includes:

The a posteriori data is decoded using the short word vocabulary using a Viterbi decoding algorithm.

In some embodiments of the present invention, the determination of the speech recognition result of the speech information to be recognized according to the long word vocabulary and the plurality of recognized short words includes:

Match the plurality of recognized short words with the long word vocabulary, if the matching fails, continue to recognize the speech information to be recognized by using the speech recognition model, and determine the corresponding speech information of the speech information to be recognized new recognition phrases;

Combining the new recognition short word with the plurality of recognition short words, and matching based on the combined recognition short word and the long word vocabulary, until the long word vocabulary is determined in the long word vocabulary The speech recognition result of the speech information to be recognized.

Another aspect of the present invention provides a speech recognition device, the device comprising:

Voice receiving module, used for receiving the voice information to be recognized;

The posterior data acquisition module is used to perform speech recognition on the speech information to be recognized by using a speech recognition model, and obtain the posterior data corresponding to the speech information to be recognized in each frame; wherein, the speech recognition model is based on short words Table training obtains, and described short word vocabulary carries out semantic segmentation and obtains for each long word in the long word vocabulary;

A short word decoding module, configured to use the short word vocabulary to decode the posterior data, and sequentially obtain a plurality of recognized short words corresponding to the speech information to be recognized;

The recognition result determination module is configured to determine the speech recognition result of the speech information to be recognized according to the long word vocabulary and the plurality of recognized short words.

In some embodiments of the present invention, the device also includes a vocabulary mapping module for: according to the source of each short word in the short word vocabulary, associate each short word in the short word vocabulary with the long word Each long word in the vocabulary establishes a mapping relationship;

The identification result determination module is specifically used for:

According to the recognition order corresponding to the multiple recognition short words, combine the multiple recognition short words to obtain the combined words;

According to the mapping relationship of each recognized short word in the combined word in the long word vocabulary, the combined word is matched with the long word vocabulary to obtain the speech recognition result.

Another aspect of the present invention provides a speech recognition device, including a processor and a memory, wherein computer instructions are stored in the memory, and the processor is used to execute the computer instructions stored in the memory, when the computer instructions are When the processor executes, the device realizes the above speech recognition method.

Another aspect of the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the above speech recognition method is realized.

A kind of speech recognition method, device, equipment and storage medium provided by the present invention cut the long word vocabulary in the speech recognition into short word vocabulary, utilize the short word vocabulary training to build the speech recognition model, when performing speech recognition , use the new speech recognition model to recognize the speech information to be recognized, and determine a plurality of recognized short words corresponding to the speech information to be recognized, based on the combination of recognized short words and the original long word vocabulary, it can be accurately determined Get the voice recognition result of the voice information to be recognized. When using the speech recognition model trained by the short word vocabulary to perform speech recognition on the speech information to be recognized, the modal particles or other irrelevant words in the speech information to be recognized can be removed to ensure that different users' speaking habits can complete the speech recognition process , ensuring the accuracy of the speech recognition result and the success rate of the speech recognition.

Additional advantages, objects, and features of the present invention will be set forth in part in the following description, and will be partly apparent to those of ordinary skill in the art after studying the following text, or can be learned from the practice of the present invention. The objectives and other advantages of the invention may be realized and attained by the structure particularly pointed out in the written description and appended drawings.

It will be understood by those skilled in the art that the objects and advantages that can be achieved by the present invention are not limited to the above specific ones, and the above and other objects that can be achieved by the present invention will be more clearly understood from the following detailed description.

Description of drawings

The drawings described here are used to provide further understanding of the present invention, constitute a part of the application, and do not limit the present invention. The components in the figures are not drawn to scale, merely illustrating the principles of the invention. For ease of illustration and description of some parts of the present invention, corresponding parts in the figures may be exaggerated, ie, may be made larger relative to other components in an exemplary device actually manufactured in accordance with the present invention. In the attached picture:

Fig. 1 is a schematic flow chart of a voice recognition method provided in an embodiment of this specification;

Fig. 2 is a schematic diagram of the training process of the speech recognition model in some embodiments of the present specification;

Fig. 3 is a schematic flow chart of speech recognition in another embodiment of the present specification;

Fig. 4 is a schematic diagram of the module structure of an embodiment of the speech recognition device provided in this specification;

Fig. 5 is a block diagram of the hardware structure of the voice recognition server in one embodiment of the present specification.

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be described in further detail below in conjunction with the embodiments and accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but not to limit the present invention.

Here, it should also be noted that, in order to avoid obscuring the present invention due to unnecessary details, only the structures and/or processing steps closely related to the solution according to the present invention are shown in the drawings, and the related Other details are not relevant to the invention.

It should be emphasized that the term "comprising/comprising" when used herein refers to the presence of a feature, element, step or component, but does not exclude the presence or addition of one or more other features, elements, steps or components.

Here, it should also be noted that, unless otherwise specified, the term "connection" herein may refer not only to a direct connection, but also to an indirect connection with an intermediate.

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar components, or the same or similar steps.

General speech recognition devices, especially low-power speech recognition devices, are equipped with recognition words. During speech recognition, the speech input by the user is matched with the preset recognition words to recognize the speech input by the user. However, speech recognition equipment may be used by various users, and some users may have their own speaking habits when speaking, which may be inconsistent with the preset recognition word content. For example: "Turn on the air conditioner" when the recognition word is preset, but actual users always use the phrases "Turn on the air conditioner" and "Turn on the air conditioner" in daily life. At this time, "Turn on the air conditioner" may not be successfully recognized. recognized. Moreover, this situation is very different, and the differences in speaking habits in different regions cannot be exhaustive, and it is impossible to add these situations one by one to the vocabulary.

In the embodiment of this specification, a speech recognition method is provided, which cuts the long word vocabulary in speech recognition into a short word vocabulary, uses the short word vocabulary to train and build a speech recognition model, and uses the new speech recognition method when performing speech recognition. The model recognizes the speech information to be recognized, and determines multiple recognized short words corresponding to the speech information to be recognized. Based on the combination of recognized short words and the original long word list, it can accurately determine the Speech recognition results. When using the speech recognition model trained by the short word vocabulary to perform speech recognition on the speech information to be recognized, the modal particles or other irrelevant words in the speech information to be recognized can be removed to ensure that different users' speaking habits can complete the speech recognition process , ensuring the accuracy of the speech recognition result and the success rate of the speech recognition.

Fig. 1 is a schematic flow chart of the speech recognition method provided in one embodiment of this specification, as shown in Fig. In terminal devices such as mobile phones and smart wearable devices, the method may include the following steps:

Step 102, receiving voice information to be recognized.

In the specific implementation process, when performing speech recognition, users can speak their own requirements to speech recognition devices such as smart wearable devices or smart speakers. After the speech recognition device recognizes the user's intention, it can output corresponding the result of. For example, in general smart homes, voice recognition devices can be connected to the Internet to control air conditioners, TVs and other household appliances. The user says "turn on the air conditioner" to the voice recognition device, and after the voice recognition device recognizes it, the air conditioner can be turned on. The voice information to be recognized can be the voice spoken by the user to the voice recognition device, it can be Mandarin, dialects or languages of other countries, it can be set according to actual needs, and the embodiment of this specification does not make specific limitations. In the example, the speech information to be recognized is generally in Chinese.

Step 104: Use a speech recognition model to perform speech recognition on the speech information to be recognized, and obtain the posterior data corresponding to each frame of the speech information to be recognized; wherein, the speech recognition model is obtained based on short word vocabulary training, and the The short word vocabulary is obtained by semantic segmentation of each long word in the long word vocabulary.

In the specific implementation process, after receiving the speech information to be recognized, the speech recognition model can be used to perform speech recognition on the speech information to be recognized. The speech recognition model is an intelligent learning model that can recognize words or related information in the speech information to be recognized. The speech recognition model in the embodiments of this specification can be a neural network model, and the specific model structure can be set according to actual needs, which is not specifically limited in the embodiments of this specification. The speech recognition model using the neural network can identify the posterior data corresponding to each frame of the speech information to be recognized, and the posterior data can generally be the probability of the pronunciation or word or recognition result of the speech information to be recognized. In addition, the difference between the speech recognition model in the embodiment of this specification and the existing speech recognition model is that the embodiment of this specification uses short word vocabulary training to obtain, and the short word vocabulary is the long-term Each long word in the vocabulary is obtained by semantic segmentation, which avoids the step of reconfiguring the vocabulary, improves the efficiency of speech recognition model training, and lays a data foundation for subsequent speech recognition. That is to say, the long words in the long word vocabulary preset in the existing voice equipment can be voice-cut to obtain the short words corresponding to each long word, and then obtain the short word vocabulary, and use the short word vocabulary training to build New Speech Recognition Model, Utilize the new Speech Recognition Model for speech recognition. In this way, the modal particles or other irrelevant words in the voice information to be recognized are eliminated, so as to ensure that different users' speaking habits can complete the voice recognition process, and ensure the accuracy of voice recognition results and the success rate of voice recognition.

In some embodiments of this specification, the training method of the speech recognition model includes:

Obtain the long word vocabulary of the original speech recognition model;

Long words in the long word vocabulary are semantically cut, each long word in the long word vocabulary is cut into a plurality of short words, and the short word vocabulary is obtained;

Collect short word speech samples based on the short word vocabulary, use the short word speech samples to perform model training, and obtain the speech recognition model.

In the specific implementation process, Fig. 2 is a schematic diagram of the training process of the speech recognition model in some embodiments of this specification, as shown in Fig. 2, first obtain the long word vocabulary preset in the existing speech recognition device, such as: original speech Identify the long word vocabulary during model training, perform semantic segmentation on the long words in the long word vocabulary, and cut the long words into multiple short words that meet the semantic conditions. For example: the verbs and nouns in the long words in the long word vocabulary can be cut out, for example: the long word "turn on the air conditioner" is cut into two short words "open" and "air conditioner". In actual use, the long word vocabulary can be split by using the semantic module, or the long word vocabulary can be split by using the semantic split algorithm, and the specific split method is not specifically limited in the embodiment of this specification. After splitting the long words in the long word vocabulary into short words, each short word forms a short word vocabulary. Collect short word voice samples based on the short word vocabulary, for example: find relevant personnel to record the voices corresponding to each short word in the short word vocabulary in turn, and obtain short word voice samples. As described in the above embodiments, the model structure of the speech recognition model can be set according to actual needs. The embodiment of this specification does not make specific limitations. After obtaining the short word speech samples, set the model structure, use the short word speech samples for model training, and construct A speech recognition model.

It should be noted that when the short word vocabulary is obtained by cutting the long word vocabulary, only one of the repeated short words needs to be reserved.

The embodiment of this specification utilizes short word vocabulary training to build a speech recognition model, and then makes the speech recognition model constructed by training can recognize the short words in the speech information to be recognized when performing speech recognition, and the modal particles in the speech information to be recognized Or irrelevant words, which are not in the short word vocabulary, the speech recognition model will not recognize them, and some modal particles that users are accustomed to can be eliminated to avoid speech recognition failures caused by users' habitual words not in the vocabulary. Combining the recognized short words can accurately and quickly determine the speech recognition result, which improves the accuracy and success rate of speech recognition.

Step 106 , using the short word vocabulary to decode the posteriori data, and sequentially obtain a plurality of recognized short words corresponding to the speech information to be recognized.

In the specific implementation process, after using the speech recognition model to recognize the speech information to be recognized, the model outputs the posterior data of each frame of the speech information to be recognized, and then uses the short word vocabulary used in speech recognition model training to carry out the posteriori data By matching and decoding, multiple recognized short words corresponding to the voice information to be recognized can be obtained. In some embodiments of this specification, during decoding, the a posteriori data may be decoded by using the short word vocabulary using a Viterbi decoding algorithm. The Viterbi decoding algorithm is the viterbi algorithm. The Viterbi algorithm is a general dynamic programming algorithm for finding the shortest path of a sequence. Using the Viterbi decoding algorithm, it can quickly and accurately determine the matching of the short word vocabulary in the posterior data output by the speech recognition model. Short words, thereby improving the efficiency and accuracy of speech recognition.

In some embodiments of this specification, the a posteriori data is the probability of a phoneme corresponding to the speech information to be recognized in each frame;

The decoding of the posterior data by using the short word vocabulary, and sequentially obtaining a plurality of recognized short words corresponding to the speech information to be recognized includes:

According to the ranking of the probability of the phonemes corresponding to the speech information to be recognized in each frame in the posteriori data, obtain a screened phoneme set corresponding to the speech information to be recognized in each frame within a specified ranking;

The short word vocabulary is used to sequentially perform matching decoding on the screened phoneme sets corresponding to the speech information to be recognized in each frame, and sequentially obtain a plurality of recognized short words corresponding to the speech information to be recognized.

In the specific implementation process, usually when performing speech recognition, it is necessary to first perform feature extraction on the speech information to be recognized, such as: extract the pronunciation features in the speech information to be recognized, and input the extracted features to the speech recognition model, model or output corresponding recognition results. As described in the above-mentioned embodiments, the speech recognition model in the embodiments of this specification can choose a neural network model, and when the speech recognition model of the neural network structure is used for speech recognition of speech information to be recognized, the model can output each frame of speech information to be recognized The probability of the corresponding phoneme. That is to say, the speech recognition model can output the probability of pronunciation corresponding to each frame of speech information to be recognized. Among them, a phoneme is the smallest speech unit divided according to the natural attributes of speech, and is analyzed according to the pronunciation action in a syllable. An action constitutes a phoneme. For example, the speech information to be recognized is "turn on the air conditioner", and the speech recognition model can output the phonemes "d, a" corresponding to the first word "打" when performing speech recognition, with a probability of 90%. Wherein, the voice corresponding to one word of the voice information to be recognized may be one or more frames of audio, depending on the actual situation, which is not specifically limited in the embodiment of this specification. There can be one or more phonemes corresponding to each frame of speech recognition information to be recognized output by the speech recognition model, and each phoneme corresponds to a posteriori probability, which can be based on the corresponding The ranking of the probability of the phonemes is to screen out the phonemes corresponding to each frame whose ranking is within the specified ranking, and obtain the screened phoneme set corresponding to the voice information to be recognized in each frame. Then, the short word vocabulary is used to sequentially match and decode the selected phoneme sets corresponding to the speech information to be recognized in each frame, and then sequentially obtain a plurality of recognized short words corresponding to the speech information to be recognized.

For example: the speech information to be recognized is "Turn on the air conditioner", and the output of speech recognition based on the speech recognition model is to obtain the filtered phoneme set corresponding to the first frame as: "d, t", and the filtered phoneme set corresponding to the second frame is : "a, ai, ao", there may be some irrelevant frames in the middle, and then the screening phoneme set corresponding to the fifth frame is: "k, g, l", and the screening phoneme set corresponding to the sixth frame is: "ai, ao, i, a" After combining the screened phoneme sets corresponding to the multi-frame speech, search in the short word vocabulary, and find the combination of "open" in the short word vocabulary and the screened phoneme set of the multi-frame speech d, a", "k, a, i" match, then it can be determined that a short word corresponding to the speech information to be recognized is "open".

In the embodiment of this specification, the speech recognition model is used to identify the pronunciation corresponding to each frame of speech information to be recognized and its posterior probability, and then the phoneme corresponding to each frame of speech information to be recognized is determined, and then the phoneme is performed using the short word list Matching decoding recognizes short words in the voice information to be recognized, laying a data foundation for subsequent voice recognition.

Step 108: Determine the speech recognition result of the speech information to be recognized according to the long word vocabulary and the plurality of recognized short words.

In the specific implementation process, after identifying a plurality of short words corresponding to the speech information to be recognized, the long word vocabulary can be used to match the recognized short words, and then determine the speech recognition result of the speech information to be recognized . For example, various recognized short words can be combined, and the combined short words can be matched with the long word vocabulary to determine the speech recognition result. The short word vocabulary in the embodiment of this specification is obtained by splitting the long word vocabulary, and the speech recognition model is obtained based on the training of the short word vocabulary, and the recognition short words recognized based on the speech recognition model belong to the short word vocabulary, and the short word The short words in the word list can be obtained in the long word list. Therefore, matching multiple recognized short words with the long word list can accurately and quickly obtain the speech recognition result of the speech information to be recognized.

In some embodiments of this specification, the method also includes:

According to the source of each short word in the short word vocabulary, each short word in the short word vocabulary is set up a mapping relationship with each long word in the long word vocabulary;

The determining the speech recognition result of the to-be-recognized speech information according to the long word vocabulary and the plurality of recognized short words includes:

According to the recognition order corresponding to the multiple recognition short words, combine the multiple recognition short words to obtain the combined words;

According to the mapping relationship of each recognized short word in the combined word in the long word vocabulary, the combined word is matched with the long word vocabulary to obtain the speech recognition result.

In the specific implementation process, after obtaining the short word vocabulary, each short word in the short word vocabulary can be established according to the source of each short word in the short word vocabulary, that is, which long word the short word is obtained from The mapping relationship with each long word in the long word vocabulary. For example: there is "open" in the short word vocabulary, and "turn on the air conditioner", "turn on the TV" and "turn on the light in the master bedroom" in the long word vocabulary, it can be seen that these three long words in the long word vocabulary all include "open ", then the source of the short word "open" in the short word vocabulary can come from these three long words, then "open" in the short word vocabulary and "turn on the air conditioner" and "open" in the long word vocabulary TV" and "Turn on the light in the master bedroom" to establish a mapping relationship. After determining a plurality of recognized short words in the speech information to be recognized, the plurality of recognized short words can be combined according to the order of the plurality of recognized short words in the speech information to be recognized or the recognition order of the plurality of recognized short words , to obtain compound words. Then, according to the mapping relationship of each recognized short word in the compound word in the long word vocabulary, the compound word can be quickly matched with the long word vocabulary to obtain the speech recognition result.

For example: multiple recognized short words determined based on the voice information to be recognized include "turn on" and "air conditioner", which are combined to "turn on the air conditioner". When matching with the long word vocabulary, it is found that the short words "open" and "air conditioner" all have a mapping relationship with the long word "turn on the air conditioner" in the long word vocabulary, and the combined combined words and the long word vocabulary If the words are the same, it can be determined that the speech recognition result is "turn on the air conditioner".

In the embodiment of this specification, when splitting the long word vocabulary to obtain the short word vocabulary, the mapping relationship between the short word vocabulary and the long word vocabulary is also established according to the process of splitting. When determining the recognition result, based on two The mapping relationship between the vocabulary can quickly and accurately find the results in the long word vocabulary, and then quickly obtain the speech recognition result, which improves the efficiency of speech recognition.

Fig. 3 is a schematic flow chart of speech recognition in another embodiment of this specification. As shown in Fig. 3, in some embodiments of this specification, according to the long word vocabulary and the plurality of recognized short words, determine the Describe the voice recognition results of the voice information to be recognized, including:

Match the plurality of recognized short words with the long word vocabulary, if the matching fails, continue to recognize the speech information to be recognized by using the speech recognition model, and determine the corresponding speech information of the speech information to be recognized new recognition phrases;

Combining the new recognition short word with the plurality of recognition short words, and matching based on the combined recognition short word and the long word vocabulary, until the long word vocabulary is determined in the long word vocabulary The speech recognition result of the speech information to be recognized.

In the specific implementation process, refer to the records of the above-mentioned embodiments. When performing speech recognition, the speech recognition model can be used to recognize and decode the recognized short words in the speech information to be recognized, and each recognized short word can be combined with the long word Vocabulary for matching, if the matching fails, that is to say, no long word matching the combination of multiple recognized short words is found in the long word vocabulary, then use the speech recognition model to continue to recognize and decode the speech information to be recognized, and obtain New recognition short words, combine the new recognition short words with multiple recognition short words previously recognized, and match the combined words with the long word vocabulary, if the matching is successful, then output the speech recognition result, if If the matching fails, continue to use the speech recognition model to continue to recognize and decode the speech information to be recognized, and repeat the above actions until the combined recognized short words are successfully matched with the long words in the long word vocabulary, then the speech recognition is completed.

In this embodiment of the present specification, the contents of the recognized words in the long word vocabulary are semantically analyzed, and these recognized words are segmented according to the semantics, so as to become shorter recognized words. In the process of speech recognition, if such a recognized word after being cut is recognized, the recognition result is not output, but wait for a certain period of time to see whether the second half of the cut word appears later. If the recognized word in the second half appears in the subsequent audio, the recognized word before outputting is output to obtain the speech recognition result. Avoid the problem that different users have different speaking habits, resulting in a mismatch between the speech to be recognized and the content of the vocabulary, and the failure of speech recognition, and improve the accuracy and success rate of speech recognition.

As shown in Figures 1-3 above, the speech recognition process in the embodiment of this specification can be referred to as follows:

Step 1. When the long word vocabulary is obtained, the long word vocabulary is split by the semantic module, and the long word is cut into multiple short words that meet the semantic conditions.

Step 2. Use these short words to create a new short word vocabulary, and establish a mapping relationship with the original long word vocabulary.

Step 3. Train the speech recognition model using the short word vocabulary.

Step 4. Decoding the posterior data output by the speech recognition model of the neural network structure with the short word vocabulary to obtain the recognized short words.

Step 5. The short word recognition result that is output each time is the short word recognition record, and the subsequent recognition result is combined with the previous recognition result to search in the long word vocabulary.

Step 6. Output the final speech recognition result after finding the recognized word in the long word vocabulary that meets the requirements.

The speech recognition method provided by the embodiment of this specification is aimed at the existing speech recognition equipment, and it is found that although different people have different speaking habits, they all contain pre-set recognition words, and only add recognition words on the basis of recognition words. structure in certain grammars. In the embodiment of this specification, the longer recognition words are split into smaller elements according to the semantics, and the words in the vocabulary are split into smaller basic vocabulary elements, and the recognition results are also recognized according to the smallest word elements, so that It can skip the special idiomatic words of the speaker in the middle, and match the actual content of the speaker with the content of the vocabulary, thereby improving the accuracy and success rate of speech recognition.

Each embodiment of the above-mentioned method in this specification is described in a progressive manner, the same and similar parts of each embodiment can be referred to each other, and each embodiment focuses on the difference from other embodiments. For relevant parts, please refer to part of the descriptions of the method embodiments.

Based on the speech recognition method described above, one or more embodiments of this specification further provide a speech recognition device. The device may include devices (including distributed systems), software (applications), modules, components, servers, clients, etc. that use the methods described in the embodiments of this specification combined with necessary implementation hardware. Based on the same innovative idea, the devices in one or more embodiments provided by the embodiments of this specification are as described in the following embodiments. Since the implementation of the device to solve the problem is similar to the method, the implementation of the specific device in the embodiment of this specification can refer to the implementation of the aforementioned method, and the repetition will not be repeated. As used below, the term "unit" or "module" may be a combination of software and/or hardware that realizes a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementations in hardware, or a combination of software and hardware are also possible and contemplated.

Specifically, FIG. 4 is a schematic diagram of the module structure of an embodiment of the speech recognition device provided in this specification. As shown in FIG. 4, the device provided in this specification may include:

Voice receiving module 41, for receiving voice information to be recognized;

The posterior data acquisition module 42 is used to perform speech recognition on the speech information to be recognized by using a speech recognition model, and obtain the posterior data corresponding to the speech information to be recognized in each frame; wherein, the speech recognition model is based on short words Vocabulary training obtains, and described short word vocabulary carries out semantic segmentation and obtains for each long word in the long word vocabulary;

The short word decoding module 43 is used to decode the posteriori data by using the short word vocabulary, and sequentially obtain a plurality of recognized short words corresponding to the voice information to be recognized;

The recognition result determination module 44 is configured to determine the speech recognition result of the speech information to be recognized according to the long word vocabulary and the plurality of recognized short words.

In some embodiments of this specification, the device also includes a vocabulary mapping module, configured to: associate each short word in the short word vocabulary with the long word word according to the source of each short word in the short word vocabulary Each long word in the table establishes a mapping relationship;

The identification result determination module is specifically used for:

According to the recognition order corresponding to the multiple recognition short words, combine the multiple recognition short words to obtain the combined words;

According to the mapping relationship of each recognized short word in the combined word in the long word vocabulary, the combined word is matched with the long word vocabulary to obtain the speech recognition result.

The speech recognition device provided by the embodiment of this specification can perform speech recognition in multiple different languages on the target text, realizes cross-language transfer, and can flexibly combine languages and timbres, which improves the flexibility and accuracy of speech recognition, reduces the the cost of speech recognition.

In some embodiments of this specification, there is also provided a language synthesis device, including a processor and a memory, the memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory, when the computer When the instruction is executed by the processor, the device implements the voice recognition method recorded in the above embodiments, such as:

Receive voice information to be recognized;

Use the speech recognition model to perform speech recognition on the speech information to be recognized, and obtain the posterior data corresponding to the speech information to be recognized in each frame; wherein, the speech recognition model is obtained based on short word vocabulary training, and the short words The vocabulary is obtained by semantic segmentation of each long word in the long word vocabulary;

Decoding the posterior data by using the short word vocabulary, sequentially obtaining a plurality of recognized short words corresponding to the speech information to be recognized;

A speech recognition result of the to-be-recognized speech information is determined according to the long word vocabulary and the plurality of recognized short words.

It should be noted that the above-mentioned apparatus and equipment may also include other implementation manners according to the description of the method embodiment. For specific implementation manners, reference may be made to descriptions of related method embodiments, and details are not repeated here.

The method embodiments provided in the embodiments of this specification may be executed in mobile terminals, computer terminals, servers or similar computing devices. Taking running on a server as an example, FIG. 5 is a block diagram of the hardware structure of the speech recognition server in an embodiment of this specification. The computer terminal may be the speech recognition server or the speech recognition device in the above embodiments. As shown in Figure 5, the server 10 may include one or more (only one is shown in the figure) processors 100 (the processor 100 may include but not limited to processing devices such as microprocessor MCU or programmable logic device FPGA, etc.), A non-volatile memory 200 for storing data, and a transmission module 300 for communication functions. Those of ordinary skill in the art can understand that the structure shown in FIG. 5 is only a schematic diagram, which does not limit the structure of the above-mentioned electronic device. For example, the server 10 may also include more or fewer components than those shown in FIG. 5 , for example, may also include other processing hardware, such as a database or multi-level cache, GPU, or have a configuration different from that shown in FIG. 5 .

The non-volatile memory 200 can be used to store software programs and modules of application software, such as program instructions/modules corresponding to the speech recognition method in the embodiment of this specification, and the processor 100 can run the software stored in the non-volatile memory 200 Programs and modules to perform various functional applications and resource data updates. The non-volatile memory 200 may include high-speed random access memory, and may also include non-volatile memories, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memories. In some examples, the non-volatile memory 200 may further include a memory that is remotely located relative to the processor 100, and these remote memories may be connected to a computer terminal through a network. Examples of the aforementioned networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

The transmission module 300 is used to receive or transmit data via a network. The specific example of the above-mentioned network may include a wireless network provided by the communication provider of the computer terminal. In one example, the transmission module 300 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In an example, the transmission module 300 may be a radio frequency (Radio Frequency, RF) module, which is used to communicate with the Internet in a wireless manner.

Corresponding to the above method, the present invention also provides an apparatus, the apparatus includes computer equipment, the computer equipment includes a processor and a memory, the memory stores computer instructions, and the processor is used to execute the instructions in the memory. Stored computer instructions, when the computer instructions are executed by the processor, the device implements the steps of the aforementioned method.

An embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the aforementioned method for deploying an edge computing server can be implemented. The computer readable storage medium may be a tangible storage medium such as random access memory (RAM), internal memory, read only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, floppy disk, hard disk, removable storage disk, CD-ROM, or any other form of storage medium known in the art.

Those of ordinary skill in the art should understand that each exemplary component, system and method described in conjunction with the embodiments disclosed herein can be implemented by hardware, software or a combination of the two. Whether it is implemented in hardware or software depends on the specific application and design constraints of the technical solution. Skilled artisans may use different methods to implement the described functions for each specific application, but such implementation should not be regarded as exceeding the scope of the present invention. When implemented in hardware, it may be, for example, an electronic circuit, an application specific integrated circuit (ASIC), suitable firmware, a plug-in, a function card, or the like. When implemented in software, the elements of the invention are the programs or code segments employed to perform the required tasks. Programs or code segments can be stored in machine-readable media, or transmitted over transmission media or communication links by data signals carried in carrier waves.

It is to be understood that the invention is not limited to the specific arrangements and processes described above and shown in the drawings. For conciseness, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the sequence of steps after understanding the spirit of the present invention.

In the present invention, features described and/or exemplified for one embodiment can be used in the same or similar manner in one or more other embodiments, and/or can be combined with features of other embodiments or replace other Features of the implementation.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. For those skilled in the art, various modifications and changes may be made to the embodiments of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims (10)

1. A speech recognition method, characterized in that the method comprises: Receive voice information to be recognized; Use the speech recognition model to perform speech recognition on the speech information to be recognized, and obtain the posterior data corresponding to the speech information to be recognized in each frame; wherein, the speech recognition model is obtained based on short word vocabulary training, and the short words The vocabulary is obtained by semantic segmentation of each long word in the long word vocabulary; Decoding the posterior data by using the short word vocabulary, sequentially obtaining a plurality of recognized short words corresponding to the speech information to be recognized; A speech recognition result of the to-be-recognized speech information is determined according to the long word vocabulary and the plurality of recognized short words. 2. method according to claim 1, is characterized in that, the training method of described speech recognition model comprises: Obtain the long word vocabulary of the original speech recognition model; Long words in the long word vocabulary are semantically cut, each long word in the long word vocabulary is cut into a plurality of short words, and the short word vocabulary is obtained; Collect short word speech samples based on the short word vocabulary, use the short word speech samples to perform model training, and obtain the speech recognition model. 3. The method according to claim 1, wherein the method further comprises: According to the source of each short word in the short word vocabulary, each short word in the short word vocabulary is set up a mapping relationship with each long word in the long word vocabulary; The determining the speech recognition result of the to-be-recognized speech information according to the long word vocabulary and the plurality of recognized short words includes: According to the recognition order corresponding to the multiple recognition short words, combine the multiple recognition short words to obtain the combined words; According to the mapping relationship of each recognized short word in the combined word in the long word vocabulary, the combined word is matched with the long word vocabulary to obtain the speech recognition result. 4. The method according to claim 1, wherein the a posteriori data is the probability of a phoneme corresponding to the speech information to be recognized in each frame; The decoding of the posterior data by using the short word vocabulary, and sequentially obtaining a plurality of recognized short words corresponding to the speech information to be recognized includes: According to the ranking of the probability of the phonemes corresponding to the speech information to be recognized in each frame in the posteriori data, obtain a screened phoneme set corresponding to the speech information to be recognized in each frame within a specified ranking; The short word vocabulary is used to sequentially perform matching decoding on the screened phoneme sets corresponding to the speech information to be recognized in each frame, and sequentially obtain a plurality of recognized short words corresponding to the speech information to be recognized. 5. The method according to claim 1, wherein said utilizing said short word vocabulary to decode said posteriori data comprises: The a posteriori data is decoded using the short word vocabulary using a Viterbi decoding algorithm. 6. The method according to any one of claims 1-5, wherein the speech recognition result of the to-be-recognized speech information is determined according to the long word vocabulary and the plurality of recognized short words ,include: Match the plurality of recognized short words with the long word vocabulary, if the matching fails, continue to recognize the speech information to be recognized by using the speech recognition model, and determine the corresponding speech information of the speech information to be recognized new recognition phrases; Combining the new recognition short word with the plurality of recognition short words, and matching based on the combined recognition short word and the long word vocabulary, until the long word vocabulary is determined in the long word vocabulary The speech recognition result of the speech information to be recognized. 7. A speech recognition device, characterized in that the device comprises: Voice receiving module, used for receiving the voice information to be recognized; The posterior data acquisition module is used to perform speech recognition on the speech information to be recognized by using a speech recognition model, and obtain the posterior data corresponding to the speech information to be recognized in each frame; wherein, the speech recognition model is based on short words Table training obtains, and described short word vocabulary carries out semantic segmentation and obtains for each long word in the long word vocabulary; A short word decoding module, configured to use the short word vocabulary to decode the posterior data, and sequentially obtain a plurality of recognized short words corresponding to the speech information to be recognized; The recognition result determination module is configured to determine the speech recognition result of the speech information to be recognized according to the long word vocabulary and the plurality of recognized short words. 8. The device according to claim 7, characterized in that, the device also includes a vocabulary mapping module for: according to the source of each short word in the short word vocabulary, each short word in the short word vocabulary Each long word in the word and described long word vocabulary establishes mapping relationship; The identification result determination module is specifically used for: According to the recognition order corresponding to the multiple recognition short words, combine the multiple recognition short words to obtain the combined words; According to the mapping relationship of each recognized short word in the combined word in the long word vocabulary, the combined word is matched with the long word vocabulary to obtain the speech recognition result. 9. A speech recognition device, comprising a processor and a memory, wherein computer instructions are stored in the memory, and the processor is used to execute the computer instructions stored in the memory, when the computer instructions are executed by the processor When executed, the device implements the steps of the method according to any one of claims 1-6. 10. A computer-readable storage medium, on which a computer program is stored, wherein, when the program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are realized.

Images