WO2021128880A1

WO2021128880A1 - Speech recognition method, device, and device for speech recognition

Info

Publication number: WO2021128880A1
Application number: PCT/CN2020/110037
Authority: WO
Inventors: 郑宏
Original assignee: 北京搜狗科技发展有限公司
Priority date: 2019-12-26
Filing date: 2020-08-19
Publication date: 2021-07-01
Also published as: CN111145756A; CN111145756B

Abstract

A speech recognition method and a device (800) for speech recognition. The method specifically comprises: receiving speech information inputted by a user (101); obtaining a personalized vocabulary of the user, wherein the personalized vocabulary is established according to historical input content generated according to a process during which the user uses an input method (102); according to the personalized vocabulary, determining a decoding path weight corresponding to the speech information (103); and according to the decoding path weight, determining a speech recognition result corresponding to the speech information (104). The described method and device (800) may improve the accuracy of speech recognition.

Description

Speech recognition method, device and device for speech recognition

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office on December 26, 2019, the application number is 201911369489.9, and the invention title is "a voice recognition method, device and device for voice recognition", and its entire contents Incorporated in this application by reference.

Technical field

This application relates to the field of computer technology, and in particular to a voice recognition method, device, and device for voice recognition.

Background technique

Voice recognition technology is a technology that allows machines to convert voice signals into corresponding texts or commands through the process of recognition and understanding. With the continuous development of science and technology, speech recognition technology has been developed rapidly, the accuracy of speech recognition has been continuously improved, and the application in the field of human-computer interaction is gradually expanding.

In practical applications, when different users speak the same pronunciation, they may correspond to different meanings. For example, when user A says "libing", the voice recognition result that user A wants is "Li Bing", and when user B says "libing", the voice recognition result that user B wants is "Li Bing". In this case, the accuracy of speech recognition may be low.

Summary of the invention

The embodiments of the present application provide a voice recognition method, device, and device for voice recognition, which can improve the accuracy of voice recognition.

In order to solve the above-mentioned problem, an embodiment of the present application discloses a voice recognition method, and the method includes:

Receive voice information input by the user;

Acquiring a personalized vocabulary of the user, the personalized vocabulary being established based on historical input content generated during the user's use of the input method;

Determining the decoding path weight corresponding to the voice information according to the personalized vocabulary;

According to the decoding path weight, the speech recognition result corresponding to the speech information is determined.

On the other hand, an embodiment of the present application discloses a voice recognition device, which includes:

The voice receiving module is used to receive the voice information input by the user;

Thesaurus acquisition module, configured to acquire the user's personalized thesaurus, the personalized thesaurus is established based on the historical input content generated during the user's use of the input method;

A weight determination module, configured to determine the decoding path weight corresponding to the voice information according to the personalized vocabulary;

The result determining module is configured to determine the voice recognition result corresponding to the voice information according to the decoding path weight.

In another aspect, an embodiment of the present application discloses a device for speech recognition, including a memory, and one or more programs, wherein one or more programs are stored in the memory and configured to be composed of one or more programs. The execution of the one or more programs by the above processor includes instructions for performing the following operations:

Receive voice information input by the user;

In another aspect, an embodiment of the present application discloses a machine-readable medium with instructions stored thereon, which when executed by one or more processors, cause the device to execute one or more of the aforementioned voice recognition methods.

The embodiments of the present application include the following advantages:

After receiving the voice information input by the user, the embodiment of the application obtains the user's personalized vocabulary, and determines the decoding path weight corresponding to the voice information according to the personalized vocabulary, and according to the decoding path weight To determine the voice recognition result corresponding to the voice information. Since the personalized thesaurus is established based on the historical input content generated during the user’s use of the input method, the personalized dictionary conforms to the user’s input habits, and can be based on the user’s input in the process of decoding the voice information input by the user. The personalized vocabulary of ”performs real-time weighting on the speech decoding path, so that the final recognition result tends to the user’s input habits, thereby improving the accuracy of speech recognition.

Description of the drawings

In order to explain the technical solutions of the embodiments of the present application more clearly, the following will briefly introduce the drawings that need to be used in the description of the embodiments of the present application. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative labor.

Fig. 1 is a flowchart of steps of an embodiment of a speech recognition method of the present application;

Fig. 2 is a structural block diagram of an embodiment of a speech recognition device of the present application;

FIG. 3 is a block diagram of a device 800 for speech recognition according to the present application; and

Fig. 4 is a schematic diagram of the structure of a server in some embodiments of the present application.

Detailed ways

The technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of them. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of this application.

Method embodiment

1, it shows a flowchart of steps of an embodiment of a voice recognition method of the present application, which may specifically include the following steps:

Step 101: Receive voice information input by a user;

Step 102: Obtain a personalized vocabulary of the user, where the personalized vocabulary is established based on historical input content generated when the user uses the input method;

Step 103: Determine the decoding path weight corresponding to the voice information according to the personalized vocabulary;

Step 104: Determine a voice recognition result corresponding to the voice information according to the decoding path weight.

The voice recognition method in the embodiments of this application can be applied to electronic devices, including but not limited to: servers, smart phones, tablets, e-book readers, MP3 (Moving Picture Experts Compression Standard Audio Level 3, Moving Picture Experts) Group Audio Layer III) Players, MP4 (Moving Picture Experts Group Audio Layer IV) Players, laptop portable computers, car computers, desktop computers, set-top boxes, smart TVs, wearables Equipment and so on.

The voice recognition method of the embodiment of the present application can be used to automatically recognize the voice information input by the user, and convert the voice information into corresponding text information. The voice information may be a continuous voice, such as a sentence, a sentence, and so on. It can be understood that the embodiment of the present application does not limit the source of the voice information. For example, the voice information may be a voice segment collected in real time through the recording function of the electronic device.

In an optional embodiment of the present application, the obtaining the voice information input by the user may specifically include: obtaining the voice information input or sent or received by the user through an instant messaging application.

Wherein, the instant messaging application is an application program that implements online chat and communication through instant messaging technology. The voice information acquired by the embodiment of the present application may include: voice information input by the user through an instant messaging application, voice information sent by the user to the communication peer through the instant messaging application, and voice information received by the user from the communication peer through the instant messaging application.

In the embodiment of the present application, the voice information input by the user may be divided into multiple frames of voice segments according to the preset window length and frame shift. Among them, the window length can be used to represent the duration of each frame of speech segment, and the frame shift can be used to represent the time difference between adjacent frames. For example, when the window length is 25ms and the frame is shifted by 15ms, the first frame of speech segment is 0-25ms, the second frame of speech segment is 15-40ms, and so on. The specific window length and frame shift can be set according to actual needs, which is not limited in the embodiment of the present application.

Optionally, before dividing the voice information into multiple frames of voice segments, the electronic device may also perform noise reduction processing on the voice information, so as to improve the subsequent system's ability to process the information.

In order to improve the accuracy of speech recognition, this embodiment of the application pre-establishes a personalized vocabulary for users. The personalized vocabulary is established based on the historical input content generated during the user's use of the input method, and can reflect the user's input. habit.

In the current speech recognition process, the main problem that causes the speech recognition to be inaccurate is the error in the replacement of homophones. For example, when user A says "libing", the voice recognition result that user A wants is "Li Bing", and when user B says "libing", the voice recognition result that user B wants is "Li Bing". For the homophone "libing", different users have different speech recognition intentions.

The embodiment of this application establishes its own personalized word database for each user based on the historical input content generated by each user in the process of using the input method. The personalized word database conforms to the user's input habits and is based on the user's personalized word database. , The decoding path weight corresponding to the voice information input by the user can be determined, and then the speech recognition result corresponding to the speech information can be determined according to the decoding path weight.

Among them, the historical input content generated during the user's use of the input method may specifically include: text content that has been on the screen before the current cursor position, or text content copied by the user. The historical input content may also be text content input by the user in an instant messaging application and sent to the communication peer, or may also be text content input by the user in an input environment such as a browser, document, microblog, and mail; It can be understood that the embodiment of the present application does not limit the specific source of the historical input content.

In an application example of this application, when the "libing" voice information input by user A is received, the voice decoder combines user A's personalized vocabulary in the process of decoding user A's voice information. It is known that the historical input content generated during the input method of user A often contains "Li Bing". Therefore, additional weight can be added to the decoding path containing "Li Bing", so that the decoding path is preferentially selected, and user A's voice is obtained. The voice recognition result of the message is "Li Bing".

Therefore, using the voice recognition method of the present application, after receiving the voice information input by the user through the microphone, the voice decoder can automatically load the user’s personalized vocabulary, and then decode the personalized vocabulary that hits the personalized vocabulary during the decoding process. The path is weighted in real time, so that the final recognition result is inclined to the user's input habits, which can improve the accuracy of speech recognition.

In an optional embodiment of the present application, before the obtaining the user's personalized vocabulary, the method may further include:

Step S11: Collect historical input content generated by the user in the process of using the input method;

Step S12, preprocessing the historical input content to obtain preprocessed historical input content;

Step S13: Filter non-personalized words in the preprocessed historical input content to obtain personalized words;

Step S14: Establish a personalized vocabulary of the user according to the personalized vocabulary.

The embodiment of the present application can save the historical input content generated by the user through the input method when the user uses the input method, and establish the user's personalized vocabulary based on the historical input content.

It can be understood that the historical input content can be stored in the user's terminal device, or can be stored in a cloud server. The embodiment of the present application may preprocess the user's historical input content, for example, perform data cleaning on the historical input content, remove noise data, etc., to obtain the preprocessed historical input content. Then filter the non-personalized vocabulary in the preprocessed historical input content to obtain the personalized vocabulary. Wherein, the non-personalized vocabulary may include conjunctions, prepositions, function words and so on. The personalized vocabulary may include common person names, place names, organization names, personal idioms, domain interest vocabulary, internet hot words, etc. It can be understood that the type of the personalized vocabulary can be preset according to actual needs, and the embodiment of the present application does not limit the type of the personalized vocabulary. The personalized vocabulary can be used as a kind of portrait description of the user.

The personalized vocabulary can be stored in the user's terminal device, or can be stored in a cloud server, and the user's personalized vocabulary has a one-to-one correspondence with the user's identity information. For example, the correspondence relationship between the user identification and the user's personalized vocabulary can be established, or the correspondence between the user's voiceprint characteristics and the user's personalized vocabulary can also be established.

After the user inputs a piece of voice into the microphone of the terminal device, the user identification of the user, such as a login account, can be obtained; or, voiceprint recognition of the voice input by the user can be performed to obtain the voiceprint characteristics of the user. According to the user’s user identification or voiceprint characteristics, the user’s personalized vocabulary can be loaded from the cloud server. In the process of voice recognition of the voice information input by the user, the weight of the decoding path that hits the personalized vocabulary It can be enhanced in real time, so that the final voice recognition result tends to the personalized vocabulary in the user's personalized vocabulary, which conforms to the user's input habits.

Optionally, the user's personalized vocabulary can be updated periodically. For example, after the user’s personalized vocabulary is established, the input content generated by the user through the input method can be obtained in real time, and the input content can be preprocessed and filtered to obtain personalized vocabulary, which is added to the user’s existing personalized vocabulary In this way, the user’s personalized vocabulary can be continuously updated to adapt to the user’s new input habits and preferences.

In an optional embodiment of the present application, after the establishment of the user's personalized vocabulary, the method may further include:

Step S21: Calculate the similarity between the personalized word databases of at least two users;

Step S22: If it is determined that the similarity satisfies the preset condition, merge the personalized vocabulary of the at least two users to obtain a combined personalized vocabulary.

In specific applications, some users have similar input habits and usually have the same or similar speech recognition results. Therefore, the embodiment of the present application calculates the similarity of the personalized vocabulary of different users to determine whether the user has similar input habits, and merges the personalized vocabulary of users with similar input habits to realize the personality of the user. The lexicon is expanded.

Specifically, if the similarity between the personalized vocabulary of at least two users meets the preset condition, indicating that the at least two users have similar input habits, the personalized vocabulary of the at least two users can be merged , Get the merged personalized vocabulary.

For example, for user U and user V, if the similarity between user U's personalized vocabulary N(u) and user V's personalized vocabulary N(v) meets a preset condition, then N(u) Merge with N(v) to obtain the merged personalized vocabulary N(uv). The merged personalized vocabulary N(uv) includes the personalized words in N(u) and the words in N(v) Personalized vocabulary. In step 102, obtaining the personalized vocabulary of the user may specifically include: obtaining the merged personalized vocabulary.

In the process of recognizing the voice information input by the user U, the merged personalized vocabulary N(uv) can be loaded, and the decoding path weight corresponding to the voice information can be determined according to the personalized vocabulary in N(uv). Similarly, in the process of recognizing the voice information input by user V, the merged personalized vocabulary N(uv) can be loaded, and the decoding path corresponding to the voice information can be determined according to the personalized vocabulary in N(uv) Weights. As a result, both the personalized vocabulary of user U and user V have been expanded, which can enrich the user's personalized vocabulary and further improve the accuracy of speech recognition.

In an optional embodiment of the present application, the calculating the similarity between the personalized vocabulary of at least two users may specifically include:

Step S31: Calculate the cosine distance between the personalized word databases of the at least two users according to the common vocabulary included in the personalized word databases of the at least two users;

Step S32: Calculate the similarity between the personalized vocabulary of the at least two users according to the cosine distance.

Generally, if the personalized vocabulary of two users contains more common personalized vocabulary, it means that the input habits of the two users are more similar, that is, the similarity of the personalized vocabulary of the two users is higher. Therefore, the embodiment of the present application can calculate the similarity between users through the common personalized vocabulary included in the user personalized vocabulary.

Specifically, the embodiment of the present application may calculate the cosine distance between the personalized word databases of at least two users according to the common vocabulary included in the personalized word databases of the at least two users. Wherein, the common vocabulary refers to the number of common personalized vocabulary.

For example, for user U and user V, let N(u) denote user U’s personalized vocabulary, including user U’s personalized vocabulary; let N(v) denote user V’s personalized vocabulary, including user V Personalized vocabulary. _{Then, the similarity W UV} between N(u) and N(v) can be calculated by the cosine distance, as follows:

According to formula (1), it can be seen that the more common vocabulary of N(u) and N(v), the higher the similarity between them. Step S22 determines that the similarity between the personalized vocabularies of the at least two users meets a preset condition, which may specifically include: if the cosine distance between the personalized vocabularies of the at least two users is less than a preset threshold, It is determined that the similarity between the personalized word databases of the at least two users meets the preset condition. That is, if W _{UV is} less than the preset threshold, it can be determined that the similarity between user U's personalized vocabulary N(u) and user V's personalized vocabulary N(v) satisfies the preset condition, and N(u ) And N(v) can be combined.

Step S41: Classify the personalized vocabulary according to the field corresponding to the personalized vocabulary in the personalized vocabulary;

Step S42: Perform statistics on the personalized vocabulary corresponding to different fields in the personalized vocabulary, and determine the personalized label corresponding to the personalized vocabulary.

In the embodiment of the present application, after establishing the user's personalized vocabulary, the field to which the personalized vocabulary belongs in the user's personalized vocabulary can also be determined, and then the user's personalized vocabulary can be marked with a personalized label indicating the field tendency.

For example, if the user’s personalized vocabulary contains personalized words such as “hot pot”, “western food”, “net celebrity restaurant”, etc., since these personalized words correspond to “food”, the user’s personalization can be determined The personalized labels corresponding to the thesaurus are "food", "food", etc. For another example, if the user’s personalized vocabulary contains personalized words such as "blockchain", "face recognition", and "cloud computing", since the corresponding field of these personalized words is "IT technology", it can be determined The personalized tags corresponding to the user's personalized vocabulary are "IT Engineer", "IT Technology", etc.

In the embodiment of the present application, personalized vocabulary corresponding to different fields can be preset, and the personalized vocabulary corresponding to different fields in the user's personalized vocabulary can be counted, and the personalized label corresponding to the personalized vocabulary can be determined. In practical applications, if a user's personalized vocabulary contains multiple personalized vocabularies corresponding to different fields, then the user's personalized vocabulary can be labeled with multiple different personalized tags. For example, the personalized vocabulary corresponding to different fields in a user’s personalized vocabulary is counted, and the user’s personalized vocabulary includes the fields of "finance", "judicial", and "medical health". The personalized vocabulary of the user can be labeled with multiple personalized labels such as "financial", "judicial", and "medical health" for the user's personalized vocabulary.

Step S22 determines that the similarity between the personalized vocabularies of the at least two users meets a preset condition, which may specifically include: if the personalized vocabularies of the at least two users have the same personalized tag, determining that all the personalized vocabularies of the at least two users have the same personalized tag. It is stated that the similarity between the personalized word databases of at least two users meets the preset condition.

If two users have the same personalized tag, it means that the two users have the same field tendency. For example, the two users may be engaged in the same profession or have the same hobbies. Therefore, if it is determined that the personalized vocabularies of at least two users have the same personalized label, it can be determined that the similarity between the personalized vocabularies of the at least two users meets the preset condition, and the at least two user The personalized vocabulary of each user is merged.

In an optional embodiment of the present application, the merging of the personalized vocabulary of the at least two users in step S22 to obtain the merged personalized vocabulary may specifically include:

Merging all the words in the personalized vocabulary of the at least two users to obtain a merged personalized vocabulary; or

The vocabulary that meets the matching condition in the personalized vocabulary of the at least two users is merged to obtain a merged personalized vocabulary.

In the embodiment of the present application, after determining the personalized vocabulary of at least two users whose similarity meets a preset condition, the personalized vocabulary of the at least two users may be merged, and two solutions may be adopted for the merging.

Specifically, all vocabularies in the personalized vocabulary to be merged can be combined. For example, for user U and user V, let N(u) denote user U's personalized vocabulary, which includes user U's personalized vocabulary; N(v) represents user V's personalized vocabulary, which includes user V's personalized vocabulary. If the similarity between N(u) and N(v) meets the preset condition, all words in N(u) and all words in N(v) can be merged. Of course, de-duplication can be performed after merging. After processing, the merged personalized vocabulary is obtained.

Or, it is also possible to merge only the words that meet the matching conditions among the personalized words to be merged. Wherein, the matching conditions may include: belonging to the same field, having the same personalized label, and so on. It can be understood that those skilled in the art can set the matching conditions according to actual needs. For example, in the above example, after determining that the similarity between N(u) and N(v) meets a preset condition, the words that meet the matching condition can be further determined in N(u) and N(v), such as determining N( U) and N(v) belong to the same field vocabulary, and then only the words belonging to the same field in N(u) and N(v) are merged. Of course, after the merge, you can perform de-duplication and other processing to obtain the merged Personalized thesaurus.

In an optional embodiment of the present application, the determining the decoding path weight corresponding to the voice information according to the personalized vocabulary may specifically include:

Step S51, matching the word sequences corresponding to each decoding path of the voice information with the personalized word database;

Step S52: Determine the weight of the decoding path corresponding to the matched vocabulary according to the word frequency information corresponding to the matched vocabulary of the word sequence in the personalized vocabulary.

In the embodiment of the present application, the voice segment corresponding to the voice information may be scored by the acoustic model and the language model by the frame by frame through a preset decoding network, so as to obtain a voice recognition result. Among them, the basic structure of the decoding network is a directed graph composed of nodes and arcs. Each arc can save an entry and its acoustic model information and/or language model information. In practice, acoustic model information is generally expressed as acoustic model scores, and language model information is generally expressed as language model scores. Speech recognition is the process of finding an optimal path on this directed graph based on the input speech data. Among them, the acoustic model is used to calculate the probability from speech to syllable, and the language model is used to calculate the probability from syllable to word. Both the acoustic model score and the language model score can be obtained through pre-model training.

In the embodiment of the present application, the electronic device on which the voice recognition method runs can obtain the voice recognition result according to the final score result after performing acoustic model scoring and language model checking on the voice segment frame by frame. Specifically, after the acoustic model scoring and the language model checking are performed, the scores of all nodes on each decoding path in the decoding network can be added to form the score of the decoding path. Then, the one or more decoding paths with the highest scores are backtracked, and the word sequence corresponding to the corresponding decoding path can be obtained. In this way, the phrase or sentence composed of the obtained word sequence can be used as the result of speech recognition.

Step 104, determining the voice recognition result corresponding to the voice information according to the decoding path weight, may specifically include: determining the voice information corresponding to the weight of the decoding path corresponding to the matching vocabulary in the word sequence of each decoding path The result of speech recognition.

In this embodiment of the application, in the foregoing decoding process, the word sequence corresponding to each decoding path is matched with the user's personalized vocabulary, if one or more of the word sequences have matching words in the personalized vocabulary , The weight of the decoding path corresponding to the matched vocabulary is determined according to the word frequency information corresponding to the matched vocabulary. For example, for the user U, who often inputs "Li Bing" through the input method, his personalized vocabulary may include the personalized vocabulary "Li Bing" and the word frequency corresponding to the personalized vocabulary. When the user U inputs the voice information "libing" through the microphone, in the process of decoding the voice information, the voice decoder will match the word sequence corresponding to each decoding path with the user's personalized vocabulary. Assuming that the word sequences "Li Bing" and "Li Bing" both have matching words in the user U's personalized thesaurus, obtain the word frequency of the matching words "Li Bing" and "Li Bing" in the personalized thesaurus, and assume the word frequency of "Li Bing" If it is greater than "Li Bing", it can be determined that the weight corresponding to the decoding path "Li Bing" is greater than the weight of the decoding path "Li Bing". Therefore, the final recognition result can be obtained as "Li Bing", which conforms to the input habit of the user U, and therefore the recognition result is more accurate.

In an optional embodiment of the present application, after determining the voice recognition result corresponding to the voice information according to the decoding path weight in step 104, the method may further include:

Step S61: Acquire context information of the matched vocabulary in the speech recognition result;

Step S62: Perform error correction on the speech recognition result according to the context information, and obtain an error-corrected speech recognition result.

In practical applications, although the personalized vocabulary in the user's personalized vocabulary can reflect the user's input habits, only determining the weight of the decoding path based on the personalized vocabulary may cause some false incentives. For example, if the voice information input by a user is "lijiewansui" (Long live understanding), since the user's personalized vocabulary contains the personalized vocabulary "Sister Li" with a high word frequency, it may be recognized as "Long live Lijie" result. However, according to the context information, it can be known that the result that the user wants to output may be "Long Live Understanding", which leads to an error in the recognition result.

Therefore, after determining the voice recognition result corresponding to the voice information in the embodiment of the present application, the context information of the matched vocabulary in the voice recognition result can also be obtained, and based on the context information, the voice recognition result Perform error correction and obtain the speech recognition result after error correction.

For example, in the above example, the voice recognition result can be determined as "Long live Sister Li" according to the user's personalized vocabulary. At this time, the context information of the matching vocabulary "Sister Li" in the voice recognition result can be obtained. If the voice recognition result of the obtained voice information is "Long live Sister Li, Silian is not guilty", according to the following information "Silian is not guilty", it can be determined that the voice recognition result is wrong, and the voice recognition result can be corrected. Specifically, the word sequence with the second highest weight in the decoding path corresponding to the voice information can be obtained. If it is "understanding" and "Long live understanding" is consistent with the current context information, you can use "understand" to correct the error of "Sister Li" , Get the speech recognition result "Long Live Understanding" after error correction, and output the speech recognition result after error correction.

In summary, after receiving the voice information input by the user, the embodiment of the present application can obtain the user's personalized vocabulary, and determine the decoding path weight corresponding to the voice information according to the personalized vocabulary, and The decoding path weight determines the voice recognition result corresponding to the voice information. Since the personalized thesaurus is established based on the historical input content generated during the user’s use of the input method, the personalized dictionary conforms to the user’s input habits, and can be based on the user’s input in the process of decoding the voice information input by the user. The personalized vocabulary of ”performs real-time weighting on the speech decoding path, so that the final recognition result tends to the user’s input habits, thereby improving the accuracy of speech recognition.

It should be noted that for the method embodiments, for the sake of simple description, they are all expressed as a series of action combinations, but those skilled in the art should know that the embodiments of the present application are not limited by the described sequence of actions, because According to the embodiments of the present application, some steps may be performed in other order or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of this application.

Device embodiment

Referring to FIG. 2, a structural block diagram of an embodiment of a speech recognition device of the present application is shown, and the device may specifically include:

The voice receiving module 201 is used to receive voice information input by the user;

The lexicon acquisition module 202 is configured to acquire a personalized lexicon of the user, and the personalized lexicon is established based on the historical input content generated during the process of using the input method by the user;

The weight determination module 203 is configured to determine the decoding path weight corresponding to the voice information according to the personalized vocabulary;

The result determining module 204 is configured to determine the voice recognition result corresponding to the voice information according to the decoding path weight.

Optionally, the device may further include:

A data collection module, which is used to collect historical input content generated by the user in the process of using the input method;

The data processing module is used to preprocess the historical input content to obtain the preprocessed historical input content;

The data filtering module is used to filter the non-personalized vocabulary in the pre-processed historical input content to obtain the personalized vocabulary;

The vocabulary establishment module is used to establish the personalized vocabulary of the user according to the personalized vocabulary.

Optionally, the device may further include:

The similarity calculation module is used to calculate the similarity between the personalized vocabulary of at least two users;

The database merging module is configured to merge the personalized lexicons of the at least two users if it is determined that the similarity between the personalized lexicons of the at least two users meets a preset condition to obtain the merged personality The lexicon

The vocabulary acquisition module is specifically configured to obtain the merged personalized vocabulary.

Optionally, the similarity calculation module may specifically include:

The distance calculation submodule is configured to calculate the cosine distance between the personalized word databases of the at least two users according to the common vocabulary included in the personalized word databases of the at least two users;

A similarity calculation sub-module, configured to calculate the similarity between the personalized vocabularies of the at least two users according to the cosine distance;

The database merging module may specifically include:

The first determining sub-module is configured to determine that if the cosine distance between the personalized vocabularies of the at least two users is less than a preset threshold, determine that the similarity between the personalized vocabularies of the at least two users meets the preset threshold Set conditions.

Optionally, the device may further include:

The vocabulary classification module is used to classify the personalized vocabulary according to the field corresponding to the personalized vocabulary in the personalized vocabulary;

The label establishment module is configured to perform statistics on the personalized vocabulary corresponding to different fields in the personalized dictionary, and determine the personalized label corresponding to the user;

Optionally, the database merging module may specifically include:

The first merging sub-module is used to merge all the words in the personalized vocabulary of the at least two users to obtain a merged personalized vocabulary; or

The first merging sub-module is used to merge the vocabulary that meets the matching condition in the personalized vocabulary of the at least two users to obtain a merged personalized vocabulary.

The database merging module may specifically include:

The second determining sub-module is configured to determine that the similarity between the personalized vocabularies of the at least two users meets a preset condition if the personalized vocabularies of the at least two users have the same personalized tag.

Optionally, the weight determination module may specifically include:

The lexicon matching sub-module is used to match the word sequence corresponding to each decoding path of the voice information with the personalized lexicon;

A weight determination submodule, configured to determine the weight of the decoding path corresponding to the matched vocabulary according to the word frequency information corresponding to the matched vocabulary of the word sequence in the personalized vocabulary;

The result determination module is specifically configured to determine the speech recognition result corresponding to the speech information according to the weight of the decoding path corresponding to the matching vocabulary in the word sequence of each decoding path.

Optionally, the device may further include:

A context acquisition module, configured to acquire context information of the matched vocabulary in the speech recognition result;

The result error correction module is configured to perform error correction on the speech recognition result according to the context information to obtain the speech recognition result after the error correction.

As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and for related parts, please refer to the part of the description of the method embodiment.

The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the various embodiments can be referred to each other.

Regarding the device in the foregoing embodiment, the specific manner in which each module performs operation has been described in detail in the embodiment of the method, and detailed description will not be given here.

An embodiment of the application provides a device for speech recognition, including a memory and one or more programs, wherein one or more programs are stored in the memory and configured to be executed by one or more processors The one or more programs include instructions for performing the following operations: receiving voice information input by a user; acquiring a personalized vocabulary of the user, and the personalized vocabulary is generated according to the process of using the input method by the user The historical input content is established; according to the personalized vocabulary, the decoding path weight corresponding to the speech information is determined; according to the decoding path weight, the speech recognition result corresponding to the speech information is determined.

Fig. 3 is a block diagram showing a device 800 for speech recognition according to an exemplary embodiment. For example, the device 800 may be a mobile phone, a computer, a digital broadcasting terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, etc.

3, the device 800 may include one or more of the following components: a processing component 802, a memory 804, a power supply component 806, a multimedia component 808, an audio component 810, an input/output (I/O) interface 812, a sensor component 814, And the communication component 816.

The processing component 802 generally controls the overall operations of the device 800, such as operations associated with display, telephone calls, data communications, camera operations, and recording operations. The processing element 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the foregoing method. In addition, the processing component 802 may include one or more modules to facilitate the interaction between the processing component 802 and other components. For example, the processing component 802 may include a multimedia module to facilitate the interaction between the multimedia component 808 and the processing component 802.

The memory 804 is configured to store various types of data to support operations in the device 800. Examples of such data include instructions for any application or method operating on the device 800, contact data, phone book data, messages, pictures, videos, etc. The memory 804 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable and Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), Magnetic Memory, Flash Memory, Magnetic Disk or Optical Disk.

The power supply component 806 provides power to various components of the device 800. The power supply component 806 may include a power management system, one or more power supplies, and other components associated with the generation, management, and distribution of power for the device 800.

The multimedia component 808 includes a screen that provides an output interface between the device 800 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen can be implemented as a touch screen to receive input signals from users. The touch panel includes one or more touch sensors to sense touch, sliding, and gestures on the touch panel. The touch sensor may not only sense the boundary of a touch or slide action, but also detect the duration and pressure related to the touch or slide operation. In some embodiments, the multimedia component 808 includes a front camera and/or a rear camera. When the device 800 is in an operation mode, such as a shooting mode or a video mode, the front camera and/or the rear camera can receive external multimedia data. Each front camera and rear camera can be a fixed optical lens system or have focal length and optical zoom capabilities.

The audio component 810 is configured to output and/or input audio signals. For example, the audio component 810 includes a microphone (MIC), and when the device 800 is in an operation mode, such as a call mode, a recording mode, and a voice information processing mode, the microphone is configured to receive external audio signals. The received audio signal may be further stored in the memory 804 or transmitted via the communication component 816. In some embodiments, the audio component 810 further includes a speaker for outputting audio signals.

The I/O interface 812 provides an interface between the processing component 802 and a peripheral interface module. The above-mentioned peripheral interface module may be a keyboard, a click wheel, a button, and the like. These buttons may include, but are not limited to: home button, volume button, start button, and lock button.

The sensor component 814 includes one or more sensors for providing the device 800 with various aspects of status assessment. For example, the sensor component 814 can detect the on/off status of the device 800 and the relative positioning of components. For example, the component is the display and the keypad of the device 800. The sensor component 814 can also detect the position change of the device 800 or a component of the device 800. , The presence or absence of contact between the user and the device 800, the orientation or acceleration/deceleration of the device 800, and the temperature change of the device 800. The sensor component 814 may include a proximity sensor configured to detect the presence of nearby objects when there is no physical contact. The sensor component 814 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor component 814 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

The communication component 816 is configured to facilitate wired or wireless communication between the device 800 and other devices. The device 800 can access a wireless network based on a communication standard, such as WiFi, 2G, or 3G, or a combination thereof. In an exemplary embodiment, the communication component 816 receives a broadcast signal or broadcast related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 816 further includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency information processing (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.

In an exemplary embodiment, the apparatus 800 may be implemented by one or more application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing equipment (DSPD), programmable logic devices (PLD), field programmable A gate array (FPGA), controller, microcontroller, microprocessor, or other electronic components are implemented to implement the above methods.

In an exemplary embodiment, there is also provided a non-transitory computer-readable storage medium including instructions, such as the memory 804 including instructions, which may be executed by the processor 820 of the device 800 to complete the foregoing method. For example, the non-transitory computer-readable storage medium may be ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, optical data storage device, etc.

Fig. 4 is a schematic diagram of the structure of a server in some embodiments of the present application. The server 1900 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPU) 1922 (for example, one or more processors) and a memory 1932, one or one The above storage medium 1930 (for example, one or one storage device with a large amount of storage) for storing application programs 1942 or data 1944. Among them, the memory 1932 and the storage medium 1930 may be short-term storage or permanent storage. The program stored in the storage medium 1930 may include one or more modules (not shown in the figure), and each module may include a series of command operations on the server. Furthermore, the central processing unit 1922 may be configured to communicate with the storage medium 1930, and execute a series of instruction operations in the storage medium 1930 on the server 1900.

The server 1900 may also include one or more power supplies 1926, one or more wired or wireless network interfaces 1950, one or more input and output interfaces 1958, one or more keyboards 1956, and/or, one or more operating systems 1941 , Such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM and so on.

A non-transitory computer-readable storage medium, when instructions in the storage medium are executed by a processor of a device (server or terminal), the device can execute the voice recognition method shown in FIG. 1.

A non-transitory computer-readable storage medium. When the instructions in the storage medium are executed by the processor of the device (server or terminal), the device can execute a voice recognition method, the method comprising: receiving user input Acquire the user’s personalized vocabulary, the personalized vocabulary is established based on the historical input content generated during the user’s use of the input method; determine the voice according to the personalized vocabulary The decoding path weight corresponding to the information; and the speech recognition result corresponding to the speech information is determined according to the decoding path weight.

The embodiments of this application are described with reference to the flowcharts and/or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of this application. It should be understood that each process and/or block in the flowchart and/or block diagram, and the combination of processes and/or blocks in the flowchart and/or block diagram can be realized by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing terminal equipment to generate a machine, so that instructions executed by the processor of the computer or other programmable data processing terminal equipment A device for realizing the functions specified in a flow or multiple flows in the flowchart and/or a block or multiple blocks in the block diagram is generated.

These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing terminal equipment to work in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including the instruction device. The instruction device implements the functions specified in one process or multiple processes in the flowchart and/or one block or multiple blocks in the block diagram.

These computer program instructions can also be loaded on a computer or other programmable data processing terminal equipment, so that a series of operation steps are executed on the computer or other programmable terminal equipment to produce computer-implemented processing, so that the computer or other programmable terminal equipment The instructions executed above provide steps for implementing functions specified in a flow or multiple flows in the flowchart and/or a block or multiple blocks in the block diagram.

After considering the specification and practicing the invention disclosed herein, those skilled in the art will easily think of other embodiments of the present application. This application is intended to cover any variations, uses, or adaptive changes of this application. These variations, uses, or adaptive changes follow the general principles of this application and include common knowledge or customary technical means in the technical field that are not disclosed in this disclosure. . The description and the embodiments are only regarded as exemplary, and the true scope and spirit of the application are pointed out by the following claims.

It should be understood that the present application is not limited to the precise structure that has been described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the application is only limited by the appended claims.

The above descriptions are only preferred embodiments of this application, and are not intended to limit this application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this application shall be included in the protection of this application. Within range.

The above provides a detailed introduction to a voice recognition method, a voice recognition device, and a device for voice recognition provided by the present application. In this article, specific examples are used to illustrate the principles and implementations of the present application. The description of the above embodiments is only used to help understand the method and core idea of the application; at the same time, for those of ordinary skill in the art, according to the idea of the application, there will be changes in the specific implementation and the scope of application. In summary, the content of this specification should not be construed as a limitation on this application.

Claims

A speech recognition method, characterized in that the method includes:

Receive voice information input by the user;

Acquiring a personalized vocabulary of the user, the personalized vocabulary being established based on historical input content generated during the user's use of the input method;

Determining the decoding path weight corresponding to the voice information according to the personalized vocabulary;

According to the decoding path weight, the speech recognition result corresponding to the speech information is determined.
The method according to claim 1, characterized in that, before said obtaining the user's personalized vocabulary, the method further comprises:

Collect historical input content generated by the user in the process of using the input method;

Preprocessing the historical input content to obtain preprocessed historical input content;

Filtering non-personalized words in the preprocessed historical input content to obtain personalized words;

According to the personalized vocabulary, a personalized vocabulary of the user is established.
The method according to claim 2, characterized in that, after the establishment of the user's personalized vocabulary, the method further comprises:

Calculate the similarity between the personalized vocabulary of at least two users;

If it is determined that the similarity between the personalized word databases of the at least two users meets a preset condition, merge the personalized word databases of the at least two users to obtain a merged personalized word database;

The obtaining the personalized vocabulary of the user includes:

Obtain the merged personalized vocabulary.
The method according to claim 3, wherein the calculating the similarity between the personalized vocabulary of at least two users comprises:

Calculating the cosine distance between the personalized word databases of the at least two users according to the common vocabulary included in the personalized word databases of the at least two users;

Calculating the similarity between the personalized vocabulary of the at least two users according to the cosine distance;

The determining that the similarity between the personalized word databases of the at least two users meets a preset condition includes:

If the cosine distance between the personalized vocabularies of the at least two users is less than the preset threshold, it is determined that the similarity between the personalized vocabularies of the at least two users meets the preset condition.
The method according to claim 3, characterized in that, after the establishment of the user's personalized vocabulary, the method further comprises:

Classify the personalized vocabulary according to the field corresponding to the personalized vocabulary in the personalized vocabulary;

Perform statistics on the personalized vocabulary corresponding to different fields in the personalized dictionary, and determine the personalized tag corresponding to the user;

The determining that the similarity between the personalized word databases of the at least two users meets a preset condition includes:

If the personalized vocabularies of the at least two users have the same personalized label, it is determined that the similarity between the personalized vocabularies of the at least two users meets a preset condition.
The method according to claim 3, wherein the merging the personalized vocabulary of the at least two users to obtain the merged personalized vocabulary comprises:

Merging all the words in the personalized vocabulary of the at least two users to obtain a merged personalized vocabulary; or

The vocabulary that meets the matching condition in the personalized vocabulary of the at least two users is merged to obtain a merged personalized vocabulary.
The method according to claim 1, wherein the determining the decoding path weight corresponding to the voice information according to the personalized vocabulary comprises:

Matching the word sequences corresponding to each decoding path of the voice information with the personalized word library;

Determine the weight of the decoding path corresponding to the matched vocabulary according to the word frequency information corresponding to the matched vocabulary of the word sequence in the personalized vocabulary;

The determining the voice recognition result corresponding to the voice information according to the decoding path weight includes:

The speech recognition result corresponding to the speech information is determined according to the weight of the matching vocabulary corresponding to the decoding path in the word sequence of each decoding path.
The method according to claim 7, wherein, after the speech recognition result corresponding to the speech information is determined according to the decoding path weight, the method further comprises:

Acquiring context information of the matched vocabulary in the speech recognition result;

According to the context information, error correction is performed on the speech recognition result to obtain an error-corrected speech recognition result.
A speech recognition device, characterized in that the device includes:

The voice receiving module is used to receive the voice information input by the user;

Thesaurus acquisition module, configured to acquire the user's personalized thesaurus, the personalized thesaurus is established based on the historical input content generated during the user's use of the input method;

A weight determination module, configured to determine the decoding path weight corresponding to the voice information according to the personalized vocabulary;

The result determining module is configured to determine the voice recognition result corresponding to the voice information according to the decoding path weight.
The device according to claim 9, wherein the device further comprises:

A data collection module, which is used to collect historical input content generated by the user in the process of using the input method;

The data processing module is used to preprocess the historical input content to obtain the preprocessed historical input content;

The data filtering module is used to filter the non-personalized vocabulary in the pre-processed historical input content to obtain the personalized vocabulary;

The vocabulary establishment module is used to establish the personalized vocabulary of the user according to the personalized vocabulary.
The device according to claim 10, wherein the device further comprises:

The similarity calculation module is used to calculate the similarity between the personalized vocabulary of at least two users;

The database merging module is configured to merge the personalized lexicons of the at least two users if it is determined that the similarity between the personalized lexicons of the at least two users meets a preset condition to obtain the merged personality The lexicon

The vocabulary acquisition module is specifically configured to obtain the merged personalized vocabulary.
The device according to claim 11, wherein the similarity calculation module comprises:

The distance calculation submodule is configured to calculate the cosine distance between the personalized word databases of the at least two users according to the common vocabulary included in the personalized word databases of the at least two users;

The similarity calculation sub-module is used to calculate the similarity between the personalized vocabulary of the at least two users according to the cosine distance;

The database merging module includes:

The first determining sub-module is configured to determine that if the cosine distance between the personalized vocabularies of the at least two users is less than a preset threshold, determine that the similarity between the personalized vocabularies of the at least two users meets the preset threshold Set conditions.
The device according to claim 11, wherein the device further comprises:

The vocabulary classification module is used to classify the personalized vocabulary according to the field corresponding to the personalized vocabulary in the personalized vocabulary;

The label establishment module is configured to perform statistics on the personalized vocabulary corresponding to different fields in the personalized dictionary, and determine the personalized label corresponding to the user;

The database merging module includes:

The second determining sub-module is configured to determine that the similarity between the personalized vocabularies of the at least two users meets a preset condition if the personalized vocabularies of the at least two users have the same personalized tag.
The device according to claim 11, wherein the database merging module comprises:

The first merging sub-module is used to merge all the words in the personalized vocabulary of the at least two users to obtain a merged personalized vocabulary; or

The first merging sub-module is used to merge the vocabulary that meets the matching condition in the personalized vocabulary of the at least two users to obtain a merged personalized vocabulary.
The device according to claim 9, wherein the weight determination module comprises:

The lexicon matching sub-module is used to match the word sequence corresponding to each decoding path of the voice information with the personalized lexicon;

A weight determination submodule, configured to determine the weight of the decoding path corresponding to the matched vocabulary according to the word frequency information corresponding to the matched vocabulary of the word sequence in the personalized vocabulary;

The result determination module is specifically configured to determine the speech recognition result corresponding to the speech information according to the weight of the decoding path corresponding to the matching vocabulary in the word sequence of each decoding path.
The device according to claim 15, wherein the device further comprises:

A context acquisition module, configured to acquire context information of the matched vocabulary in the speech recognition result;

The result error correction module is configured to perform error correction on the speech recognition result according to the context information to obtain the speech recognition result after the error correction.
A device for speech recognition, characterized by comprising a memory and one or more programs, wherein one or more programs are stored in the memory and configured to be executed by one or more processors One or more programs contain instructions for the following operations:

Receive voice information input by the user;

Acquiring a personalized vocabulary of the user, the personalized vocabulary being established based on historical input content generated during the user's use of the input method;

Determining the decoding path weight corresponding to the voice information according to the personalized vocabulary;

According to the decoding path weight, the speech recognition result corresponding to the speech information is determined.
The device of claim 17, wherein the device is further configured to execute the one or more programs by one or more processors including instructions for performing the following operations:

Collect historical input content generated by the user in the process of using the input method;

Preprocessing the historical input content to obtain preprocessed historical input content;

Filtering non-personalized words in the preprocessed historical input content to obtain personalized words;

According to the personalized vocabulary, a personalized vocabulary of the user is established.
The device of claim 18, wherein the device is further configured to execute the one or more programs by one or more processors including instructions for performing the following operations:

Calculate the similarity between the personalized vocabulary of at least two users;

If it is determined that the similarity between the personalized word databases of the at least two users meets a preset condition, merge the personalized word databases of the at least two users to obtain a merged personalized word database;

The obtaining the personalized vocabulary of the user includes:

Obtain the merged personalized vocabulary.
The device according to claim 19, wherein said calculating the similarity between the personalized word databases of at least two users comprises:

Calculating the cosine distance between the personalized word databases of the at least two users according to the common vocabulary included in the personalized word databases of the at least two users;

Calculating the similarity between the personalized vocabulary of the at least two users according to the cosine distance;

The determining that the similarity between the personalized word databases of the at least two users meets a preset condition includes:

If the cosine distance between the personalized vocabularies of the at least two users is less than the preset threshold, it is determined that the similarity between the personalized vocabularies of the at least two users meets the preset condition.
The apparatus of claim 19, wherein the apparatus is further configured to be executed by one or more processors to execute the one or more programs comprising instructions for performing the following operations:

Classify the personalized vocabulary according to the field corresponding to the personalized vocabulary in the personalized vocabulary;

Perform statistics on the personalized vocabulary corresponding to different fields in the personalized dictionary, and determine the personalized tag corresponding to the user;

The determining that the similarity between the personalized word databases of the at least two users meets a preset condition includes:

If the personalized vocabularies of the at least two users have the same personalized label, it is determined that the similarity between the personalized vocabularies of the at least two users meets a preset condition.
The device according to claim 19, wherein said merging the personalized word databases of the at least two users to obtain a merged personalized word database comprises:

Merging all the words in the personalized vocabulary of the at least two users to obtain a merged personalized vocabulary; or

The vocabulary that meets the matching condition in the personalized vocabulary of the at least two users is merged to obtain a merged personalized vocabulary.
The device according to claim 17, wherein the determining the decoding path weight corresponding to the voice information according to the personalized vocabulary comprises:

Matching the word sequences corresponding to each decoding path of the voice information with the personalized word library;

Determine the weight of the decoding path corresponding to the matched vocabulary according to the word frequency information corresponding to the matched vocabulary of the word sequence in the personalized vocabulary;

The determining the voice recognition result corresponding to the voice information according to the decoding path weight includes:

The speech recognition result corresponding to the speech information is determined according to the weight of the matching vocabulary corresponding to the decoding path in the word sequence of each decoding path.
The device of claim 23, wherein the device is further configured to execute the one or more programs by one or more processors including instructions for performing the following operations:

Acquiring context information of the matched vocabulary in the speech recognition result;

According to the context information, error correction is performed on the speech recognition result to obtain an error-corrected speech recognition result.
A machine-readable medium with instructions stored thereon, which when executed by one or more processors, causes a device to execute the voice recognition method according to one or more of claims 1 to 8.