CN110310631A - Speech recognition method, device, server and storage medium - Google Patents

Abstract

The embodiment of the invention discloses a voice recognition method, device, server and storage medium. The method includes: performing map information search on the current user's voice, and determining at least one matching candidate information; performing disambiguation processing on the at least one candidate information according to the current user's map domain characteristics, so as to determine the map of the current user's voice information identification results. The embodiment of the present invention disambiguates the candidate information obtained from the special search of the map domain information, which not only removes the possible interference of general domain knowledge on the map search, but also avoids misjudgments caused by ambiguity and accents, making the search The obtained map information recognition results are more in line with the user's habits and needs, which greatly improves the voice recognition accuracy of map voice search.

Description

Speech recognition method, device, server and storage medium

technical field

The embodiments of the present invention relate to the technical field of voice recognition, and in particular, to a voice recognition method, device, server and storage medium.

Background technique

Map voice search is an important function of maps at present. By using voice input and interaction instead of manual input to search and query map information, it can greatly facilitate user input and is more suitable for map driving scenarios.

At present, a third-party mature voice input method interface can be called to provide voice recognition support for map voice search. Among them, the speech recognition model used is usually obtained based on large-scale Internet data training, and has the versatility of speech recognition. Alternatively, retrain a version of the model specifically for map speech recognition using a map corpus with experience in the map domain.

However, general-purpose speech recognition models lack map domain experience and are not suitable for map speech search scenarios. When the general-purpose speech recognition model is used in the map scene, most of the names of map information in the map are low-frequency, rare, or even non-existent unfamiliar words in the general-purpose speech recognition model, and the general-purpose speech is used directly The recognition model accuracy will be very poor. In addition, the cost of retraining a dedicated speech recognition model based on map corpus is high, and it is difficult to overcome the problem of map information recognition errors caused by noise and regional differences, and the accuracy of map voice search is low.

Contents of the invention

Embodiments of the present invention provide a voice recognition method, device, server and storage medium, which can improve the voice recognition accuracy rate of map voice search.

In a first aspect, an embodiment of the present invention provides a speech recognition method, including:

Carrying out a map information search on the current user's voice, and determining at least one matching candidate information;

Disambiguation processing is performed on the at least one piece of candidate information according to the characteristics of the current user's map domain, so as to determine the map information recognition result of the current user's voice.

In a second aspect, an embodiment of the present invention provides a speech recognition device, including:

Candidate information determination module, used to search the current user voice for map information, and determine at least one matching candidate information;

The voice recognition disambiguation module is configured to perform disambiguation processing on the at least one candidate information according to the current user's map domain characteristics, so as to determine the map information recognition result of the current user's voice.

In a third aspect, an embodiment of the present invention provides a server, including:

one or more processors;

memory for storing one or more programs;

When the one or more programs are executed by the one or more processors, the one or more processors implement the speech recognition method in any embodiment of the present invention.

In a fourth aspect, an embodiment 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 speech recognition method described in any embodiment of the present invention is implemented.

In the embodiment of the present invention, a plurality of candidate information matching the user's voice is determined by performing a special map information search on the user's voice, and disambiguation processing is performed on the multiple candidate information according to the current user's map domain characteristics, and the multiple candidate information The candidate information that best matches the user is screened out as the map information recognition result. The embodiment of the present invention disambiguates the candidate information obtained by the special search of the map domain information, which not only removes the possible interference of the general domain knowledge on the map search, but also avoids misjudgment caused by ambiguity and accent, making the search The obtained map information recognition results are more in line with the user's habits and needs, which greatly improves the voice recognition accuracy of map voice search.

Description of drawings

FIG. 1 is a flow chart of a speech recognition method provided by Embodiment 1 of the present invention;

FIG. 2 is a flow chart of a speech recognition method provided by Embodiment 2 of the present invention;

FIG. 3 is an overall architecture diagram of speech recognition provided by Embodiment 2 of the present invention;

FIG. 4 is a schematic structural diagram of a voice recognition device provided in Embodiment 3 of the present invention;

FIG. 5 is a schematic structural diagram of a server provided by Embodiment 4 of the present invention.

Detailed ways

The embodiments of the present invention will be further described in detail below in conjunction with the drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the embodiments of the present invention, rather than to limit the present invention. In addition, it should be noted that, for the convenience of description, the drawings only show some but not all structures related to the embodiments of the present invention.

In addition, it should be noted that, for the convenience of description, only parts relevant to the present application are shown in the drawings but not all content. Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe various operations (or steps) as sequential processing, many of the operations may be performed in parallel, concurrently, or simultaneously. In addition, the order of operations can be rearranged. The process may be terminated when its operations are complete, but may also have additional steps not included in the figure. The processing may correspond to a method, function, procedure, subroutine, subroutine, or the like.

Embodiment one

Fig. 1 is a flow chart of a voice recognition method provided by Embodiment 1 of the present invention. This embodiment is applicable to the situation of searching for map information based on user voice, and the method can be executed by a voice recognition device, which can adopt It is realized by means of software and/or hardware, and is preferably configured on a server. The method specifically includes the following:

S110. Search the current user's voice for map information, and determine at least one matching candidate information.

In specific embodiments of the present invention, map products provide a search function for map information, especially a voice search function for driving and navigation. The current user voice refers to the voice search request input by the user to the map product for map information query when the user uses the map product. The user voice may include information about at least one location to be searched, restrictions on the location to be searched, etc. For the voice search request submitted by the user, the map product searches for relevant POI (Point of Interest) data and returns it to the user. Each piece of POI data in the map can include various information such as name, category, latitude and longitude, and importance. Generally, map search can include precise search and fuzzy search, wherein, precise search refers to that in map search, the voice search request submitted by the user is a search for a specific POI data point, and fuzzy search can be based on the sug engine, according to the voice search request A broad search for partial information, similarities, or corrected speech in .

In this embodiment, the speech recognition model is used to recognize the user's speech, and the speech recognition model may generally include an acoustic sub-model and a language sub-model. Among them, in order to improve the accuracy of the map information search, the general corpus and the map corpus are used to train the speech recognition model at the same time. In order to reduce the cost of speech recognition model training in this implementation, on the basis of the traditional general-purpose speech recognition model based on general-purpose corpus training, since a large part of the POI data in the map belongs to low-frequency and rare unfamiliar words, it can be directly The language sub-model in the general speech recognition model is trained twice by using the map corpus to strengthen the recognition ability of the speech recognition model for the map corpus.

In this embodiment, it is different from the traditional map search method in which the user's voice is recognized as text and the search is performed based on the text. In the new scheme, speech recognition does not directly output text, but instead outputs phoneme representations composed of multiple phonemes as fuzzy sounds, so that map information can be searched based on fuzzy sounds. Among them, fuzzy sounds can refer to pinyin that is not strictly required, that is, fuzzy sounds are similar to pinyin in form, but there may be differences in fuzzy sounds that do not conform to the spelling rules of pinyin, and fuzzy sounds still retain sound-related features as a whole . Therefore, a search is performed based on the fuzzy sound, and at least one candidate information matching the pronunciation of the fuzzy sound is obtained. The candidate information may be a candidate recognition text that is the same as or similar to the user's pronunciation, or a candidate recognition text that is consistent with or associated with the user's accent.

Specifically, in the speech recognition model, usually only when the corresponding text information is recognized can it be regarded as a normal speech recognition process, and then the intermediate information of any link can be extracted from it. Therefore, in this embodiment, the recognition hit rate of the map-type information is improved through secondary training of the language sub-model in the general speech recognition model, so as to reduce the abnormal situation that the map-type information cannot be recognized. Correspondingly, when the speech recognition module trained with map corpus is used to recognize the user's speech to obtain the recognized text, that is, after a normal speech recognition, the phoneme representation of the user's speech based on the acoustic sub-model recognition can be obtained, and the phoneme representation of the user's speech is used as Fuzzy sounds, and do not need to consider the text recognition results or as little as possible. Then search based on fuzzy sounds. In the process of searching, the error correction and expansion of phoneme representation can be realized, and multiple different phoneme representations can be obtained as variant phoneme representations, and the search results match the original phoneme representation and the variant phoneme representation respectively. candidate information.

For example, assuming that the fuzzy sound is output as "cangshangcun" based on the user's speech recognition, and the map information search is performed based on the fuzzy sound, the candidate information that can be matched includes "Cangshang Village, Tongzhou District", "Cangshang Village, Tongzhou District" and "Tongzhou District Cangchang Village" and so on.

S120. Perform disambiguation processing on at least one piece of candidate information according to the map domain characteristics of the current user, so as to determine a map information recognition result of the current user's voice.

In the specific embodiment of the present invention, in view of the fact that voice recognition can only be searched based on the user's pronunciation, there may be search results in the voice search that are the same as the user's voice pronunciation but not related to the content, and there may also be erroneous search results caused by the user's accent deviation . Therefore, in this embodiment, after obtaining at least one candidate information that matches the user's voice, it is necessary to disambiguate the candidate information to filter out interference information in the candidate information and improve the accuracy of map information recognition results.

In this embodiment, the current user's map domain characteristics can be used to disambiguate the candidate information. Specifically, the current user's map domain feature includes at least one of the current map search scene feature, the current user behavior feature, and the map information search quality feature of the candidate text.

Among them, the current map search scene feature may include the scene feature where the user is in when the user is currently searching, such as the user's current location, and then it may be based on the spatial position relationship between the user's current location and the location represented by the candidate information, such as the spatial affiliation , spatial proximity and spatial distance, etc., to screen candidate information. Exemplarily, the city, administrative area, etc. where the user is currently located is determined according to the current GPS information of the user, and candidate information other than the city and administrative area may be filtered out. In addition, the current map search scene feature may also include the spatial location description in the user's voice, such as the spatial location description of POI area restrictions in the user's voice, and then filter out candidate information that does not meet the restriction according to the spatial location description. Exemplarily, the user voice includes restriction information "POI point B in city A", and candidate information includes "POI point B in city S", and then the candidate information is filtered out.

Secondly, the current user behavior characteristics can include the user's historical search behavior of map search. Since the user has a high probability of repeated search for the same POI, the repeated search probability of each candidate information can be determined according to the user's historical search behavior, and then filtered out Candidate information with lower probability is repeatedly searched. In addition, the current user behavior characteristics may also include the user's accent characteristics. Usually, according to regional differences, users may not distinguish front and back nasal sounds, such as an and ang, en and eng, in and ing, ian and iang, uan and uang, etc., and flat tongues and warped tongues, such as z and zh, c and ch, s and sh, etc., or for users in individual areas, users may have cases where f and h, r and l are not distinguished, etc. Therefore, the map voice search in this embodiment can determine the pronunciation habits of the user, such as the biased characteristics of the user's pronunciation, based on the user's historical search behavior, so as to retain and expand the phonemes that the user has a biased pronunciation, so as to avoid only being able to use the user's biased phoneme. Pronunciation leads to misidentification. Exemplarily, it is assumed that candidate information is screened by scoring. If it is determined that the current user's pronunciation preference for f and h is f, then when f phoneme candidate information is detected, f phoneme is expanded to h phoneme, and At the same time, the scoring of the candidate information of the h phoneme is set to be the same as that of the candidate information of the f phoneme, so as to include the possible correct candidate information as much as possible.

In addition, map corpus or map information may be pre-determined as a map information base, and information related to real POIs is stored in the map information base. Correspondingly, the map information search quality feature of the candidate text may include the similarity between the candidate information and the map information in the preset map information database, that is, by calculating the text similarity between the candidate information and the real POI, It can be further determined whether the candidate information belongs to a real POI, so as to avoid the interference of words with the same or similar pronunciation in common corpus on the map search. Correspondingly, if the candidate information is successfully matched with the map information in the preset map information database or the similarity is greater than a certain threshold, it indicates that the candidate information is a better map search result. Then filter out the candidate information with poor map search quality. In addition, the map information search quality feature of the candidate text may also include the historical search demand distribution of the majority of users in the map search for the location represented by the candidate information. For example, users have lower demand for map search in sparsely populated areas, but higher demand for map search in prosperous areas. Therefore, according to the historical search behavior of the majority of users, the search requirements of each POI can be analyzed in real time or periodically, so as to filter out candidate information with low search requirements.

It should be noted that this embodiment enumerates three map domain features respectively, and each map domain feature includes at least two specific cases. Wherein, the map domain features in this embodiment are not limited to the above examples, and any feature that can reasonably filter candidate information to eliminate ambiguity can be applied to this embodiment, and the division of situations in each map domain feature is not limited. Limited to the examples above.

In this embodiment, the degree of association between the candidate information and the current user can be determined according to one or more of the map domain features, and the ambiguity between the candidate information and the current user's voice can be determined according to the degree of association between the candidate information and the current user ambiguity information, so as to filter out the ambiguity information in the candidate information, so as to obtain the map information recognition result of the current user's voice.

Exemplarily, assuming that the candidate information is disambiguated according to the map information search quality of the candidate text, and the candidate information includes "Wang Jiawei" and "Wang Jiawei", then the candidate information "Wang Jiawei" can be determined by detecting the map information search quality is the name of the director in the general corpus, and the candidate information "Wang Jiawei" is the name of a restaurant, which belongs to the actual map POI, and the candidate information "Wang Jiawei" is ambiguous information, which should be filtered out to avoid interference with the recognition results.

In the technical solution of this embodiment, a plurality of candidate information matching the user's voice is determined by performing a special map information search on the user's voice, and disambiguation processing is performed on the multiple candidate information according to the current user's map domain characteristics, from multiple The candidate information that best matches the user is selected from the candidate information as the map information recognition result. The embodiment of the present invention disambiguates the candidate information obtained by the special search of the map domain information, which not only removes the possible interference of the general domain knowledge on the map search, but also avoids misjudgment caused by ambiguity and accent, making the search The obtained map information recognition results are more in line with the user's habits and needs, which greatly improves the voice recognition accuracy of map voice search.

Embodiment two

On the basis of the first embodiment above, this embodiment provides a preferred implementation of the voice recognition method, which can perform fuzzy search for map information based on the phoneme representation of the user's voice. Fig. 2 is a flow chart of a speech recognition method provided by Embodiment 2 of the present invention. As shown in Fig. 2, the method specifically includes the following:

S210. Perform acoustic feature recognition on the current user's voice to determine a phoneme representation of the current user's voice.

In a specific embodiment of the present invention, a speech recognition model is used to recognize user speech, and the speech recognition model may generally include an acoustic sub-model and a language sub-model. Among them, in order to improve the accuracy of the map information search, the general corpus and the map corpus are used to train the speech recognition model at the same time. That is, on the basis of the traditional general-purpose speech recognition model based on general-purpose corpus training, since a large part of the POI data in the map belongs to low-frequency and rare unfamiliar words, it is possible to directly use the map corpus to analyze the language subclasses in the general-purpose speech recognition model. The model undergoes secondary training to strengthen the recognition ability of the speech recognition model for map corpus.

In this embodiment, the speech recognition model trained on the map corpus is used to recognize the user's speech to obtain the recognized text. After it is determined to be normal speech recognition, the phoneme representation of the user's speech based on the acoustic sub-model for acoustic feature recognition can be obtained, and the user's speech The phonemes of are represented as fuzzy sounds. Among them, a phoneme is the smallest unit of speech divided according to the natural attributes of speech. From the perspective of acoustic properties, a phoneme is the smallest unit of speech divided from the perspective of sound quality; from the perspective of physiological properties, one pronunciation action forms a phoneme. For example, ma contains two pronunciation actions of m and a, which are two phonemes. The sounds produced by the same pronunciation action are the same phoneme, and the sounds produced by different pronunciation actions are different phonemes. For example, in ma-mi, the two m pronunciation actions are the same and are the same phoneme, while a and i are different phonemes because of different pronunciation actions. In phonetics, the basic unit of speech structure consisting of one or several phonemes is called a syllable. In Chinese, the sound of a Chinese character is usually a syllable, and the basic syllable of Mandarin is composed of one or more phonemes according to certain combination rules.

In this embodiment, the phonemes obtained by acoustic feature recognition through the acoustic sub-model can be represented as fuzzy sounds, and fuzzy sounds can refer to non-strictly required pinyin, that is, fuzzy sounds are similar in form to pinyin, but there may be Differences that do not conform to the spelling rules of Pinyin, and fuzzy sounds still retain sound-related features as a whole. Exemplarily, assuming that the user's voice is "Cangshangcun", based on the recognition of the user's actual voice, the phoneme representation can be obtained as "cangshangcun", "canshangcun", "changshangcun" or "cagshangcun".

S220. Search the map information according to the phoneme representation, and determine at least one piece of candidate information that matches the pronunciation of the phoneme representation.

In a specific embodiment of the present invention, the possible correct pronunciation representation is included in the phoneme representation as much as possible, so that the map information search based on the phoneme representation can obtain the candidate recognition text that is the same as or similar to the user's pronunciation, or it can be consistent with the user's accent or associated extended candidate recognition text.

Optionally, perform error correction and phoneme expansion on the phoneme representation, determine at least one variant phoneme representation; obtain at least one candidate information that matches the pronunciation of the phoneme representation and the at least one variant phoneme representation.

In this embodiment, the phoneme representation recognized by the acoustic sub-model can be used as a basic phoneme representation, and on the basis of the basic phoneme representation, the basic phoneme representation is corrected and extended to obtain at least one variant phoneme representation that is different from the basic phoneme representation . Wherein, the error correction of the basic phoneme representation refers to correcting the phonemes in the basic phoneme representation that do not conform to the pinyin rules. For example, in the above embodiment, the phoneme expression "cagshangcun" can be modified to a variant phoneme expression "cangshangcun". The extension of the basic phoneme representation refers to the associated expansion of the phonemes that may have pronunciation deviations in the basic phoneme representation, that is, the various possible deviation results of the phonemes that may have pronunciation deviations are associated to obtain different variant phoneme representations to avoid user accent problems Causes speech recognition root error. By obtaining the variant phoneme representation, possible correct pronunciation representations can be included as much as possible, so as to improve the accuracy of map voice search. Therefore, the map search is performed based on the phoneme representation and the variant phoneme representation respectively, and at least one candidate information matching the pronunciation of the phoneme representation and at least one variant phoneme representation is obtained through matching.

Exemplarily, a phoneme fuzzy matching table can be preset, and matching rules are defined in the phoneme fuzzy matching table, for example: z=zh, c=ch, s=sh, an=ang, en=eng, in=ing, ian =iang, uan=uang, iong=ing, f=h, r=l and l=n etc. For example, when the user's voice is "Hu Jian", its basic phoneme can be expressed as "hujian", and based on phoneme expansion, at least one variant phoneme can be obtained and expressed as "fujian". The expansion of the phoneme representation solves problems such as speech recognition or recognition errors caused by the user's slurred speech and inaccurate pronunciation, and further improves the accuracy of map information search in this embodiment.

Usually map search can include precise search and fuzzy search, wherein precise search means that the voice search request submitted by the user in the map search is a search for a specific POI data point, fuzzy search can be based on the sug engine, according to the voice search request Broad search for partial information, similarity, etc. It is worth noting that the error correction and expansion process of phoneme representation can be an independent processing process independent of the map information search, or it can be a preprocessing process integrated with the map information search function.

S230. Determine the degree of association between at least one piece of candidate information and the current user according to the characteristics of the map domain.

In a specific embodiment of the present invention, the map domain characteristics of the current user include at least one of the current map search scene characteristics, the current user behavior characteristics, and the map information search quality characteristics of the candidate text.

Optionally, the current map search scene feature is determined in the following manner: according to the spatial position relationship between the current user's current position and the position represented by the candidate information, determine the current map search scene feature; and/or, the current user voice for The spatial position description of the position represented by the candidate information is used as the feature of the current map search scene.

In this embodiment, in the map information search, the candidate information generally indicates a specific map POI, and further, the candidate information may indirectly indicate a specific position or position range of the POI. In view of the fact that the user is usually going to or planning to go to a certain destination when searching for map information, and then the user's current location can be centered and radiated outward to search for POIs that match the phoneme representation. For example, priority is given to recalling POIs from this city as candidate information. Therefore, this embodiment can determine the characteristics of the current map search scene according to the spatial position relationship between the current user's current location and the location represented by the candidate information, such as spatial affiliation, spatial adjacency, and spatial distance, so as to filter the candidate information. Exemplarily, if the user's current location belongs to city A, the location represented by candidate information 1 belongs to city A, and the location represented by candidate information 2 belongs to city B, then the difference between the current user's current location and the location represented by candidate information 1 can be determined. The spatial position relationship between them is spatially adjacent, and the spatial positional relationship between the current position of the current user and the position represented by candidate information 2 is determined as spatial distance.

In addition, in the map information search, the user's voice may include a spatial location description of the POI to be searched, and the spatial location description includes a relationship between the POI to be searched and at least one other location or location range. Therefore, in this embodiment, the spatial position description of the position represented by the candidate information in the current user's voice can be used as the feature of the current map search scene to filter the candidate information. Exemplarily, assuming that the user's voice is "restaurant S in city A", the constraint condition that restaurant S belongs to the spatial location description of city A is taken as the feature of the current map search scene.

Optionally, the behavior characteristics of the current user are determined in the following manner: determine the historical search behavior of the current user for the position represented by the candidate information, and the pronunciation habits of the current user; Behavioral characteristics.

In this embodiment, for a user who is currently searching for map information, the user's historical search behavior can be obtained, and the current user's historical search behavior for the location represented by the candidate information can be determined, and the current user's historical search behavior for the location represented by the candidate information can be obtained. Search behavior is used as the current user behavior characteristics. For example, the historical search time and historical search times of the current user for the location indicated by the candidate information can be obtained. In addition, based on the user's historical search behavior, pronunciation habits such as biased features of the user's pronunciation can be determined, and the current user's pronunciation habit can be used as the current user behavior feature. For example, it is determined that the current user's pronunciation preference for f and h is h.

Optionally, the map information search quality feature of the candidate text is determined in the following manner: determining the similarity between the candidate information and the map information in the preset map information database, and the map search user's perception of the position indicated by the candidate information Historical search demand distribution; according to the similarity and/or historical search demand distribution, determine the map information search quality characteristics of the candidate text.

In this embodiment, map corpus or map information may be pre-determined as a map information base, and information related to real POIs is stored in the map information base. Wherein, if the candidate information matches the map information in the preset map information database, it can be determined that the candidate information is an actual POI, rather than an interfering word with similar pronunciation in the common corpus. Therefore, in this embodiment, the similarity between the candidate information and the map information in the preset map information database can be used as the map information search quality feature of the candidate text. Exemplarily, the similarity threshold can be preset, and if the similarity between the candidate information and the map information in the preset map information library meets the preset similarity threshold, the map information search quality of the candidate information can be determined higher.

In addition, the historical search behavior of map search users can reflect the search trend of the majority of users for map information from a macro perspective, so it is possible to obtain the historical search behavior of map search users in real time or regularly, and determine the map search users' preference for candidate information. The historical search demand distribution of the represented location takes the historical search demand distribution of the map search users for the location represented by the candidate information as the map information search quality feature of the candidate text. Exemplarily, map users have higher search requirements for POIs in the downtown area of city A, but lower search requirements for development areas around city A. For another example, with the development of small videos, the search demand of map users for a certain Internet celebrity POI has increased significantly in the near future.

In this embodiment, according to at least one of the map domain characteristics, the degree of association between each candidate information and the current user is determined. Exemplarily, for the scene features of the current map search, the region or city to which the user currently belongs may be determined based on the spatial affiliation, and for candidate information having the same affiliation with the region or city, determine the relationship between the candidate information and the user's The degree of association is a larger value. For example, if the current location of the current user belongs to city A, then the degree of association between the candidate information belonging to city A and the user is set to a larger value. In addition, for candidate information that satisfies the spatial position description in the user's voice, determine that the degree of association between the candidate information and the current user is a relatively large value, and conversely, for candidate information that does not satisfy the spatial position description, determine that the candidate information The degree of association with the current user is small, or even zero.

Exemplarily, for the current user behavior characteristics, based on the principle that the user has a higher probability of repeated searches, according to the historical search behavior of the current user, for the historical search time and historical search quantity of the location represented by the candidate information, for a certain historical search time Candidate information with a higher number of historical searches in the database is determined to be more relevant to the current user. In addition, in combination with the current user's pronunciation habits, if it is determined that the current user is confused about the pronunciation of at least two phonemes, it may be determined that the candidate information corresponding to the at least two phonemes has the same degree of association with the current user.

For another example, for the map information search quality of the candidate text, if it is determined that the candidate information is an actually existing map POI according to the similarity between the candidate information and the map information in the preset map information database, then the The map search quality is high, and the degree of association between the candidate information and the current user can be set to a relatively large value. In addition, according to the historical search demand distribution, if the historical search demand of the location represented by the candidate information is higher, the degree of association between the candidate information and the current user can be set to be greater.

In this embodiment, the degree of association between the candidate information determined based on various characteristics in the map domain characteristics and the current user may be integrated to obtain the degree of association determined based on various characteristics of the candidate information. Among them, the degree of association of candidate information under various characteristics can be synthesized based on big data and machine learning models to obtain the degree of association between candidate information and the current user; The preset weights of each feature of each feature are weighted and summed based on the scoring results and weights of each feature to obtain the degree of association between the candidate information and the current user.

S240. Determine ambiguous information that is ambiguous with the current user's voice according to the degree of association between the at least one candidate information and the current user.

In a specific embodiment of the present invention, the candidate information may be sorted according to the degree of association between the candidate information and the current user, and the candidate information with a lower degree of association may be determined as ambiguity information that interferes with the search results in the map search. Wherein, an association degree threshold or a percentage threshold may be preset, and candidate information lower than the association degree threshold or a preset percentage threshold number with a lower association degree may be determined as ambiguous information.

S250. Filter out ambiguous information from at least one piece of candidate information, so as to determine a map information recognition result of the current user voice.

In a specific embodiment of the present invention, the ambiguous information is filtered out from the candidate information, and the filtered candidate information can be used as the identification result of the map information, and displayed to the user in order of the degree of association from high to low, so that the user can preferentially view To the identification result of the map information with the highest degree of correlation with the person, you can also refer to the candidate information with a relatively low degree of correlation. Alternatively, the candidate information with the highest degree of association may be directly used as the map information identification result and displayed to the user.

Exemplarily, FIG. 3 is an overall architecture diagram of speech recognition in this embodiment. As shown in Figure 3, the user inputs a voice search request to the map search client, and the voice recognition model performs voice recognition based on the received user voice, and determines the phoneme representation as fuzzy sound. Among them, the speech recognition model is based on the general speech recognition model obtained from the general corpus training, and the speech recognition model obtained after the second training using the map corpus, thus avoiding the high cost of retraining the speech recognition model, not only maintaining the Knowledge in the general field, and improve the recognition accuracy of the speech recognition model for map information. Secondly, fuzzy sounds are used for map search, which includes error correction and expansion of phoneme representation, so that multiple candidate information can be searched. Among them, the powerful matching capability based on map search replaces simple pinyin-to-text matching, and directly filters out the possible interference of general domain knowledge. Finally, the current user's map domain characteristics are used to disambiguate the candidate information to obtain the top N recognition texts with the strongest correlation with the current user, and the recognition text or the recognition text with the highest correlation is selected as the accurate text feedback to the user .

In the technical solution of this embodiment, through the speech recognition model optimized based on the map corpus, the acoustic feature recognition of the current user's voice is carried out, the phoneme representation of the current user's voice is determined, and the error correction and expansion of the phoneme representation are performed as fuzzy sounds, and based on the fuzzy sounds Carry out map information search to obtain at least one possible candidate information, and finally determine the degree of association between the candidate information and the current user according to the current user's map domain characteristics, disambiguate the multiple candidate information, and filter out the candidate information from the multiple candidate information. The candidate information that best matches the user is used as the map information recognition result. Through the recognition of fuzzy sounds, the embodiment of the present invention not only retains the characteristics of the user's voice and sound, but also avoids the problem of wrong text selection. Through the optimization of the speech recognition model and the map search of fuzzy sounds, it replaces the simple pinyin-to-text Matching removes the possible interference of general domain knowledge on map search, avoids misjudgments caused by ambiguity and accents, and greatly improves the speech recognition accuracy of map speech search.

Embodiment three

Fig. 4 is a schematic structural diagram of a speech recognition device provided by Embodiment 3 of the present invention. This embodiment is applicable to the situation where map information is searched based on the user's voice. The device can be configured in a server to implement any embodiment of the present invention. The speech recognition method described above. The device specifically includes the following:

Candidate information determination module 410, configured to search for map information on the current user voice, and determine at least one matching candidate information;

The voice recognition disambiguation module 420 is configured to disambiguate the at least one piece of candidate information according to the current user's map domain characteristics, so as to determine the map information recognition result of the current user's voice.

Optionally, the speech recognition disambiguation module 420 is specifically used for:

determining the degree of association between the at least one candidate information and the current user according to the map domain characteristics;

determining ambiguous information that is ambiguous with the current user's voice according to the degree of association between the at least one candidate information and the current user;

Filter out the ambiguous information from the at least one candidate information to determine a map information recognition result of the current user voice.

Optionally, the map domain feature of the current user includes at least one of the current map search scene feature, the current user behavior feature, and the map information search quality feature of the candidate text.

Optionally, the current map search scene feature is determined in the following manner:

According to the spatial relationship between the current user's current location and the location represented by the candidate information, determine the current map search scene features; and/or,

The spatial position description of the position represented by the candidate information in the current user's voice is used as the feature of the current map search scene.

Optionally, the current user behavior characteristics are determined in the following manner:

Determining the historical search behavior of the current user for the location represented by the candidate information, and the pronunciation habits of the current user;

According to the historical search behavior and/or the pronunciation habit of the current user, the behavior characteristics of the current user are determined.

Optionally, the map information search quality feature of the candidate text is determined in the following manner:

Determine the similarity between the candidate information and the map information in the preset map information database, and the historical search demand distribution of map search users for the positions represented by the candidate information;

According to the similarity and/or the historical search demand distribution, the map information search quality feature of the candidate text is determined.

Optionally, the candidate information determining module 410 includes:

A phoneme recognition unit 4101, configured to perform acoustic feature recognition on the current user's voice, and determine the phoneme representation of the current user's voice;

A map search unit 4102, configured to search for map information according to the phoneme representation, and determine at least one candidate information that matches the pronunciation of the phoneme representation.

Optionally, the map search unit 4102 is specifically used for:

performing error correction and phoneme extension on the phoneme representation, and determining at least one variant phoneme representation;

Obtain at least one candidate information that matches the pronunciation of the phoneme representation and the at least one variant phoneme representation.

The technical scheme of this embodiment realizes the identification of phoneme representation (i.e. fuzzy sound), the correction and expansion of fuzzy sound, the search of map information based on fuzzy sound, the determination of map domain characteristics, Disambiguation of candidate information and feedback of accurate text recognition. The embodiment of the present invention disambiguates the candidate information obtained by the special search of the map domain information, which not only removes the possible interference of the general domain knowledge on the map search, but also avoids misjudgment caused by ambiguity and accent, making the search The obtained map information recognition results are more in line with the user's habits and needs, which greatly improves the voice recognition accuracy of map voice search.

Embodiment four

FIG. 5 is a schematic structural diagram of a server provided by Embodiment 4 of the present invention, and FIG. 5 shows a block diagram of an exemplary server suitable for implementing the embodiment of the present invention. The server shown in FIG. 5 is only an example, and should not limit the functions and scope of use of this embodiment of the present invention.

The server 12 shown in FIG. 5 is only an example, and should not limit the functions and scope of use of this embodiment of the present invention.

As shown in FIG. 5, server 12 takes the form of a general-purpose computing device. Components of server 12 may include, but are not limited to: one or more processors 16, system memory 28, bus 18 connecting various system components (including system memory 28 and processor 16).

Bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor, or a local bus using any of a variety of bus structures. These architectures include, by way of example, but are not limited to Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MAC) bus, Enhanced ISA bus, Video Electronics Standards Association (VESA) local bus, and Peripheral Component Interconnect ( PCI) bus.

Server 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by server 12 and include both volatile and nonvolatile media, removable and non-removable media.

System memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and/or cache memory 32 . The server 12 may further include other removable/non-removable, volatile/nonvolatile computer system storage media. By way of example only, storage system 34 may be used to read and write to non-removable, non-volatile magnetic media (not shown in FIG. 5, commonly referred to as a "hard drive"). Although not shown in FIG. 5, a disk drive for reading and writing to removable nonvolatile disks (e.g., "floppy disks") may be provided, as well as for removable nonvolatile optical disks (e.g., CD-ROM, DVD-ROM or other optical media) CD-ROM drive. In these cases, each drive may be connected to bus 18 via one or more data media interfaces. System memory 28 may include at least one program product having a set (eg, at least one) of program modules configured to perform the functions of various embodiments of the present invention.

Program/utility 40 may be stored, for example, in system memory 28 as a set (at least one) of program modules 42 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include the implementation of the network environment. The program module 42 generally executes the functions and/or methods in the embodiments described in the embodiments of the present invention.

Server 12 may also communicate with one or more external devices 14 (e.g., keyboards, pointing devices, displays 24, etc.), and may also communicate with one or more devices that enable users to interact with Server 12 is capable of communicating with any device (eg, network card, modem, etc.) that communicates with one or more other computing devices. Such communication may occur through input/output (I/O) interface 22 . Moreover, the server 12 can also communicate with one or more networks (eg, a local area network (LAN), a wide area network (WAN) and/or a public network, such as the Internet) through the network adapter 20 . As shown, network adapter 20 communicates with other modules of server 12 via bus 18 . It should be appreciated that although not shown, other hardware and/or software modules may be used in conjunction with server 12, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID systems, tape drives, and Data backup storage system, etc.

The processor 16 executes various functional applications and data processing by running the programs stored in the system memory 28 , such as implementing the speech recognition method provided by the embodiment of the present invention.

Embodiment five

Embodiment 5 of the present invention also provides a computer-readable storage medium, on which a computer program (or called computer-executable instructions) is stored. When the program is executed by a processor, it is used to perform a speech recognition method. The method includes :

Carrying out a map information search on the current user's voice, and determining at least one matching candidate information;

Disambiguation processing is performed on the at least one piece of candidate information according to the characteristics of the current user's map domain, so as to determine the map information recognition result of the current user's voice.

The computer storage medium in the embodiments of the present invention may use any combination of one or more computer-readable media. The computer readable medium may be a computer readable signal medium or a computer readable storage medium. A computer readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (non-exhaustive list) of computer readable storage media include: electrical connections with one or more leads, portable computer disks, hard disks, random access memory (RAM), read only memory (ROM), Erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

A computer readable signal medium may include a data signal carrying computer readable program code in baseband or as part of a carrier wave. Such propagated data signals may take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. .

Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including - but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

Computer program code for carrying out operations of embodiments of the present invention may be written in one or more programming languages, or combinations thereof, including object-oriented programming languages—such as Java, Smalltalk, C++, including A conventional procedural programming language—such as "C" or a similar programming language. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In cases involving a remote computer, the remote computer can be connected to the user computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (such as through an Internet service provider). Internet connection).

Note that the above are only preferred embodiments of the present invention and applied technical principles. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the protection scope of the present invention. Therefore, although the embodiments of the present invention have been described in detail through the above embodiments, the embodiments of the present invention are not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention. , and the scope of the present invention is determined by the scope of the appended claims.

Claims (11)

1. A speech recognition method, characterized in that, comprising: Carrying out a map information search on the current user's voice, and determining at least one matching candidate information; Disambiguation processing is performed on the at least one piece of candidate information according to the characteristics of the current user's map domain, so as to determine the map information recognition result of the current user's voice. 2. The method according to claim 1, wherein the at least one candidate information is disambiguated according to the map domain characteristics of the current user, so as to determine the map information recognition result of the current user's voice, include: determining the degree of association between the at least one candidate information and the current user according to the map domain characteristics; determining ambiguous information that is ambiguous with the current user's voice according to the degree of association between the at least one candidate information and the current user; Filter out the ambiguous information from the at least one candidate information to determine a map information recognition result of the current user voice. 3 . The method according to claim 2 , wherein the map domain characteristics of the current user include at least one of current map search scene characteristics, current user behavior characteristics, and map information search quality characteristics of candidate texts. 4. The method according to claim 3, wherein the current map search scene feature is determined in the following manner: According to the spatial relationship between the current user's current location and the location represented by the candidate information, determine the current map search scene features; and/or, The spatial position description of the position represented by the candidate information in the current user's voice is used as the feature of the current map search scene. 5. The method according to claim 3, wherein the current user behavior characteristics are determined in the following manner: Determining the historical search behavior of the current user for the location represented by the candidate information, and the pronunciation habits of the current user; According to the historical search behavior and/or the pronunciation habit of the current user, the behavior characteristics of the current user are determined. 6. The method according to claim 3, wherein the map information search quality feature of the candidate text is determined in the following manner: Determine the similarity between the candidate information and the map information in the preset map information database, and the historical search demand distribution of map search users for the positions represented by the candidate information; According to the similarity and/or the historical search demand distribution, the map information search quality feature of the candidate text is determined. 7. The method according to claim 1, wherein said searching for map information on the current user voice and determining at least one matching candidate information includes: performing acoustic feature recognition on the current user's voice, and determining the phoneme representation of the current user's voice; A map information search is performed according to the phoneme representation, and at least one candidate information matching the pronunciation of the phoneme representation is determined. 8. The method according to claim 7, wherein said searching for map information according to said phoneme representation and determining at least one candidate information matching the pronunciation of said phoneme representation comprises: performing error correction and phoneme extension on the phoneme representation, and determining at least one variant phoneme representation; Obtain at least one candidate information that matches the pronunciation of the phoneme representation and the at least one variant phoneme representation. 9. A speech recognition device, characterized in that, comprising: Candidate information determination module, used to search the current user voice for map information, and determine at least one matching candidate information; The voice recognition disambiguation module is configured to perform disambiguation processing on the at least one candidate information according to the current user's map domain characteristics, so as to determine the map information recognition result of the current user's voice. 10. A server, characterized in that, comprising: one or more processors; memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors are made to implement the speech recognition method according to any one of claims 1-8. 11. A computer-readable storage medium, on which a computer program is stored, wherein, when the program is executed by a processor, the speech recognition method according to any one of claims 1-8 is implemented.