WO2013178002A1 - 一种语音识别匹配的方法和设备,以及计算机程序和存储介质 - Google Patents
一种语音识别匹配的方法和设备,以及计算机程序和存储介质 Download PDFInfo
- Publication number
- WO2013178002A1 WO2013178002A1 PCT/CN2013/074933 CN2013074933W WO2013178002A1 WO 2013178002 A1 WO2013178002 A1 WO 2013178002A1 CN 2013074933 W CN2013074933 W CN 2013074933W WO 2013178002 A1 WO2013178002 A1 WO 2013178002A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- character information
- determined
- fields
- similarity
- character
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 38
- 238000004590 computer program Methods 0.000 title claims abstract description 8
- 238000006243 chemical reaction Methods 0.000 claims abstract description 10
- 230000011218 segmentation Effects 0.000 claims description 23
- 238000004364 calculation method Methods 0.000 claims description 11
- 238000005516 engineering process Methods 0.000 abstract description 8
- 238000010586 diagram Methods 0.000 description 4
- 238000011156 evaluation Methods 0.000 description 4
- 241000282414 Homo sapiens Species 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 229910002651 NO3 Inorganic materials 0.000 description 1
- NHNBFGGVMKEFGY-UHFFFAOYSA-N Nitrate Chemical compound [O-][N+]([O-])=O NHNBFGGVMKEFGY-UHFFFAOYSA-N 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000010076 replication Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/08—Speech classification or search
Definitions
- the present invention relates to the field of artificial intelligence technologies in computer science, and in particular, to a method and apparatus for voice recognition matching, and a computer program and a storage medium.
- Voice is not only the most natural, effective and convenient tool for information exchange between human beings, but also an important tool for communication between people and machines.
- the mobile terminal searches for the phone number information of the contact stored in the mobile terminal according to the voice instruction of the user, and illustrates the application of the voice recognition technology in the prior art.
- Step 1 Receive a voice command of the name of the contact person sent by the user, and determine the pinyin information corresponding to the voice command according to the converted voice signal.
- Step 2 Determine the contact name corresponding to the pinyin information from the stored contact phone number according to the exact matching algorithm of the pinyin information.
- the perfect matching algorithm refers to comparing the received pinyin information with the locally stored pinyin information to determine whether the received pinyin information is completely consistent with the locally stored pinyin information.
- the pinyin information is compared with the locally stored pinyin information by using a perfect matching algorithm of the pinyin information.
- the comparison result is that the received pinyin information is completely consistent with the locally stored pinyin information
- the local pinning information is completely established according to the local
- the correspondence between the pinyin information and the contact name will determine the contact name of the contact corresponding to the locally stored pinyin information that is completely consistent with the received pinyin information. Name.
- Step 3 According to the correspondence between the locally stored contact name and the phone number, the phone number information of the contact corresponding to the received voice command is obtained.
- Embodiments of the present invention provide a method and a device for voice recognition matching, and a computer program and a storage medium, which are used to solve the problem that a voice recognition rate is low in a prior voice recognition technology.
- a method for speech recognition matching comprising:
- the character information stored in the form of pinyin and Chinese characters in the local database is fuzzyly matched according to the pinyin, and the character information in the form of Chinese characters matching the converted character information in the local database is obtained. .
- a voice recognition matching device comprising:
- a determining module configured to determine character information in a pinyin form obtained by converting the voice information
- a fuzzy matching module configured to convert the character obtained from the local database in the character information stored in the form of pinyin and Chinese characters according to the fuzzy pinyin matching strategy The information is fuzzy matched according to the pinyin, and the character information in the form of Chinese characters matching the converted character information in the local database is obtained.
- the embodiment of the present invention converts the character information stored in the form of pinyin and Chinese characters from the local database according to the fuzzy matching strategy.
- the character information obtained by the method is fuzzy matched according to pinyin, and the character information in the form of Chinese characters matching the converted character information in the local database is obtained, and a single perfect matching strategy in the prior art is extended to the character in the form of pinyin converted.
- the information is fuzzy matched according to the pinyin, which effectively increases the speech recognition rate of the transformed character information, thereby improving the efficiency of the speech recognition technology.
- FIG. 1 is a flowchart of a method for voice recognition matching according to Embodiment 1 of the present invention
- Figure 2 is a flow chart of the fuzzy perfect matching strategy
- Figure 3 is a schematic flow chart of a partial fuzzy matching strategy
- FIG. 4 is a flowchart of a method for voice recognition matching according to Embodiment 2 of the present invention.
- FIG. 5 is a schematic structural diagram of a voice recognition matching device according to Embodiment 3 of the present invention.
- an embodiment of the present invention provides a method and a device for voice recognition matching, and a computer program and a storage medium, which determine character information in a pinyin form obtained by converting voice information, and according to a fuzzy pinyin matching strategy,
- the character information stored in the form of pinyin and Chinese characters in the local data the character information obtained by the conversion is fuzzyly matched according to the pinyin, and the character information in the form of Chinese characters matching the converted character information in the local database is obtained.
- the character information of the converted character information is extracted from the character information in the form of pinyin and Chinese characters stored in the local database.
- Pinyin is used for fuzzy matching, and a single perfect matching strategy is used in the prior art to extend the character information of the converted pinyin form according to the pinyin fuzzy matching, thereby effectively increasing the speech recognition rate of the transformed character information, thereby improving The efficiency of speech recognition technology.
- Example 1 Example 1:
- FIG. 1 is a flowchart of a method for voice recognition matching according to an embodiment of the present invention.
- the method includes: Step 101: Determine character information in a pinyin form obtained by converting the voice information.
- step 101 the user sends a voice message to the terminal that can recognize the voice information, and when receiving the voice message, the terminal can analyze the voice message by itself, and determine the character information in the pinyin form obtained by converting the voice information;
- the received voice information may be uploaded to the voice recognition server, and the voice recognition server parses the received voice information, and sends the character information in the pinyin form converted by the determined voice information to the terminal.
- the voice information includes contact information and/or current operation information to be executed, for example: a voice message is: Calling Zhang San, wherein Zhang 3 belongs to contact information; "calling" belongs to current L. Line operation information.
- a voice message is: Go to Zhongguancun Square, where Zhongguancun belongs to similar contact information; "go" belongs to current execution: for information.
- voice information may be information in the form of a voice command, which is not specifically limited herein.
- the terminal and/or the voice recognition server parse the received voice information, initially identify the voice information, and convert the voice information indicating the contact information into character information in a pinyin form.
- the speech recognition server parses the received speech information, it can only receive the received speech information according to the preset sound model.
- the character information in the form of pinyin obtained by converting the voice information is not completely consistent with the voice information sent by the user, and the collected voice information may be incomplete. Therefore, the pinyin converted from the voice information is here.
- the character information of the form is regarded as the ambiguous character information, that is, the indeterminate character information.
- Step 102 According to the fuzzy pinyin matching strategy, the character information stored in the form of pinyin and Chinese characters in the local database is fuzzyly matched according to the pinyin, and the Chinese character form matching the converted character information in the local database is obtained. Character information.
- step 102 from the character information stored in the local database in the form of pinyin and kanji, the character information obtained by the conversion is fuzzy matched according to the pinyin.
- the character information obtained by the conversion is fuzzy matched according to the pinyin.
- Fuzzy exact match as shown in the figure, is the flow diagram of the fuzzy perfect match strategy, which includes:
- Step U Find the character information of the same field number in the form of pinyin from the local database according to the determined number of fields of the character information.
- the field refers to character information in the form of pinyin that can uniquely determine a Chinese character form, for example: ong "determines a Chinese character "east” or other Chinese characters that emit the same sound. At this time, "dong” is regarded as a measure.
- the number of the fields refers to the number of words included in the determined character information, for example: "dong xi mm bei" is the determined character information, wherein “dong” determines a Chinese character; “xi” determines a Chinese character;
- character information in the form of pinyin having the same number of fields is searched from the local database based on the determined number of fields of the character information. For example, find character information in the form of Pinyin with 4 fields.
- Step 12 Perform similarity calculation on the determined character information and the found character information respectively, and determine, from the found character information, character information whose similarity satisfies the first threshold condition.
- Step 1 Perform the following operations on each field in the determined character information and the corresponding field in a searched character information, until each field in the determined character information and the corresponding field in the found character information Similarity:
- the preset pinyin pair list refers to: an exception of the Chinese pinyin according to the difference in the sounding vowels, but the pronunciation characteristics are similar, or the sounding mother distinguishing criteria are small but the pronunciation is very different.
- Initials 1 r are generally considered to be similar, but when they carry the final i, the pronunciation of ri and li is very different, so ⁇ ri, li ⁇ belongs to a set of pinyin pairs, stored in the list of pinyin pairs, The similarity is small, corresponding to a similarity value; in addition, hui and fei are different regardless of the initial or the final, but the pronunciation is very similar, so ⁇ hui, fei ⁇ also belongs to a set of pinyin pairs, stored in the list of pinyin pairs. Among them, the similarity/replication is larger, corresponding to a similarity value.
- the initials and finals of the field are separated, and the initial and the final phase of the field in the searched character information are respectively determined, and the field and the found character information are obtained.
- the similarity between the corresponding fields are obtained.
- the corresponding field refers to a position of a field in the determined character information in the determined character information and a position of a field in the found character information in the searched character information.
- !b dongxi" and "torigsbi” where "dong” and Ixmg” are fields corresponding to each other in the character information, and "dong” and “shi” are not fields corresponding to each other in the character information.
- the phase set for the preset pinyin pair list means that the similarity between two pinyin that is close to or distant from each other can be determined according to the pronunciation of the initials.
- the quantized data representation is stored locally in the form of a table, and the degree of similarity can also be determined by probability, that is, the probability of a pinyin error in which two pronunciations are close is determined.
- a field in the determined character information is "bill”
- the corresponding field in the found character information is "fei”
- the similarity table set in the pinyin pair stored in the list is the preset pinyin pair in the list. Find and determine the similarity between "hui” and "fei”.
- the determined character information is preprocessed, and the unrecognized pinyin contained therein is converted into recognizable pinyin.
- u and V are often used to refer to ii in Chinese Pinyin, such as lv (Lv), yuan (yuan), for convenience of handling, Uniformly ii corresponds to v, in particular, when the initials are jq, x and y, when ⁇ contains ⁇ , u is converted to V,
- the method of comprehensive evaluation is that the weighted summation can be used to obtain the comprehensive similarity, and the comprehensive evaluation result can be determined according to the determined relationship between the initial similarity and the finality of the final: when the determined initial similarity and finality of the final When the similarity is higher or at least one similarity is higher, the initial similarity and the final similarity are added to obtain a comprehensive evaluation result; when the determined initial similarity and final similarity are lower, The initial similarity of the initials and the similarity of the finals are added together with a weighting factor to obtain a comprehensive evaluation result.
- Step 2 After the similarity between each field in the character information determined by Chad and the corresponding field in the found character information, the similarity between the character information is determined according to the phase between the fields.
- the similarity between the obtained fields is comprehensively calculated, and the determined character information is obtained.
- Step 3 From the found character information, determine the character information whose similarity satisfies the first threshold condition.
- the first threshold condition refers to that the similarity reaches a set threshold.
- the set threshold may be determined by the data collected by the practice, or may be determined according to the probability value of the voice model, and is not limited.
- the similarity is compared with the first threshold condition.
- the found character information whose similarity satisfies the first threshold condition is determined; when the obtained similarity does not satisfy the first threshold condition At the time, you can continue the second partial fuzzy match or return the search failure score.
- Step 13 Convert the character information satisfying the first threshold condition into a Chinese character form, and use the character information in the Chinese character form as the character information in the matched Chinese character form.
- Partial fuzzy matching is a schematic diagram of the process of the partial fuzzy matching strategy, which specifically includes:
- Step 21 Find, according to the number of the fields in the determined character information, character information in a pinyin form that is not equal to (greater than or smaller than) the number of the fields, and the number of fields of the character information found is greater than the determined character.
- step 22 is performed; when the number of fields of the character information found is less than the number of fields of the determined character information, step 24 is performed.
- the number of fields of the character information to be searched is greater than the number of fields in the determined character information, that is, if the number of fields in the determined character information is 4, the number of search fields in the local database is greater than 4 or less than 4.
- Step 22 When the number of the field of the character information that is found is greater than the number of the field of the determined character information, the character information that is found is split, where the content of each word segment after the same character information is split is different, and The number of fields in the word segmentation is the same as the number of fields in the determined character information.
- each of the found character information is split, wherein the principle of splitting is that the content of each word segment after the same character information is split is different, and the number of fields in the word segment and the number of fields in the determined character information are different. the same.
- the determined character information is "yong tao”
- the character information found is "zhang yong tao” which will be split to find out the character information.
- the split results are: “zhangycmg”, “zhangtao” and " Yongtao "three participles",
- the similarity between each word segmentation after the found character information is split and the determined character information is determined.
- the character information still determined is “yong tao”, and the character information found is “zliang yong tao”, and the 4 pieces are used to split the character information, and the split result is: "zhaiigyong”, “zhangtao” and "yongtao” is divided into three parts. At this time,
- each participle obtained after the split is only part of the character information found, the similarity between each of the split words and the determined character information is obtained, and the similarity of each part after the split is obtained.
- the similarity of a participle with the highest similarity is selected as the degree of correlation between the found character information and the nitrated character information.
- a weighting coefficient may be selected according to the value of the difference between the number of fields of the found character information and the number of fields of the determined character information, and the found character information is similar to the determined character information.
- the degree is obtained by weighting the similarity of each participle after splitting.
- the rule for determining the weighting coefficient is: if the value of the difference between the number of fields of the character information found and the number of fields of the determined character information is smaller, the smaller the weighting coefficient, if the number of fields of the character information found is determined The larger the value of the difference between the number of fields of the character information, the larger the weighting coefficient.
- Step 2 3 If the similarity between the segmented word segmentation and the determined character information meets the second threshold condition, the found character information is converted into a Chinese character form, and the Chinese character form is The character information is used as the character information of the matched Chinese character form.
- the root can determine the similarity between each piece of the segmented character information and the determined character information, and obtain the similarity between the found character information and the determined character information, and the similarity will be obtained. Comparing with the second threshold condition, when the obtained similarity satisfies the second threshold condition, determining the found character information that the similarity satisfies the second threshold condition, and converting the found character information into a Chinese character form, The character information in the form of a Chinese character is used as the character information in the form of the matched Chinese character; when the obtained similarity does not satisfy the second threshold condition, the result of the search failure is returned, indicating that the voice information is re-entered.
- the second threshold condition refers to that the similarity reaches a set threshold.
- the set value can be determined according to the data collected by the practice, or can be determined according to the probability value of the voice model, and is not limited.
- the "first" and “second” in the first threshold condition and the second threshold condition have no special meaning, only that this is two different thresholds.
- determining between each word segmentation after the found character information and the determined character information determining between each word segmentation after the found character information is split and the determined character information Whether the similarity is greater than the set threshold, if yes, determining the searched character information that is greater than the set threshold, and converting the found character information into a Chinese character form, the character information in the Chinese character form Character information in the form of the matched Chinese character; otherwise, the result of the search failure is returned, indicating that the voice information is re-entered.
- the set threshold value refers to a value at which the similarity reaches the set value.
- the set threshold value may be determined according to the data collected by the practice, or may be determined according to the probability value of the voice model, and is not limited.
- the root ⁇ determines the similarity between each of the segmented character information and the determined character information
- the similarity between the found character information and the determined character information is obtained, wherein two search results exist.
- the character information is the same as the determined character information, at this time, it will look up
- the character information with a smaller number of word segments obtained by splitting out the character information is preferentially compared.
- Step 24 When the number of the character information fields found is less than the determined number of character information fields, the determined character information is split, wherein the content of each word segment after the same character information is split is different, and the word segmentation is The number of fields is the same as the number of fields in the found character information.
- Step 25 If the similarity between the determined word segmentation and the found character information satisfies the second threshold condition, the found character information is converted into a Chinese character form, and the character in the Chinese character form is The information is used as the character information of the matched Chinese character form.
- step 25 The specific implementation of this step 25 is the same as that of step 23, and will not be described in detail herein.
- the fuzzy perfect matching method and the partial fuzzy matching method may be a progressive relationship. When the matching fuzzy character matching method does not determine the matching character information, the character matching is continued through the partial fuzzy matching manner. Operation; fuzzy perfect matching method and partial fuzzy matching method may also be a side-by-side relationship. When character information of a Chinese character form is determined for the character information of the pinyin form obtained by converting a certain voice information, one of the methods is selected for matching dry work. .
- the fuzzy perfect matching and/or the partial fuzzy matching manner is adopted, and the character information in the form of the matching Chinese character is searched from the local database;
- fuzzy perfect matching method for fuzzy matching not only the similarity between initials and finals is considered, but also the similarity of the pronunciations of some special letters in ordinary Chinese life in ordinary life.
- the speech is performed by such fuzzy perfect matching. Recognition, improving the recognition rate of speech recognition, and enhancing the accuracy of character information in the form of Chinese characters determined by character information in the form of pinyin
- Embodiment 2 is a diagrammatic representation of Embodiment 1:
- FIG. 4 it is a flowchart of a method for voice recognition matching according to Embodiment 2 of the present invention.
- the second embodiment is a detailed description of the steps in the first embodiment, and the method specifically includes:
- Step 201 Determine character information in a pinyin form obtained by converting the voice information.
- Step 202 Determine whether the determined character information can be entered, and the line is completely matched. If yes, Then, the character information in the form of a Chinese character corresponding to the character information of the nitrate is returned; otherwise, step 203 is performed. In this step 202, all the character information in the form of pinyin and Chinese characters included in the local database are compared with the converted character information, and it is determined whether the character information in the local database has one-to-one correspondence with the converted character information.
- the character information in the form of a Chinese character corresponding to the character information in the pinyin form in the local database corresponding to the corresponding relationship is used as the character information in the form of the Chinese character corresponding to the determined character information, and is returned to the user for viewing.
- Step 203 Determine whether the partial character matching operation can be performed on the determined character information, and if yes, return the character information in the form of a Chinese character corresponding to the determined character information. If not, execute step 204.
- the partial complete matching operation comprises:
- the searched character information is split, wherein the content of each word segment after the same character information is split is different, and the number of fields in the word segmentation is determined.
- the number of fields in the character information is the same, and the similarity between the segmented word segmentation and the determined character information of the found character information is determined;
- the determined character information is split, wherein the content of each word segment after the same character information is split is different, and the number of words in the word segment is It is the same as the number of fields in the found character information, and determines the similarity between the found character information and the segmentation of the determined character information.
- the finding is obtained according to the degree of similarity between each piece of the segmented word information and the determined character information determined by the found character information or the similarity between the found character information and the segmented word segment of the determined character information. Comparing the extracted character information with the determined character information, comparing the obtained similarity with the third threshold condition, and determining that the similarity satisfies the third threshold condition when the obtained phase satisfies the third threshold condition Character information, and converting the found character information into a Chinese character form, and using the character information in the Chinese character form as the character information of the matched Chinese character form; When the obtained similarity does not satisfy the third threshold condition, step 204 is performed.
- the third threshold condition refers to a threshold value at which the degree of convergence reaches a set value.
- the threshold value may be determined according to the data collected by the practice, or may be determined according to the probability value of the voice model, and is not limited.
- the first threshold condition, the second threshold condition and the second threshold condition of "first”, “second” and “third” have no special meaning, only that this is three different thresholds.
- Step 204 Determine whether the determined character information can be subjected to fuzzy complete matching, and if yes, return the character information in the form of Chinese characters corresponding to the determined character information, and if no, execute step 205.
- step 203 and step 204 may also be performed simultaneously. If steps 203 and 204 are performed simultaneously, the Chinese characters corresponding to the second threshold condition character information are returned. The character information in the form and the character information in the form of a Chinese character corresponding to the character information satisfying the third threshold condition.
- Step 205 Determine whether the determined partial character information can be subjected to a partial fuzzy matching operation. If yes, the character information in the form of a Chinese character corresponding to the determined character information is returned, and if not, a matching failure message is returned, indicating that the voice information is resent.
- Embodiment 3 is a diagrammatic representation of Embodiment 3
- FIG. 5 is a schematic structural diagram of a voice recognition matching device according to Embodiment 3 of the present invention.
- the voice recognition matching device includes: a determining module 31 and a fuzzy matching module 32, wherein:
- a determining module 31 configured to determine character information in a pinyin form obtained by converting the voice information
- a fuzzy matching module 32 configured to convert the character information stored in the form of pinyin and Chinese characters from the local database according to the fuzzy pinyin matching strategy The character information is fuzzy matched according to the pinyin, and the character information in the form of a Chinese character matching the converted character information in the local database is obtained.
- the fuzzy matching module 32 specifically includes: a first character information searching unit 41, a similarity meter The calculation unit 42 and the first matching result determination unit 43, wherein:
- the first character information searching unit 41 is configured to search, according to the determined number of fields in the character information, the character information in the pinyin form of the same number of fields from the local database;
- the similarity calculation unit 42 is configured to perform a similarity operation on the determined character information and the found character information, and determine, from the found character information, the character information that the similarity satisfies the first threshold condition;
- the first matching result determining unit 43 is configured to convert the character information whose similarity satisfies the first threshold condition into a Chinese character form, and use the character information in the Chinese character form as the character information in the matched Chinese character form.
- the similarity calculation unit 42 the character information specific to each of the determined fields to find out a character information fields in the following? Di obtained until a determined character information in each field and find out The similarity of the fields in the character information:
- the similarity of the setting is used as the degree between the field and the corresponding field in the found character information
- the similarity between the character information is determined according to the similarity between the fields.
- the fuzzy matching module 32 further includes: a second character information searching unit 44, a splitting unit 45, and a second matching result determining unit 46, where:
- the second character information searching unit 44 is configured to search, according to the determined number of fields in the character information, character information in a pinyin form different from the number of the fields from the local database;
- the splitting unit 45 is configured to: when the number of character information fields found is greater than the determined character information word When the number of segments is divided, the found character information is split, wherein the content of each word segment after the same character information is split is different, and the number of fields in the word segment is the same as the number of fields in the determined character information, and If the number of the character information fields found is less than the determined number of character information fields, the determined character information is split, wherein the content of each word segment after the same character information is split is different, and the number of fields in the word segment is found. The number of fields in the character information is the same;
- the second matching node determining unit 46 is configured to: when the number of the character information fields found is greater than the determined number of character information fields, if the similarity between the segmented word segmentation and the determined character information is satisfied
- the second threshold condition converts the found character information into a Chinese character form, and uses the character information in the Chinese character form as the character information of the matched Chinese character form, and when the number of the searched character information fields is less than the determined character If the number of information fields is different, if the similarity between the determined word segmentation and the character information in the search satisfies the second threshold condition, the found character information is converted into a Chinese character form, and the Chinese character form is The character information is used as the character information of the matched Chinese character form.
- the device further includes: a first determining module 33 and a second determining module 34, wherein: the first determining module 33 is configured to determine whether the determined character information can be completely completed before performing fuzzy matching on the converted character information. Matching operation
- the second judging module 34 further determines whether the partial matching operation can be performed on the determined character information when the judgment result of the first judging module is no, and if not, triggers the fuzzy matching module 12.
- the second determining module 34 is configured to trigger the fuzzy matching module 32 when the determining result of the first determining module is negative, and perform a partial perfect matching operation on the determined character information.
Landscapes
- Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Document Processing Apparatus (AREA)
Abstract
一种语音识别匹配的方法和设备,以及计算机程序和存储介质,其主要内容包括:在确定语音信息转化得到的拼音形式的字符信息(101)后,根据模糊匹配策略,从本地数据库中以拼音和汉字形式存储的字符信息中,对转化得到的字符信息根据拼音进行模糊匹配(102),将现有技术中采用单一的完全匹配策略扩展至对转化得到的拼音形式的字符信息根据拼音进行模糊匹配,有效地增加了对转化得到的字符信息的语音识别率,进而提高了语音识别技术的效率。
Description
一种语音识别匹配的方法和设备, 以及计算机程序和存储介质
技术领域 本发明涉及计算机科学中人工智能技术领域,尤其涉及一种语音识别匹配 的方法和设备, 以及计算机程序和存储介质。
背景技术
语音不仅是人类之间信息交流最自然、 最有效、 最方便的工具, 而且也 成为人与机器之间进行通信的重要工具。
随着科学技术的不断发展, 人工智能作为计算机科学的分支, 致力于研 发一种新的能以人类智能相似的方式做出反应的智能机器, 该领域的研究包 括机器人、 语言识別、 图像识别、 自然语言处理和专家系统等, 其中, 语音 识別作为一个分支, 以语音为研究对象, 其目标是将人类的语音中的词汇内 容转换为计算机可执行的输入符号进 实现语音识别。
以移动终端根据用户的语音指令, 查找移动终端中存储的联系人电话号 码信息为例, 说明现有技术中语音识別技术的应用。
第一步: 接收用户发出的包舍联系人姓名的语音指令, 并根据语音指令 转化后的语音信号确定该语音指令对应的拼音信息。
第二步: 根据拼音信息的完全匹配算法, 从存储的联系人电话号码中确 定该拼音信息对应的联系人姓名。
所述完全匹配算法是指将接收到的拼音信息与本地存储的拼音信息进行 比较, 确定接收到的拼音信息与本地存储的拼音信息是否完全一致。
具体地, 利用拼音信息的完全匹配算法, 将接收到的拼音信息与本地存 储的拼音信息进^"比较, 当比较结果为接收到的拼音信息与本地存储的拼音 信息完全一致时, 根据本地建立的拼音信息与联系人姓名之间的对应关系, 将确定与接收到的拼音信息完全一致的本地存储的拼音信息对应的联系人姓
名。
第三步: 根据本地存储的联系人姓名和电话号码之间的对应关系, 得到 接收到的语音指令对应的联系人的电话号码信息。
由于中文自身的特殊性以及不同的地方方言的多样性, 不同地方的用户 针对同一中文词汇发出的语音指令也存在差异 , 在语音识别服务器将语音指 令转化为拼音信息过.程中,并不能考虑到发出该语音指令的用户的口音特性, 简单的侬据本地存储的大词汇表进行语音到拼音的转化, 使得转化后的拼音 相对于接收到的语音指令存在误差, 而后再简单的根据拼音信息之间的对应 关系确定联系人姓名 , 将使得语音识别率大大降低。
由此可见, 在现有技术中, 语音识.别技术中存在语音识別率较低的问题。 发明内容
本发明实施例提供了一种语音识別匹配的方法和设备,以及计算机程序和 存储介质, 用于解决 前语音识别技术中存在的语音识别率较低的问题。
一种语音识别匹配的方法, 所述方法包括:
确定语音信息转化得到的拼音形式的字符信息;
根据模糊拼音匹配策略, 从本地数据库中以拼音和汉字形式存储的字符 信息中, 为转化得到的字符信息根据拼音进行模糊匹配, 得到本地数据库中 与转化后的字符信息匹配的汉字形式的字符信息。
—种语音识别匹配设备, 所述设备包括:
确定模块, 用于确定语音信息转化得到的拼音形式的字符信息; 模糊匹配.模块, 用于根据模糊拼音匹配策略, 从本地数据库中以拼音和 汉字形式存储的字符信息中,为转化得到的字符信息根据拼音进行模糊匹配 , 得到本地数据库中与转化后的字符信息匹配的汉字形式的字符信息。
本发明有益效莱如下:
本发明实施例在确定语音信息转化得到的拼音形式的字符信息后, 根据 模糊匹配策略, 从本地数据库中以拼音和汉字形式存储的字符信息中, 对转
化得到的字符信息根据拼音进行模糊匹配, 得到本地数据库中与转化后的字 符信息匹配的汉字形式的字符信息, 将现有技术中采用单一的完全匹配策略 扩展至对转化得到的拼音形式的字符信息根据拼音进行模糊匹配, 有效地增 加了对转化得到的字符信息的语音识别率 ,进而提高了语音识别技术的效率。 附图说明
图 1为本发明实施例一的一种语音识别匹配的方法的流程图;
图 2为模糊完全匹配策略的流程图;
图 3为部分模糊匹配策略的流程示意图;
图 4为本发明实施例二的一种语音识別匹配的方法的流程图;
图 5为本发明实施例三的一种语音识别匹配设备的结构示意图。
为了实现本发明的目的,本发明实施例提供了一种语音识别匹配的方法和 设备, 以及计算机程序和存储介质, 确定语音信息转化得到的拼音形式的字 符信息, 并根据模糊拼音匹配策略, 从本地数据戽中以拼音和汉字形式存储 的字符信息中, 为转化得到的字符信息根据拼音进行模糊匹配, 得到本地数 据库中与转化后的字符信息匹配的汉字形式的字符信息。
与现有技术相比, 在确定语音信息转化得到的拼音形式的字符信息后, 根据模糊匹配策略,从本地数据库中存储的以拼音和汉字形式的字符信息中, 对转化得到的字符信息根椐拼音进行模糊匹配, 将现有技术中采用单一的完 全匹配策略扩展至对转化得到的拼音形式的字符信息根据拼音进行模糊匹 配, 有效地增加了对转化得到的字符信息的语音识别率 , 进而提高了语音识 別技术的效率。
下面结合说明书附图对本发明各实施例进行详细描述。
实.施例一:
如图 1所示, 为本发明实施例―的一种语音识别匹配的方法的流程图。 该方法包括:
步骤 101 : 确定语音信息转化得到的拼音形式的字符信息。
在步骤 101 中, 用户向可识别语音信息的终端发出语音信息 , 终端在接 收到该语音信息时, 可以自身对该语音倌息进行解析, 确定该语音信息转化 得到的拼音形式的字符信息; 还可以将接收到的语音信息上传至语音识别服 务器, 由语音识别服务器对接收到的该语音信息进行解析, 并将确定的语音 信息转化得到的拼音形式的字符信息发送给终端
所述语音信息中包含了联系人信息和 /或当前待执行操作信息, 例如: 一 条语音信息为: 给张三打电话, 其中, 张三属于联系人信息; "打电话"属于 当前待 L.行操作信息。 再例如一条语音信息为: 去中关村广场, 其中, 中关 村属于类似联系人信息; "去" 属于当前执行 :作信息。
需要说明的是, 所述语音信息可以是语音指令形式的信息, 这里不做具 体限定。
具体地, 终端和 /或语音识別服务器对接收到的语音信息进行解析、 初步 识別该语音信息, 将其中表示联系人信息的语音信息转化为拼音形式的字符 信息。
由于用户之间发音存在差别, 以及汉语中一些文字在不同地区发音存在 差异, 因此, 语音识别服务器对接收到的语音信息进行解析时, 只能根据预 先设定的声音模型对接收到的语音信息进行解析, 存在将语音信息转化得到 的拼音形式的字符信息与用户发出的语音信息不完全一致的情况, 可能还存 在采集的语音信息是不完整的, 因此, 在这里将由语音信息转化得到的拼音 形式的字符信息看作是.椟糊的字符信息, 即不确定的字符信息。
步驟 102: 根据模糊拼音匹配策略, 从本地数据库中以拼音和汉字形式 存储的字符信息中, 为转化得到的字符信息根据拼音进行模糊匹配, 得到本 地数据库中与转化后的字符信息匹配的汉字形式的字符信息。
在步骤 102中, 根据模糊拼音匹配策略, 从本地数据库中以拼音和汉字 形式存储的字符信息中, 为转化得到的字符信息根据拼音进行模糊匹配的方
式有两种: 一种方式是模糊完全匹配; 另一种方式是部分模糊匹配。
笫一种方式: 模糊完全匹配, 如图 所示, 为模糊完全匹配策略的流程 图, 具体包括:
步骤 U : 根据确定的字符信息的字段数量, 从本地数据库中查找出相同 字段数量的拼音形式的字符信息。
所述字段是指拼音形式的字符信息中能唯一确定一个汉字形式的字符信 息,例如: ong"确定一个汉字 "东"或者发相同音的其他汉字,此时, "dong" 被看作是措音形式的字符信息中的一个字段。
所述字段数量是指确定的字符信息中包含字的个数, 例如: "dong xi mm bei" 是确定的字符信息, 其中, "dong" 确定一个汉字; "xi" 确定一个汉字;
"纖" 确定一个汉字; "bei" 确定一个汉字; 因此, 该确定的字符信息中的 字段数量为 4,
具体地, 根据确定的字符信息的字段数量, 从本地数据库中查找具有相 同字段数量的拼音形式的字符信息。 例如, 查找具有 4个字段数量的拼音形 式的字符信息。
步骤 12: 将确定的字符信息分别与查找出的字符信息进行相似度运算 , 从查找出的字符信息中, 确定相似度满足第一阈值条件的字符信息。
其中, 将确定的字符信息分別与查找出的字符倌息进行相似度运算的具 体方式为:
第一步: 将确定的字符信息中的每个字段与一个查找出的字符信息中的 相应字段进行以下操作, 直至荻得确定的字符信息中每个字段与查找出的字 符信息中的相应字段的相似度:
首先: 判断确定的字符信息中的一个字段是否与查找出的字符信息中的 相应字段是否在预设的拼音对列表中。
所述预设的拼音对列表是指: 中文拼音依据声韵母区分准则差别较大但 发音特性相近或依据声韵母区分准则差别较小但发音差别很大的例外情况。
例如: 声母 1、 r通常认为比较相近, 但是当它们带上韵母 i时, ri和 li的发 音差别就很大, 因此 {ri, li}属于一组拼音对, 存储在拼音对列表中 , 其相似 度较小, 对应一个相似度值; 另外, hui和 fei其无论声母还是.韵母都不相同, 但发音却很相近, 因此 {hui, fei}也属于一组拼音对, 存储在拼音对列表中, 其相似 /复较大, 对应一个相似度值。
其次: 若判断结杲为是, 则根据拼音对列表中为预设的拼音对设定的相 似度, 将该设定的相似度作为该字段与查找出的字符信息中的相应字段之间 的相似度;
若判断结果为否, 则分离该字段的声母和韵母, 分别确定该字段与查找 出的字符信息中的字段的声母相^度和韵母相 ^度, 并得到该字段与查找出 的字符信息中的相应字段之间的相似度。
其中, 所述相应字段是指确定的字符信息中的一个字段在确定的字符信 息中的位置与一个查找出的字符信息中的一个字段在查找出的字符信息中的 位置. 对应, '例如: !bdongxi" 和 "torigsbi" , 其中, "dong" 和 Ixmg" 是 字符信息中互为相应的字段, "dong"和 "shi" 不是字符信息中互为相应的字 段。
所述为预设的拼音对列表设定的相 ^度是指根据实践中无法依椐声母韵 母区分准则只能依椐读音确定某两个发音接近或相远的拼音之间的相似度, 通过量化的数据表示, 以表格的形式存储在本地, 也可以通过概率的方式确 定相似度, 即确定某两个发音接近的拼音出错的概率。
例如: 确定的字符信息中的一个字段为 "bill" , 查找出的字符信息中的 相应字段为 "fei" ,根椐本地存储的拼音对列表中为预设的拼音对设定的相似 度表, 查找并确定 "hui" 与 "fei" 之间的相似度。
较优地, 在分离该字段的声母和韵母之前, 对确定的字符信息进行预处 理, 将其中包含的无法识別的拼音转化成可识別的拼音。 例如: 电脑拼音中 经常用 u和 V指代汉语拼音中的 ii, 如 lv (吕), yuan (元) , 为了处理方便,
统一将 ii对应成 v, 特殊地, 在声母分別为 j q, x和 y时, 韵母中包含 ϋ时, 将 u转换.为 V、
分别分离确定的字符信息中每一个字段的声母和韵母与查找的字符信息 中相应字段的声母和韵母,并根据本地存储的声母相似度表和韵母'相似度表, 分别确定该字段与查找出的字符信息中的字段的声母相似度和韵母相似度, 并将确定的声母相似度和韵母相似度进.行综合评估, 得到该字段与查找出的 字符信息中的相应字段之间的相似度。
所述综合评估的方式为可以进行加权求和得到综合相似度, 也可以根据 确定的声母相似度与韵母相似度之间的大小关系, 确定综合评估结果: 当确 定的声母相似度与韵母相似度都属于相似度较高或至少有一项相似度较高 时, 則将声母相似度与韵母相似度进行加法运算得到综合评估结果; 当确定 的声母相似度与韵母相似度都较低时, 則将声母相似度与韵母相似度进行加 法运算同时加上一个加权因子得到综合评估结果。
假如相似度的馭值范围是 0〜1时, 大于 0.6的为相似度较高, 小于 0.4为 相似度较低,
第二步: 在荻得确定的字符信息中每个字段与查找出的字符信息中的相 应字段的相似度后, 根据各字段间的相 度, 确定字符信息间的相似度。
具体地, 在获得确定的字符信息中每个字段与一个查找出的字符信息中 的相应字符的相似度后 ,将得到的各字段间的相似度进行相似度的综合计算, 得到确定的字符信息与一个查找出的字符信息之间的相似度。
第三步: 从查找出的字符信息中, 确定相似度满足第一阈值条件的字符 信息。
具体地, 所述第一阈值条件是指相似度达到设定的阈值。 其中, 设定的 阈值可裉椐实践采集的数据确定, 也可以根据语音模型的概率值确定, 具体 不做限定。
裉据上述步驟得到的确定的字符信息与至少一个查找出的字符信息之间
的相似度与第一阈值条件进行比较, 当得到的相似度满足第一阈值条件时, 确定相似度满足第―阈值条件的查找出的字符信息; 当得到的相似度都不满 足第一阈值条件时, 可以继续第二种部分模糊匹配或者返回查找失败结杲。
步骤 13 : 将所述相^度满足第一阈值条件的字符信息转换为汉字形式, 并将该汉字形式的字符信息作为所述匹配的汉字形式的字符信息。
第二种方式: 部分模糊匹配, 如图 3所示, 为部分模糊匹配策略的流程 示意图, 具体包括:
步骤 21 : 根据确定的字符信息中的字段数量, 从本地数据库中查找出不 等于 (大于或小于) 所述字段数量的拼音形式的字符信息, 当查找出的字符 信息的字段数量大于确定的字符信息的字段数量时, 执行步骤 22; 当查找出 的字符信息的字段数量小于确定的字符信息的字段数量时, 执行步驟 24。
在本步中, 若查找的字符信息的字段数量大于确定的字符信息中的字段 数量, 即假设确定的字符信息中的字段数量为 4, 那么从本地数据库中查找 字段数量大于 4或者小于 4的拼音形式的字符信息。
步骤 22: 当查找出的字符信息的字段数量大于确定的字符信息的字段数 量时, 分别对查找出的字符信息进行拆分, 其中, 同一字符信息拆分后的每 个分词内容不相同, 且分词中的字段数量与确定的字符信息中的字段数量相 同。
具体地,针对查找出的每一个大于所述字段数量的拼音形式的字符信息, U亍以下操作:
首先, 将每一个查找出的字符信息进行拆分, 其中, 拆分的原则是同一 字符信息拆分后的每个分词内容不相同, 且分词中的字段数量与确定的字符 信息中的字段数量相同。
例如: 确定的字符信息为 "yong tao" , 查找出的一个字符信息为 "zhang yong tao ", 将对查找出该字符信息进行拆分, 拆分结果为: " zhangycmg "、 "zhangtao" 和 "yongtao" 三个分词',
其次, 针对查找出的字符信息拆分后的分词, 确定查找出的字符信息拆 分后的每一个分词与确定的字符信息之间的相似度。
仍以确定的字符信息为 "yong tao" ,查找出的一个字符信息为 "zliang yong tao" , 4夺对查找出该字符信息进行拆分,拆分结果为: "zhaiigyong"、 "zhangtao" 和 "yongtao" 三个分词为倒, 此时,
4 "zhaiigyong" 与 "yong tao" 进 ί亍相 乂/¾¾算, 确定 "zhangyong" 的 相似度 A1;
将 "zhangtao" 与 "yong tao" 进行相似度运算, 确定 "zhangtao" 的相似 度 A2;
将 "yongtao" 与 "yong tao" 进行相似度运算, 确定 "yongtao" 的相 4¾ 度 A3。
由于拆分后得到的每一个分词只是查找到的字符信息的一部分, 因此, 将拆分后的每一个分词与确定的字符信息进行相似度运算, 得到拆分后每一 个分词的相似度, 从中选出相似度最高的一个分词的相似度作为查找出的字 符信息与硝定的字符信息的相^度。
为了提高识别的精度, 还可以依据查找出的字符信息的字段数量与确定 的字符信息的字段数量之间差的数值大小选定一个加权系数, 则查找出的字 符信息与确定的字符信息的相似度通过拆分后每一个分词的相似度的进行加 权运算得到。
所述加权系数确定的规则为: 若查找出的字符信息的字段数量与确定的 字符信息的字段数量之间差的数值越小, 加权系数越小, 若查找出的字符信 息的字段数量与确定的字符信息的字段数量之间差的数值越大, 则加权系数 越大。
H i "zhangyong" , "zhangtao" 和 "yongtao" —三个分词与确定的字符信' 息的相似度为 AL A2和 A3 , 则 "zhangyongtao" 与确定的字符信息的最终 相似度为 min{Al , A2, A3 加权系数。
步骤 2,3 : 若查找出的字符信息拆分后的分词与确定的字符信息之间的相 似度满足第二阈值条件, 則将该查找出的字符信息转换为汉字形式, 并将该 汉字形式的字符信息作为所述匹配的汉字形式的字符信息。
具体地, 根椐确定查找出的字符信息拆分后的每一个分词与确定的字符 信息之间的相似度, 得到该查找出的字符信息与确定的字符信息的相似度, 将得到的相似度与第二阈值条件进行比较, 当得到的相似度满足第二阈值条 件时, 确定相似度满足第二阈值条件的查找出的字符信息, 并将该查找出的 字符信息转换为汉字形式, 将该汉字形式的字符信息作为所述匹配的汉字形 式的字符信息; 当得到的相似度都不满足第二阈值条件时, 返回查找失败结 果, 指示重新输入语音信息。
需要说明的是, 所述第二阈值条件是指相似度达到设定的阈值。 其中, 设定的阐值可根据实践采集的数据确定 ,也可以根据语音模型的概率值确定, 具体不傲限定。 第一阈值条件与第二阈值条件中的 "第一', 和 "第二" 没有 什么特别意义, 只表示这是两个不同的阈值。
较优地, 根据确定查找出的字符信息拆分后的每一个分词与确定的字符 信息之间的相似度, 判断查找出的字符信息拆分后的每一个分词与确定的字 符信息之间的相似度是否都大于设定的门限值, 若是, 则确定大于设定的门 限值的查找出的字符信息, 并将该查找出的字符信息转换为汉字形式, 将该 汉字形式的字符信息作为所述匹配的汉字形式的字符信息; 否则, 返回查找 失败结果, 指示重新输入语音信息。
所述设定的门限值是指相似度达到设定的数值。 其中, 设定的门限值可 根据实践采集的数据确定, 也可以根据语音模型的概率值确定, 具体不做限 定。
假设根椐确定查找出的字符信息拆分后的每一个分词与确定的字符信息 之间的相似度, 得到该查找出的字符信息与确定的字符信息的相似度, 其中, 存在两个查找出的字符信息与确定的字符信息的相似度相同, 此时, 将查找
出的字符信息中拆分得到的分词个数较少的字符信息优先进行比较。
步骤 24:当查找出的字符信息字段数量小于确定的字符信息字段数量时 , 则对确定的字符信息进行拆分, 其中, 同一字符信息拆分后的每个分词内容 不相同, 且分词中的字段数量与查找出的字符信息中的字段数量相同。
本步骤 24的具体实施方式与步骤 22的相同, 这里不再做具体描述。 步骤 25: 若确定的字符信息拆分后的分词与查找出的字符信息之间的相 似度满足第二阈值条件, 則将该查找出的字符信息转换为汉字形式, 并将该 汉字形式的字符信息作为所述匹配的汉字形式的字符信息。
本步骤 25的具体实施方式与步骤 23的相同, 这里不再做具体描述。 需要说明的是, 模糊完全匹配方式和部分模糊匹配方式可以是递进的关 系, 在通过.模糊完全匹配方式没有确定出相匹配的字符信息时, 继续通过部 分模糊匹配方式进.行字符信息匹配操作; 模糊完全匹配方式和部分模糊匹配 方式还可以是并列的关系, 在为某语音信息转化得到的拼音形式的字符信息 确定相应的汉字形式的字符信息时, 选择其中一种方式进行匹配燥作。
通过实施例一的方案, 在确定语音信息转化得到的拼音形式的字符信息 后, 采用了模糊完全匹配和 /或部分模糊匹配的方式, 从本地数据库中查找与 其匹配的汉字形式的字符信息; 在利用模糊完全匹配方式进行模糊匹配时, 不仅考虑到声母和韵母的相似度, 还考虑到中文语音中一些特殊字母在曰常 生活中存在的发音的相似性, 通过这样的模糊完全匹配方式进行语音识别, 提高了语音识别的识别率, 并且增强了由拼音形式的字符信息确定汉字形式 的字符信息的准确性„
实施例二:
如图 4所示, 为本发明实施例二的一种语音识別匹配的方法的流程图。 本实施例二是实施例一中各步骤的详细描述, 该方法具体包括:
步骤 201 : 确定语音信息转化得到的拼音形式的字符信息。
步驟 202: 判断是否能够对确定的字符信息进,行完全匹配搡作, 若是,
则返回硝定的字符信息对应的汉字形式的字符信息; 否则, 执行步驟 203。 在本步骤 202中, 将本地数据库中包含的所有的以拼音和汉字形式的字 符信息, 与转化得到的字符信息进行比较, 确定本地数据库中是否存在字符 信息与转化得到的字符信息一一对应, 当存在完全匹配的字符信息时, 将满 足——对应关系的本地数据库中的拼音形式的字符信息对应的汉字形式的字 符信息作为确定的字符信息对应的汉字形式的字符信息, 返回给用户进行查 看。
步骤 203: 判断是否能够对确定的字符信息进行部分完全匹配操作, 若 是,则返回确定的字符信息对应的汉字形式的字符信息,若否,执行步骤 204。
其中, 所述部分完全匹配操.作包括:
根据确定的字符信息的字段数量, 从本地数据库中查找出与所述字段数 量不同的拼音形式的字符信息;
若查找出的字符信息字段数量大于确定的字符信息字段数量, 对查找 出的字符信息进行拆分, 其中, 同一字符信息拆分后的每个分词内容不相同, 且分词中的字段数量与确定的字符信息中的字段数量相同, 并确定查找出的 字符信息拆分后的分词与确定的字符信息之间的相似度;
若查找出的字符信息字段数量小于确定的字符信息字段数量, 则对确定 的字符信息进行拆分, 其中, 同一字符信息拆分后的每个分词内容不相同, 且分词中的字.段数量与查找出的字符信息中的字段数量相同 , 并确定查找出 的字符信息与确定的字符信息拆分后的分词之间的相似度。
根据确定查找出的字符信息拆分后的每一个分词与确定的字符信息之间 的相^度或者查找出的字符信息与确定的字符信息拆分后的分词之间的相似 度, 得到该查找出的字符信息与确定的字符信息的相似度, 将得到的相似度 与第三阈值条件进行比较, 当得到的相^度满足第三阈值条件时, 确定相似 度满足第三阈值条件的查找出的字符信息, 并将该查找出的字符信息转换为 汉字形式, 将该汉字形式的字符信息作为所述匹配的汉字形式的字符信息;
当得到的相似度都不满足第三阈值条件时, 执行步驟 204。
需要说明的是, 所述第三阈值条件是指相^度达到设定的阈值。 其中, 设定的阈值可根据实践采集的数据确定,也可以根据语音模型的概率值确定, 具体不做限定。 第一阈值条件, 第二阈值条件和第二阈值条件中的 "第 ·一 ", "第二" 和 "第三" 没有什么特別意义, 只表示这是三个不同的阈值。
步骤 204: 判断是否能够对确定的字符信息进行模糊完全匹配搡作, 若 是,則返回确定的字符信息对应的汉字形式的字符信息,若否,执行步骤 205。
其中, 模糊完全匹配操作的具体实现方式见实施例一图 2对应的文字部 分, 这里不再做具体描述。
需要说明的是, 在本实施方案中, 除了上述实施顺序外, 步骤 203与步 骤 204还可以是同时执行的 , 若步骤 203和 204同时执行, 则返回满足第二 阔值条件字符信息对应的汉字形式的字符信息和满足第三阈值条件的字符信 息对应的汉字形式的字符信息。
步骤 205: 判断是否能够对确定的字符信息进—行部分模糊匹配操作, 若 是, 则逸回确定的字符信息对应的汉字形式的字符信息 , 若否, 返回匹配失 败消息, 指示重新发送语音信息。
其中, 部分模糊匹配操作的具体实现方式见实施例一图 3对应的文字部 分, 这里不再做具体描述。
实施例三:
如图 5所示,为本发明实施例三的一种语音识別匹配设,备的结构示意图。 所述语音识別匹配设备包括: 确定模块 31和模糊匹配模块 32, 其中:
确定模块 31 , 用于确定语音信息转化得到的拼音形式的字符信息; 模糊匹配模块 32, 用于根据模糊拼音匹配策略, 从本地数据库中以拼音 和汉字形式存储的字符信息中, 为转化得到的字符信息根据拼音进行模糊匹 配, 得到本地数据库中与转化后的字符信息匹配的汉字形式的字符信息。
所述模糊匹配模块 32, 具体包括: 第一字符信息查找单元 41 , 相似度计
算单元 42和第一匹配结果确定单元 43, 其中:
第一字符信息查找单元 41 , 用于根据确定的字符信息中的字段数量, 从 本地数据库中查找出相同字段数量的拼音形式的字符信息;
相似度计算单元 42, 用于将确定的字符信息分别与查找出的字符信息进 行相似度运算, 从查找出的字符信息中, 确定相似度满足第一阈值条件的字 符信息;
第一匹配结果确定单元 43, 用于将所述相似度满足第一阈值条件的字符 信息转换为汉字形式, 并将该汉字形式的字符信息作为所述匹配的汉字形式 的字符信息。
所述相似度计算单元 42, 具体用于将确定的字符信息中的每个字段与一 个查找出的字符信息中的字段进行以下操作 ? 直至荻得确定的字符信息中每 个字段与查找出的字符信息中的字段的相似度:
判断确定的字符信息中的一个字段与查找出的字符信息中的相应字段是 否在预设的拼音对列表中;
若是, 则根据拼音对列表为预设的拼音对设定的相似度, 将该设定的相 似度作为该字段与查找出的字符信息中的相应字段之间的相 · 度;
若否, 则分离该字段的声母和韵母, 分别确定该字段与查找出的字符信 息中的字段的声母相似度和韵母相似度, 并得到该字段与查找出的字符信息 中的相应字段之间的相^度;
在获得确定的字符信息中每个字段与查找出的字符信息中的字段的相似 度后, 根据各字段间的相似度, 确定字符信息间的相似度。
所述模糊匹配模块 32, 还包括: 第二字符信息查找单元 44、 拆分单元 45和第二匹配结果确定单元 46, 其中:
第二字符信息查找单元 44, 用于根据确定的字符信息中的字段数量, 从 本地数据库中查找出与所述字段数量不同的拼音形式的字符信息;
拆分单元 45 , 用于当查找出的字符信息字段数量大于确定的字符信息字
段数量时, 则对查找出的字符信息进行拆分, 其中, 同一字符信息拆分后的 每个分词内容不相同, 且分词中的字段数量与确定的字符信息中的字段数量 相同, 以及当查找出的字符信息字段数量小于确定的字符信息字段数量, 則 对确定的字符信息进行拆分, 其中, 同一字符信息拆分后的每个分词内容不 相同, 且分词中的字段数量与查找出的字符信息中的字段数量相同;
第二匹配结杲确定单元 46, 用于当查找出的字符信息字段数量大于确定 的字符信息字段数量时, 若查找出的字符信息拆分后的分词与确定的字符信 息之间的相似度满足第二阈值条件, 则将该查找出的字符信息转换为汉字形 式, 并将该汉字形式的字符信息作为所述匹配的汉字形式的字符信息, 以及 当查找出的字符信息字段数量小于确定的字符信息字段数量时, 若确定的字 符信息拆分后的分词与查找中的字符信息之间的相似度满足第二阈值条件, 则将该查找出的字符信息转换为汉字形式, 并将该汉字形式的字符信息作为 所述匹配的汉字形式的字符信息。
所述设备还包括: 第一判断模块 33和第二判断模块 34, 其中: 第一判断模块 33, 用于在为转化得到的字符信息进行模糊匹配之前, 判 断是否能够对确定的字符信息进行完全匹配操作;
第二判断模块 34 , 周于在第一判断模块的判断结果为否时, 进一步判断 是否能够对确定的字符信息进行部分完全匹配操作, 若否, 則触发模糊匹配 模块 12。
较优地, 第二判断模块 34 , 用于在第一判断模块的判断结果为否时, 触 发模糊匹配模块 32 , 同时, 执行对确定的字符信息进行部分完全匹配操作。
显然, 本领域的技术人员可以对本发明进行各种改动和变型而不脱离本 发明的精神和范围。 这样, 倘若本发明的这些修改和变型属于本发明权利要 求及其等同技术的范围之内, 則本发明也意图包含这些改动和变型在内。
Claims
1、 一种语音识別匹配的方法, 其特征在于, 所述方法包括:
确定语音信息转化得到的拼音形式的字符信息;
根据模糊拼音匹配策略, 从本地数据库中以拼音和汉字形式存储的字符 信息中, 为转化得到的字符信息根据拼音进.行模 匹配, 得到本地数据库中 与转化后的字符信息匹配的汉字形式的字符信息。
2、 如权利要求 1所述的语音识别匹配的方法, 其特征在于, 为转化得到 的字符信息进行模糊匹配, 具体包括:
根据确定的字符信息的字段数量, 从本地数据库中查找出相同字段数量 将确定的字符信息分别与查找出的字符信息进行相似度运算, 从查找出 的字符信息中, 确定相似度满足第一阈值条件的字符信息;
将所述相似度满足第一阈值条件的字符信息转换为汉字形式, 并将该汉 字形式的字符信息作为所述匹配的汉字形式的字符信息。
3、 如权利要求 2所述的语音识别匹配的方法, 其特征在于, 将确定的拼 音形式的字符信息分别与查找出的字符信息进行相似度运算, 具体包括: 将确定的字符信息中的每个字段与一个查找出的字符信息中的相应字段 进行以下搡作 , 直至获得确定的字符信息中每个字段与查找出的字符信息中 的相应字段的相似度:
判断确定的字符信息中的一个字段与查找出的字符信息中的相应字段是 否在预设的拼音对列表中;
若是, 则根据拼音对列表中为预设的拼音对设定的相似度, 将该设定的 相似度.作为该字段与査找出的字符信息中的相应字段之间的相似度;
若否, 则分离该字段的声母和韵母, 分别确定兹字段与查找出的字符信
息中的字段的声母相似度和韵母相似度, 并得到该字段与查找出的字符信息 中的相应字段之间的相^度;
在获得确定的字符信息中每个字段与查找出的字符信息中的相应字段的 相似度后, 根据各字段间的相似度, 确定字符信息间的相似度。
4、 如权利要求 i或 2所述的语音识别匹配的方法, 其特征在于, 为转化 得到的字符信息进行模糊匹配, 具体还包括:
裉据确定的字符信息的字段数量, 从本地数据戽中查找出与所述字段数 量不同的措音形式的字符信息;
当查找出的字符信息的字段数量大于确定的字符信息的字段数量时, 则 对查找出的字符信息进行拆分, 其中, 同一字符信息拆分后的每个分词内容 不相同, 且分词中的字段数量与确定的字符信息中的字段数量相同, 若查找 出的字符信息拆分后的分词与确定的字符信息之间的相似度满足第二阈值条 件、 則将该查找出的字符信息转换为汉字形式, 并将该汉字形式的字符信息 作为所述匹配的汉字形式的字符信息;
当查找出的字符信息字段数量小于确定的字符信息字段数量时, 則对确 定的字符信息进行拆分, 其中, 同一字符信息拆分后的每个分词内容不相同, 且分词中的字段数量与查找出的字符信息中的字段数量相同, 若确定的字符 信息拆分后的分词与查找出的字符信息之间的相似度满足第二阈值条件, 到 将该查找出的字符信息转换为汉字形式, 并将该汉字形式的字符信息作为所 述匹配的汉字形式的字符信息。
5、 如权利要求 i所述的语音识別匹配的方法, 其特征在于, 为转化得到 的字符信息进行模糊匹配之前, 所述方法还包括:
判断是否能够对确定的字符信息进行完全匹配操作;
若否,则进一步判断是否能够对确定的字符信息进行部分完全匹配搡作, 若否, 则执行对确定的字符信息进行模糊匹配搡作。
6、 如权利要求 5所述的语音识別匹配的方法, 其特征在于, 所述部分完
全匹配操作包括:
根据确定的字符信息的字段数量, 从本地数据库中查找出与所述字段数 量不同的4^音形式的字符信息;
若查找出的字符信息字段数量大于确定的字符信息字段数量, 则对查找 出的字符信息进行拆分, 其中, 同一字符信息拆分后的每个分词内容不相同, 且分词中的字段数量与确定的字符信息中的字段数量相同, 并确定查找出的 字符信息拆分后的分词与确定的字符信息之间的相似度;
若查找出的字符信息字段数量小于确定的字符信息字段数量, 则对确定 的字符信息进行拆分, 其中, 同一字符信息拆分后的每个分词内容不相同, 且分词†的字段数量与查找出的字符信息中的字段数量相同, 并确定查找出 的字符信息与确定的字符信息拆分后的分词之间的相似度。
7、 一种语音识別匹配设备, 其特征在于, 所述设备包括:
确定模块, 用于确定语音信息转 -化得到的拼音形式的字符信息; 模糊匹配模块, 用于根据模糊拼音匹配策略, 从本地数据库中以拼音和 汉字形式存储的字符信息中,为转化得到的字符信息根据拼音进行模糊匹配, 得到本地数据库中与转化后的字符信息匹配的汉字形式的字符信息。
8、 如权利要求 7所述的语音识別匹配设备, 其特征在于, 所述模糊匹配 模块, 具体包括:
第一字符信息查找单元, 用于根据确定的字符信息中的字段数量, 从本 地数据戽中查找出相同字段数量的拼音形式的字符信息;
相似度计算单元, 用于将确定的字符信息分别与查找出的字符信息进行 相似度运算, 从查找出的字符信息中, 确定相似度满足第一阈值条件的字符 信息;
第一匹配结果确定单元, 用于将所述相似度满足第一阈值条件的字符信 息转换为汉字形式, 并将该汉字形式的字符信息作为所述匹配的汉字形式的 字符信息。
9、 如权利要求 8所述的语音识别匹配设备, 其特征在于,
所述相似度计算单元 , 具体用于将确定的字符信息中的每个字段与一个 查找出的字符信息中的相应字段进行以下操作, 直至获得确定的字符信息中 每个字段与查找出的字符信息中的字段的相似度:
判断确定的字符信息中的一个字段与查找出的字符信息中的相应字段是 否在预设的拼音对列表中;
若是, 则根据拼音对列表为预设的拼音对设定的相似度, 将该设定的相 似度作为该字段与查找出的字符信息中的相应字段之间的相似度;
若否, 则分离该字段的声母和韵母, 分別确定该字段与查找出的字符信 息中的字段的声母相似度和韵母相似度, 并得到该字段与查找出的字符信息 中的相应字段之间的相似度;
在获得确定的字符信息中每个字段与查找出的字符信息中的字段的相似 度后, 根据各字段间的相似度, 确定字符信息间的相似度。
ί0、 如权利要求 7或 8所述的语音识别匹配设备, 其特征在于, 所述模 糊匹配模块, 还包括:
第二字符信息查找单元, 用于根据确定的字符信息中的字段数量, 从本 地数据库中查找出与所述字段数量不同的拼音形式的字符信息;
拆分单元, 用于当查找出的字符信息字段数量大于确定的字符信息字段 数量时, 则对查找出的字符信息进行拆分, 其中, 同一字符信息拆分后的每 个分词内容不相同, 且分词中的字段数量与确定的字符信息中的字段数量相 同, 以及当查找出的字符信息字段数量小于确定的字符信息字段数量, 则对 确定的字符信息进行拆分, 其中, 同一字符信息拆分后的每个分词内容不相 同, 且分词中的字段数量与查找出的字符信息中的字段数量相同;
第二匹配结果确定单元, 用于当查找出的字符信息字段数量大于确定的 字符信息字段数量时, 若查找出的字符信息拆分后的分词与确定的字符信息 之间的相^度满足第二阈值条件,剩将该查找出的字符信息转换为汉字形式,
并将该汉字形式的字符信息作为所述匹配的汉字形式的字符信息, 以及当查 找出的字符信息字段数量小于确定的字符信息字段数量时, 若确定的字符信 息拆分后的分词与查找中的字符信息之间的相^度满足第二阈值条件, 则将 该查找出的字符信息转换为汉字形式, 并将该汉字形式的字符信息作为所述 匹配的汉字形式的字符信息。
11 , 如权利要求 7所述的语音识別匹配设备, 其特征在于, 所述设备还 包括: 第一判断模块和第二判断模块, 其中:
第一判断模块, 用于在为转化得到的字符信息进行模糊匹配之前, 判断 是否能够对确定的字符信息进行完全匹配操作;
第二判断模块, 用于在第一判断模块的判断结果为否时, 判断是否能够 对确定的字符信息进行部分完全匹配操作 , 若否, 则触发模糊匹配模块。
12. 一种包括指令的计算机程序, 所述指令在由处理器执行时被设置成 使所述处理器执行如权利要求 1-6中任一项所述的方法。
1 3, 一种存储了如权利要求 12所述计算机程序的存儲介质。
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210171583.5 | 2012-05-29 | ||
CN201210171583.5A CN103456297B (zh) | 2012-05-29 | 2012-05-29 | 一种语音识别匹配的方法和设备 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2013178002A1 true WO2013178002A1 (zh) | 2013-12-05 |
Family
ID=49672387
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2013/074933 WO2013178002A1 (zh) | 2012-05-29 | 2013-04-28 | 一种语音识别匹配的方法和设备,以及计算机程序和存储介质 |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN103456297B (zh) |
WO (1) | WO2013178002A1 (zh) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109299471A (zh) * | 2018-11-05 | 2019-02-01 | 广州百田信息科技有限公司 | 一种文本匹配的方法、装置及终端 |
CN110209892A (zh) * | 2019-04-17 | 2019-09-06 | 深圳壹账通智能科技有限公司 | 敏感信息识别方法、装置、电子设备及存储介质 |
CN110232914A (zh) * | 2019-05-20 | 2019-09-13 | 平安普惠企业管理有限公司 | 一种语义识别方法、装置以及相关设备 |
CN110334348A (zh) * | 2019-06-28 | 2019-10-15 | 珍岛信息技术(上海)股份有限公司 | 一种基于纯文本中的文字校验方法 |
CN110489754A (zh) * | 2019-08-20 | 2019-11-22 | 杭州数澜科技有限公司 | 快速生成标准语料的方法和系统 |
CN110992959A (zh) * | 2019-12-06 | 2020-04-10 | 北京市科学技术情报研究所 | 一种语音识别方法及系统 |
CN112885348A (zh) * | 2021-01-25 | 2021-06-01 | 广州中汇信息科技有限公司 | 结合ai智能语音电销方法 |
CN115618451A (zh) * | 2022-12-06 | 2023-01-17 | 广东时谛智能科技有限公司 | 基于模糊匹配进行鞋体模型设计的方法及装置 |
Families Citing this family (22)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105187656B (zh) * | 2015-09-23 | 2018-09-07 | 百度在线网络技术(北京)有限公司 | 通讯录联系人匹配方法和装置 |
CN107016994B (zh) * | 2016-01-27 | 2020-05-08 | 阿里巴巴集团控股有限公司 | 语音识别的方法及装置 |
CN106021504A (zh) * | 2016-05-20 | 2016-10-12 | 深圳Tcl数字技术有限公司 | 字符串模糊匹配方法及装置 |
CN106297799A (zh) * | 2016-08-09 | 2017-01-04 | 乐视控股(北京)有限公司 | 语音识别处理方法及装置 |
CN106653031A (zh) * | 2016-10-17 | 2017-05-10 | 海信集团有限公司 | 语音唤醒方法及语音交互装置 |
CN106548777B (zh) * | 2016-11-25 | 2020-11-10 | 北京光年无限科技有限公司 | 一种用于智能机器人的数据处理方法及装置 |
CN106710585B (zh) * | 2016-12-22 | 2019-11-08 | 云知声(上海)智能科技有限公司 | 语音交互过程中的多音字播报方法及系统 |
CN107274894A (zh) * | 2017-05-19 | 2017-10-20 | 杭州仁盈科技股份有限公司 | 一种提高嘈杂环境下语音识别率的方法 |
CN107273359A (zh) * | 2017-06-20 | 2017-10-20 | 北京四海心通科技有限公司 | 一种文本相似度确定方法 |
CN107562907B (zh) * | 2017-09-11 | 2020-10-02 | 武汉科技大学 | 一种智能律师专家案件应答装置 |
CN108281144B (zh) * | 2018-01-23 | 2020-12-08 | 浙江国视科技有限公司 | 一种语音识别方法和系统 |
CN108363745B (zh) | 2018-01-26 | 2020-06-30 | 阿里巴巴集团控股有限公司 | 机器人客服转人工客服的方法和装置 |
CN108597537A (zh) * | 2018-04-06 | 2018-09-28 | 东莞市华睿电子科技有限公司 | 一种音频信号相似度检测方法 |
CN109377993A (zh) * | 2018-10-12 | 2019-02-22 | 上海庆科信息技术有限公司 | 智能语音系统及其语音唤醒方法及智能语音设备 |
CN109963022A (zh) * | 2018-12-27 | 2019-07-02 | 广州云趣信息科技有限公司 | 一种基于soundex算法的语音电话本找人方法及流程 |
CN110399608B (zh) * | 2019-06-04 | 2023-04-25 | 深思考人工智能机器人科技(北京)有限公司 | 一种基于拼音的对话系统文本纠错系统及方法 |
CN111105799B (zh) * | 2019-12-09 | 2023-07-07 | 国网浙江省电力有限公司杭州供电公司 | 基于发音量化和电力专用词库的离线语音识别装置及方法 |
CN111354334B (zh) * | 2020-03-17 | 2023-09-15 | 阿波罗智联(北京)科技有限公司 | 语音输出方法、装置、设备和介质 |
CN113539247B (zh) * | 2020-04-14 | 2024-06-18 | 京东科技控股股份有限公司 | 语音数据处理方法、装置、设备及计算机可读存储介质 |
CN111554297B (zh) | 2020-05-15 | 2023-08-22 | 阿波罗智联(北京)科技有限公司 | 语音识别方法、装置、设备及可读存储介质 |
CN112133295B (zh) * | 2020-11-09 | 2024-02-13 | 北京小米松果电子有限公司 | 语音识别方法、装置及存储介质 |
CN116798408A (zh) * | 2022-03-14 | 2023-09-22 | 中国移动通信集团设计院有限公司 | 语音识别方法、终端设备及计算机可读存储介质 |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1674091A (zh) * | 2005-04-18 | 2005-09-28 | 南京师范大学 | 地理信息的语音识别方法及其在导航系统中的应用 |
CN101206859A (zh) * | 2007-11-30 | 2008-06-25 | 清华大学 | 语音点歌方法 |
CN102254557A (zh) * | 2011-07-04 | 2011-11-23 | 深圳市子栋科技有限公司 | 基于自然语音识别的导航方法和系统 |
CN102347026A (zh) * | 2011-07-04 | 2012-02-08 | 深圳市子栋科技有限公司 | 基于自然语音识别的音频/视频点播方法和系统 |
-
2012
- 2012-05-29 CN CN201210171583.5A patent/CN103456297B/zh active Active
-
2013
- 2013-04-28 WO PCT/CN2013/074933 patent/WO2013178002A1/zh active Application Filing
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1674091A (zh) * | 2005-04-18 | 2005-09-28 | 南京师范大学 | 地理信息的语音识别方法及其在导航系统中的应用 |
CN101206859A (zh) * | 2007-11-30 | 2008-06-25 | 清华大学 | 语音点歌方法 |
CN102254557A (zh) * | 2011-07-04 | 2011-11-23 | 深圳市子栋科技有限公司 | 基于自然语音识别的导航方法和系统 |
CN102347026A (zh) * | 2011-07-04 | 2012-02-08 | 深圳市子栋科技有限公司 | 基于自然语音识别的音频/视频点播方法和系统 |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109299471A (zh) * | 2018-11-05 | 2019-02-01 | 广州百田信息科技有限公司 | 一种文本匹配的方法、装置及终端 |
CN109299471B (zh) * | 2018-11-05 | 2022-12-27 | 广州百田信息科技有限公司 | 一种文本匹配的方法、装置及终端 |
CN110209892A (zh) * | 2019-04-17 | 2019-09-06 | 深圳壹账通智能科技有限公司 | 敏感信息识别方法、装置、电子设备及存储介质 |
CN110232914A (zh) * | 2019-05-20 | 2019-09-13 | 平安普惠企业管理有限公司 | 一种语义识别方法、装置以及相关设备 |
CN110334348A (zh) * | 2019-06-28 | 2019-10-15 | 珍岛信息技术(上海)股份有限公司 | 一种基于纯文本中的文字校验方法 |
CN110334348B (zh) * | 2019-06-28 | 2022-11-15 | 珍岛信息技术(上海)股份有限公司 | 一种基于纯文本中的文字校验方法 |
CN110489754A (zh) * | 2019-08-20 | 2019-11-22 | 杭州数澜科技有限公司 | 快速生成标准语料的方法和系统 |
CN110489754B (zh) * | 2019-08-20 | 2023-01-03 | 杭州数澜科技有限公司 | 快速生成标准语料的方法和系统 |
CN110992959A (zh) * | 2019-12-06 | 2020-04-10 | 北京市科学技术情报研究所 | 一种语音识别方法及系统 |
CN112885348A (zh) * | 2021-01-25 | 2021-06-01 | 广州中汇信息科技有限公司 | 结合ai智能语音电销方法 |
CN112885348B (zh) * | 2021-01-25 | 2024-03-08 | 广州中汇信息科技有限公司 | 结合ai智能语音电销方法 |
CN115618451A (zh) * | 2022-12-06 | 2023-01-17 | 广东时谛智能科技有限公司 | 基于模糊匹配进行鞋体模型设计的方法及装置 |
Also Published As
Publication number | Publication date |
---|---|
CN103456297B (zh) | 2015-10-07 |
CN103456297A (zh) | 2013-12-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2013178002A1 (zh) | 一种语音识别匹配的方法和设备,以及计算机程序和存储介质 | |
CN107240398B (zh) | 智能语音交互方法及装置 | |
EP4068280A1 (en) | Speech recognition error correction method, related devices, and readable storage medium | |
WO2017215122A1 (zh) | 多语音识别模型切换方法、装置和存储介质 | |
Chen et al. | Query-by-example keyword spotting using long short-term memory networks | |
CN107016994B (zh) | 语音识别的方法及装置 | |
CN108183844B (zh) | 一种智能家电语音控制方法、装置及系统 | |
CN108694940B (zh) | 一种语音识别方法、装置及电子设备 | |
WO2014117645A1 (zh) | 信息的识别方法和装置 | |
WO2014101826A1 (zh) | 一种提高语音识别准确率的方法及系统 | |
CN104036774A (zh) | 藏语方言识别方法及系统 | |
JPH0470640B2 (zh) | ||
CN109036471B (zh) | 语音端点检测方法及设备 | |
CN105551485A (zh) | 语音文件检索方法及系统 | |
WO2021098318A1 (zh) | 应答方法、终端及存储介质 | |
CN113436614B (zh) | 语音识别方法、装置、设备、系统及存储介质 | |
CN111724766B (zh) | 语种识别方法、相关设备及可读存储介质 | |
Peng et al. | Search results based n-best hypothesis rescoring with maximum entropy classification | |
CN114155839A (zh) | 一种语音端点检测方法、装置、设备及存储介质 | |
JP5112978B2 (ja) | 音声認識装置、音声認識システムおよびプログラム | |
CN111862963B (zh) | 语音唤醒方法、装置和设备 | |
CN112669836B (zh) | 命令的识别方法、装置及计算机可读存储介质 | |
CN105244024A (zh) | 一种语音识别方法及装置 | |
KR20130068621A (ko) | 발화검증 기반 대용량 음성 데이터 자동 처리 장치 및 방법 | |
EP4030335A1 (en) | Artificial intelligence-based semantic identification method, device, and semantic identification apparatus |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 13796236 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 13796236 Country of ref document: EP Kind code of ref document: A1 |