WO2017016126A1 - 语音识别语法树的构图方法、装置、终端设备及存储介质 - Google Patents

语音识别语法树的构图方法、装置、终端设备及存储介质 Download PDF

Info

Publication number
WO2017016126A1
WO2017016126A1 PCT/CN2015/096624 CN2015096624W WO2017016126A1 WO 2017016126 A1 WO2017016126 A1 WO 2017016126A1 CN 2015096624 W CN2015096624 W CN 2015096624W WO 2017016126 A1 WO2017016126 A1 WO 2017016126A1
Authority
WO
WIPO (PCT)
Prior art keywords
slot
word
class
application scenario
path
Prior art date
Application number
PCT/CN2015/096624
Other languages
English (en)
French (fr)
Inventor
彭守业
贾磊
Original Assignee
百度在线网络技术（北京）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 百度在线网络技术（北京）有限公司 filed Critical 百度在线网络技术（北京）有限公司
Publication of WO2017016126A1 publication Critical patent/WO2017016126A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • 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
    • G10L15/06—Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice
    • 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
    • G10L15/28—Constructional details of speech recognition systems

Definitions

  • the embodiments of the present invention relate to the field of voice recognition technologies, and in particular, to a method, a device, a terminal device, and a storage medium for a voice recognition syntax tree.
  • Local speech recognition also requires syntactic tree composition for pre-identified text before recognition, that is, establishing all possible paths of user input speech.
  • speech recognition traversing the entire composition space, combined with speech recognition algorithm, finding and inputting the best speech. The matching path that will return the final result of the recognition.
  • composition is composed of a weighted finite state machine speech recognition composition algorithm, which will exhaust all possible paths. For example, for the above three statements “calling Zhang San”, “calling Zhang San” and “calling Zhang San's phone”, using the weight finite state machine voice recognition composition calculation When constructing a map, three separate paths will be constructed, corresponding to the statements “Call to Zhang San”, “Call Zhang San”, and “Call Zhang San”.
  • the corresponding acoustic model and voice model need to be repeatedly stored. Therefore, when the amount of data is large, the memory consumed by the composition is also large. In addition, in order to make the composition more compact, it is necessary to perform multiple traversal, and find the same node for merging, resulting in a long consumption of the composition.
  • the embodiment of the invention provides a composition method, a device, a terminal device and a storage medium of a speech recognition syntax tree, which can save the memory space occupied by the composition and the consumption time of the composition.
  • an embodiment of the present invention provides a method for composing a speech recognition syntax tree, including:
  • the embodiment of the present invention further provides a composition device for a speech recognition syntax tree, including:
  • a slot determining module configured to determine a slot corresponding to the application scenario, and assign a corresponding at least one word to each slot;
  • An acoustic model module configured to form a grammar sentence according to a slot order in a predetermined grammar path and a word corresponding to the slot, and parse the utterance of the word in the grammar statement to determine a corresponding acoustic model
  • a syntax tree building module configured to construct a syntax tree according to a slot in a predetermined syntax path of the application scenario, where the slot corresponds to a node in the syntax tree, and the slot index of the slot is stored in the node, where the slot index is used
  • the index corresponds to the acoustic model storage location of the word.
  • the embodiment of the present invention further provides a terminal device for implementing composition of a speech recognition syntax tree, including:
  • One or more processors are One or more processors;
  • One or more modules the one or more modules being stored in the memory, and when executed by the one or more processors, performing the following operations:
  • the acoustic model of the corresponding word stores the location.
  • an embodiment of the present invention further provides a non-volatile computer storage medium, where the computer storage medium stores one or more modules, when the one or more modules are configured by a speech recognition syntax tree.
  • the device of the method is executed, the device is caused to perform the following operations:
  • the slot corresponding to the application scenario is determined, and at least one word is assigned to each slot, and the syntax tree is constructed based on the slot, and information such as an acoustic model of the word corresponding to the same slot does not need to be repeatedly stored, and the slot may be used.
  • the index is indexed to a different syntax path, so when the composition process is completed, memory space and composition consumption time can be saved.
  • FIG. 1A is a schematic flowchart of a method for composing a speech recognition syntax tree according to Embodiment 1 of the present invention
  • FIG. 1B is a schematic structural diagram of a first syntax tree in a method for composing a speech recognition syntax tree according to Embodiment 1 of the present invention
  • FIG. 1C is a schematic structural diagram of a second syntax tree in a method for composing a speech recognition syntax tree according to Embodiment 1 of the present invention.
  • FIG. 2 is a schematic structural diagram of a composition device of a speech recognition syntax tree according to Embodiment 2 of the present invention.
  • FIG. 3 is a schematic structural diagram of a terminal device for implementing composition of a speech recognition syntax tree according to Embodiment 3 of the present invention.
  • the execution body of the composition method of the speech recognition syntax tree provided by the embodiment of the present invention may be the present invention
  • the composition device of the speech recognition syntax tree provided by the embodiment the device may be integrated into a mobile terminal device (for example, a smart phone, a tablet computer, etc.), or may be integrated into a server, and the composition device of the speech recognition syntax tree may adopt hardware or Software Implementation.
  • the composition method of the speech recognition syntax tree provided by the embodiment of the invention is particularly suitable for local speech recognition, and can adapt to the content recognition of a limited number of voice commands, but can also be applied to online speech recognition, correspondingly increasing the predetermined grammatical path and the number of slots. Yes, the following description will be made in conjunction with the embodiments.
  • FIG. 1A is a schematic flowchart of a method for composing a speech recognition syntax tree according to Embodiment 1 of the present invention. As shown in FIG. 1A, the method specifically includes:
  • S11 Determine a slot corresponding to the application scenario, and assign a corresponding at least one word to each slot.
  • the application scenario in this embodiment is a common scenario in which the user controls the operation of the terminal.
  • Typical application scenarios are, for example, calling, navigating, texting, and searching.
  • Each application scenario may involve at least one control instruction, usually involving multiple control instructions, and the substantive content of the instructions is often the same.
  • the embodiment of the present invention will be described by taking a call application scenario as an example.
  • the operation instructions generally involved include making a call to Zhang Yuanyuan, calling Zhang Yuanyuan, dialing a round call, etc., and determining the slot in the application scenario according to manual experience or data mining. : $SIL, $to, $action, $de, $name, and $phone, and assign each slot at least one word associated with the calling application scenario:
  • $SIL sil, the slot is a special starting slot for marking the starting point of the predetermined syntax path;
  • each slot can be represented by an array, and the right side of the equal sign is the word information associated with the slot, including at least one word.
  • the predetermined syntax path may be preset, and specifically, the slots determined in the above step 11 are configured in a certain syntax order, and the words in the slots are read in order to form a syntax statement.
  • the words in the grammar sentence are parsed to obtain corresponding voice information.
  • the speech information of each of the above three grammar sentences is obtained by parsing, and the corresponding acoustic model is determined, so as to quickly locate the corresponding grammar sentence in the speech recognition.
  • the determined acoustic model is as follows:
  • S13 Construct a syntax tree according to a slot in a predetermined syntax path of the application scenario, where the slot corresponds to a node in the syntax tree, and the slot index of the slot is stored in the node, where the slot index is used to index the corresponding word.
  • Acoustic model storage location Construct a syntax tree according to a slot in a predetermined syntax path of the application scenario, where the slot corresponds to a node in the syntax tree, and the slot index of the slot is stored in the node, where the slot index is used to index the corresponding word.
  • the slot index of the slot is stored in each node.
  • $SIL, $to, $action, $de, $name, and $phone can be used as the slot index for the slot, which can be a pointer or storage address, an acoustic model of the word in the slot, and other information.
  • the slot corresponding to the application scenario is determined, and at least one word is assigned to each slot, and a syntax tree is constructed based on the slot, and information such as an acoustic model of the word corresponding to the same slot does not need to be repeatedly stored, and may be indexed by slot. It is in a different syntax path, so when the composition process is completed, it can save memory space and composition time.
  • the method further includes:
  • the slot joint Storing the slot join index of the slot connection in a node corresponding to the sub-slot in the syntax tree, the slot joint The index is used to index the storage location of the acoustic model of the slot joint.
  • the slot connection represents association information between the parent slot and the child slot.
  • the word "call” in the above example is taken as an example, wherein the last acoustic information of "hit” is d- a+*, the first acoustic information of “telephone” is *-d+ian, * means the meaning of any word, but the individual acoustic information cannot confirm the contents of *, only by slot connection can they be displayed: d- a+d (* is shown as the initials of the phone), a-d+ian (* is shown as the finals of the call).
  • the acoustic context of the words is recorded to facilitate speech recognition. So for "calling” you need to build an acoustic model like "d-a+d".
  • the slot joints of adjacent slots will include multiple acoustic models. For example, the slot connection of $to$name will include multiple acoustic models such as "Give Zhang Yuanyuan”, “Give Li Si”, and “Give Wang Wu”.
  • the slot join index is stored in the node corresponding to the sub-slot. This saves memory and composition time spent on repetitive storage slot coupling acoustic models.
  • parent slot $to and child slot $name in the above-mentioned predetermined syntax path ($SIL$to$name$action$phone), and parent slot $to and child in ($SIL$action$phone$to$name)
  • the slot $name only needs to store the slot connection of the parent slot $to and the child slot $name in the predetermined syntax path ($SIL$to$name$action$phone), in the store ($SIL$action$phone$to$
  • the parent slot $to in name) and the slot of child slot $name are joined, since the slot join of $to$name is already built and stored, only the above predetermined syntax path ($SIL$to$name$action$phone) is stored.
  • the slot of the parent slot $to and the slot of the child slot $name can be joined by the corresponding slot join index.
  • the slot construction syntax tree in the predetermined syntax path specifically includes:
  • each of the predetermined syntax paths into a depth path of the syntax tree, the slots in the predetermined syntax path correspond to nodes in the depth path, and storing the slot index and the slot join index with the parent slot in the node, Store an end identifier at the tail node of each depth path;
  • the nodes corresponding to the same slot in the same layer in each depth path are merged.
  • the call application scenario in the foregoing embodiment is taken as an example.
  • the SIL of each predetermined syntax path is used as a starting slot to form a syntax tree.
  • the root node and merges the nodes corresponding to the same slot of the same layer in the predetermined syntax path, for example, the slot $action and the predetermined syntax path ($SIL) in the predetermined syntax path ($SIL$action$phone$to$name)
  • the slot $action in $action$name$de$phone) can be merged.
  • the start slot may also be an empty slot, and an empty slot is automatically formed as the root node of each predetermined syntax path in the application scenario.
  • the characteristics of the slots in each preset grammar path can be summarized, and corresponding identifiers are added to further save the composition resources. Specifically, at least one identifier is added, and an identifier may be added to the slot, or multiple at the same time:
  • Add a data identifier (_CORE), such as $name_CORE, that is, determine a slot corresponding to the application scenario, and assign at least one corresponding word to each slot includes:
  • Determining a data class slot corresponding to the application scenario adding a data identifier to the data class slot as the node information of the slot, and assigning a word in the setting database to the data class slot, where the setting database includes at least : Address book name library, address book phone number library, public phone library, or building name library.
  • the address book name library, the address book telephone number library, the public telephone library, or the building name library can all be obtained from the mobile terminal.
  • the data class slot includes a directory of address book names, which can be obtained from the address book of the terminal, for example, including: Zhang Yuanyuan, Zhang Yayuan, Zhu Dayuan, and Zhou Xiaoyuan.
  • Add a loop identifier (_LOOP), such as $number_LOOP, that is, determine a slot corresponding to the application scenario, and assign at least one corresponding word to each slot includes:
  • Determining a loop class slot corresponding to the application scenario adding a loop identifier to the loop class slot as node information of the slot, and assigning at least two words to the loop class slot, wherein the loop identifier is used to indicate
  • the loop uses the acoustic model of the loop-like slot for word recognition.
  • numbers in 0-9 can be stored in the class slot, and used for cyclically identifying the phone number information input by the user during voice recognition.
  • the mute slot $SIL can be set after the loop class slot, and when it is recognized that silence occurs, the loop ends.
  • Adding an optional identifier (_OPT), that is, determining a slot corresponding to the application scenario, and assigning each slot a corresponding at least one word includes:
  • An optional class slot corresponding to the application scenario is determined, an optional identifier is added to the optional class slot as the node information of the slot, and the optional class slot is assigned a corresponding optional word, where the optional The identification is used to indicate that the acoustic model of the optional slot is simultaneously speech-recognized with the acoustic model of the next slot in the same depth path during the identification process.
  • Such slots are typically used in simple situations where only one or two optional words are included in a single syntax path.
  • a predetermined grammar path is expanded into two or more.
  • the above slot $de can be identified as an optional class slot, that is, when constructing the syntax tree, the following two syntax statements "call three calls” and "three calls” are constructed.
  • the statement "call three calls” and "play three calls” will be supported at the same time, ie, for $SIL$action $hame$de$phone reserves the depth path of the syntax tree corresponding to the syntax path, and can simultaneously recognize the slot $de and the slot $phone to support two speech recognitions.
  • Add a jump identifier (_JMP&_TAG), that is, determine a slot corresponding to the application scenario, and assign at least one corresponding word to each slot includes:
  • Determining a jump class slot corresponding to the application scenario adding a jump identifier to the jump class slot as the node information of the slot, and assigning a corresponding word to the jump class slot, where the jump identifier is used
  • the jump to the designated slot is performed for voice recognition.
  • divide; $number 0
  • $SIL$number_LOOP$yunsuan_TAG1$number_LOOP_JMP1$SIL you can implement any number of calculation functions.
  • the recognition process is that the number 1 is first recognized in the digital slot $number_LOOP, then the + is recognized in the operation slot $yunsuan, and then jumped to the digital slot $number_LOOP to recognize the loop.
  • the number 56 then jump to the computation slot $yunsuan recognizes -, and so on, and so on, to identify the above formula.
  • this class of slots can be used to identify more complex syntax statements. Under some syntax statements, this class slot is equivalent to multiple optional slots. For example, the following two syntaxes are equivalent: $SIL$action$name$de_OPT$phone;$SIL$action$name_JMP2$de$phone_TAG2.
  • the slot, and assigning each slot a corresponding at least one word includes:
  • Determining an acoustic enhancement type slot corresponding to the application scene adding an acoustic reinforcement identifier to the acoustic reinforcement type slot as node information of the slot, and assigning a corresponding word to the acoustic enhancement type slot, wherein the acoustic enhancement indicator is used to indicate In the identification process, the acoustic recognition score of the acoustically enhanced groove is increased.
  • Such slots are generally applied when the syntax path of the syntax tree is very large (for example, several hundred), and the user may pay more attention to the detection rate of one of the predetermined syntax paths, and then the acoustic enhancement may be added to the slots of the syntax path.
  • the identification when the voice recognition is performed, the path is more likely to be detected due to the high score, so that the recognition accuracy of the path can be improved.
  • Add a language enhancement identifier such as $ime_LOOP_LM, that is, determine a slot corresponding to the application scenario, and assign at least one corresponding word to each slot includes:
  • Determining a language enhancement class slot corresponding to the application scenario adding a language enhancement identifier to the language enhancement class slot as node information of the slot, and assigning a corresponding word to the language enhancement class slot, wherein the language enhancement identifier is used to indicate In the recognition process, the language recognition score of the language enhancement class slot is increased.
  • the slot corresponding to the application scenario is determined, and at least one word is assigned to each slot, and the syntax tree is constructed based on the slot, and information such as an acoustic model of the word corresponding to the slot does not need to be stored repeatedly.
  • the slot index is in a different syntax path, so memory space and composition consumption can be saved when the composition process is completed. time.
  • the speech recognition rate can be increased by adding a marker to the slot.
  • FIG. 2 is a schematic structural diagram of a composition device of a speech recognition syntax tree according to Embodiment 2 of the present invention, as shown in FIG. 2, specifically including: a slot determination module 21, an acoustic model determination module 22, and a syntax tree construction module 23;
  • the slot determining module 21 is configured to determine a slot corresponding to the application scenario, and assign a corresponding at least one word to each slot;
  • the acoustic model module 22 is configured to form a grammar sentence according to a slot order in a predetermined grammar path and a word corresponding to the slot, and parse the voice of the word in the grammar sentence to determine a corresponding acoustic model;
  • the syntax tree construction module 23 is configured to construct a syntax tree according to a slot in a predetermined syntax path of the application scenario, where the slot corresponds to a node in the syntax tree, and the slot index of the slot is stored in the node, the slot index An acoustic model storage location for indexing corresponding words.
  • the speech recognition composition device described in this embodiment is used to perform the speech recognition composition method described in the above embodiments, and the technical principle and the generated technical effect are similar, and are not described here.
  • the device further comprising: a slot coupling determining module 24 and a slot connecting storage module 25;
  • the slot coupling determining module 24 is configured to determine a slot joint between the parent slot and the sub slot according to an order of the slots in the predetermined syntax path, and store an acoustic model of the slot coupling;
  • the slot connection storage module 25 is configured to store the slot join index of the slot connection in a syntax tree Among the nodes corresponding to the sub-slots, the slot join index is used to index the storage location of the acoustic model of the slot joint.
  • syntax tree construction module 23 is specifically configured to:
  • each predetermined syntax path of the application scenario Determining a starting slot of each predetermined syntax path of the application scenario as a root node of the syntax tree; forming each of the predetermined syntax paths into a depth path of the syntax tree, in a slot in the predetermined syntax path and in the depth path Corresponding to the node, and storing the slot index and the slot join index with the parent slot in the node, storing the end identifier at the tail node of each depth path; merging the nodes corresponding to the same slot in the same layer in each depth path .
  • the slot determining module 21 is specifically configured to:
  • Determining a data class slot corresponding to the application scenario adding a data identifier to the data class slot as the node information of the slot, and assigning a word in the setting database to the data class slot, where the setting database includes at least : Address book name library, address book phone number library, public phone library, or building name library.
  • the slot determining module 21 is specifically configured to:
  • Determining a loop class slot corresponding to the application scenario adding a loop identifier to the loop class slot as node information of the slot, and assigning at least two words to the loop class slot, wherein the loop identifier is used to indicate
  • the loop uses the acoustic model of the loop-like slot for word recognition.
  • the slot determining module 21 is specifically configured to:
  • An optional class slot corresponding to the application scenario is determined, an optional identifier is added to the optional class slot as the node information of the slot, and the optional class slot is assigned a corresponding optional word, where the optional The identification is used to indicate that the acoustic model of the optional slot is simultaneously speech-recognized with the acoustic model of the next slot in the same depth path during the identification process.
  • the slot determining module 21 is specifically configured to:
  • Determining a jump class slot corresponding to the application scenario adding a jump identifier to the jump class slot as the node information of the slot, and assigning a corresponding word to the jump class slot, where the jump identifier is used
  • the jump to the designated slot is performed for voice recognition.
  • the slot determining module 21 is specifically configured to:
  • Determining an acoustic enhancement type slot corresponding to the application scene adding an acoustic reinforcement identifier to the acoustic reinforcement type slot as node information of the slot, and assigning a corresponding word to the acoustic enhancement type slot, wherein the acoustic enhancement indicator is used to indicate In the identification process, the acoustic recognition score of the acoustically enhanced groove is increased.
  • the slot determining module 21 is specifically configured to:
  • Determining a language enhancement class slot corresponding to the application scenario adding a language enhancement identifier to the language enhancement class slot as node information of the slot, and assigning a corresponding word to the language enhancement class slot, wherein the language enhancement identifier is used to indicate In the recognition process, the language recognition score of the language enhancement class slot is increased.
  • the speech recognition composition device described in the above embodiments is also used to perform the speech recognition composition method described in the above embodiments.
  • the technical principle and the generated technical effects are similar, and are not described here.
  • FIG. 3 is a schematic diagram of a hardware structure of a terminal device for implementing composition of a speech recognition syntax tree according to Embodiment 3 of the present invention, where the terminal device includes one or more processors 31, a memory 32, and one or more modules, One or more modules (for example, the slot coupling determination module 21, the acoustic model determination module 22, the syntax tree construction module 23, the slot coupling determination module 24, and the slot coupling storage module 25 in the voice recognition composition device shown in FIG. 2) Stored in the memory 32; in FIG. 2, a processor 31 is taken as an example; the processor 31 and the memory 32 in the terminal device can be connected by a bus or other means, and the bus connection is taken as an example in FIG.
  • the acoustic model of the corresponding word stores the location.
  • the foregoing terminal device can perform the method provided by Embodiment 1 of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.
  • the processor 31 determines a slot connection between the parent slot and the sub slot according to the order of slots in the predetermined syntax path, and stores an acoustic model of the slot joint; storing the slot join index of the slot joint in Among the nodes corresponding to the sub-slots in the syntax tree, the slot join index is used to index the storage location of the acoustic model of the slot joint.
  • the processor 31 uses a starting slot of each predetermined syntax path of the application scenario as a root node of the syntax tree; forming each of the predetermined syntax paths into a depth path of the syntax tree, a predetermined syntax
  • the slot in the path corresponds to the node in the depth path, and the slot index and the slot join index with the parent slot are stored in the node, and the end identifier is stored at the tail node of each depth path;
  • the nodes corresponding to the same slot of the layer are merged.
  • the processor 31 determines a data class slot corresponding to the application scenario, adds a data identifier to the data class slot as the node information of the slot, and assigns a word in the setting database to the data class slot, where
  • the setting database includes at least: an address book name library, a directory telephone number library, a public telephone library, or a building name library.
  • the processor 31 determines a loop class slot corresponding to the application scenario, and is the loop class.
  • the slot adds a loop identifier as the node information of the slot, and assigns at least two words to the loop class slot, wherein the loop identifier is used to indicate that the loop adopts an acoustic model of the loop slot to perform words in the identification process. Word recognition.
  • the processor 31 determines an optional class slot corresponding to the application scenario, adds an optional identifier to the optional class slot as the node information of the slot, and assigns the corresponding optional slot to the optional class slot.
  • the processor 31 determines a jump class slot corresponding to the application scenario, adds a jump identifier to the jump class slot as node information of the slot, and assigns a corresponding word to the jump class slot.
  • the jump identifier is used to indicate that, in the identification process, after the identification of the jump class slot, jump to the designated slot for voice recognition.
  • the processor 31 determines an acoustic enhancement class slot corresponding to the application scenario, adds an acoustic enhancement flag to the acoustic enhancement class slot as node information of the slot, and assigns a corresponding word to the acoustic enhancement class slot, wherein
  • the acoustic enhancement indicator is used to indicate that an acoustic recognition score of the acoustic enhancement type slot is increased during the identification process.
  • the processor 31 determines a language enhancement class slot corresponding to the application scenario, adds a language enhancement identifier to the language enhancement class slot as the node information of the slot, and assigns a corresponding word to the language enhancement class slot, where The language enhancement identifier is used to indicate that the language recognition score of the language enhancement class slot is increased during the recognition process.
  • Embodiments of the present invention also provide a non-volatile computer storage medium storing one or more modules when one or more modules are executed by a speech recognition grammar
  • the device of the tree's composition method is executed, the device is caused to perform the following operations:
  • the method further preferably includes:
  • the slot join index of the slot join is stored in a node corresponding to the sub-slot in the syntax tree, the slot join index being used to index the storage location of the acoustic model of the slot join.
  • the syntax tree is preferably constructed according to a slot in a predetermined syntax path of the application scenario:
  • each of the predetermined syntax paths into a depth path of the syntax tree, the slots in the predetermined syntax path correspond to nodes in the depth path, and storing the slot index and the slot join index with the parent slot in the node, Store an end identifier at the tail node of each depth path;
  • the nodes corresponding to the same slot in the same layer in each depth path are merged.
  • the slot corresponding to the application scenario is determined, and the corresponding at least one word is assigned to each slot:
  • Determining a data class slot corresponding to the application scenario adding a data identifier to the data class slot as the node information of the slot, and assigning a word in the setting database to the data class slot, wherein the setting the database to Less include: address book name library, address book phone number library, public phone library, or building name library.
  • the slot corresponding to the application scenario is determined, and the corresponding at least one word is assigned to each slot:
  • Determining a loop class slot corresponding to the application scenario adding a loop identifier to the loop class slot as node information of the slot, and assigning at least two words to the loop class slot, wherein the loop identifier is used to indicate
  • the loop uses the acoustic model of the loop-like slot for word recognition.
  • the slot corresponding to the application scenario is determined, and the corresponding at least one word is assigned to each slot:
  • An optional class slot corresponding to the application scenario is determined, an optional identifier is added to the optional class slot as the node information of the slot, and the optional class slot is assigned a corresponding optional word, where the optional The identification is used to indicate that the acoustic model of the optional slot is simultaneously speech-recognized with the acoustic model of the next slot in the same depth path during the identification process.
  • the slot corresponding to the application scenario is determined, and the corresponding at least one word is assigned to each slot:
  • Determining a jump class slot corresponding to the application scenario adding a jump identifier to the jump class slot as the node information of the slot, and assigning a corresponding word to the jump class slot, where the jump identifier is used
  • the jump to the designated slot is performed for voice recognition.
  • the slot corresponding to the application scenario is determined, and the corresponding at least one word is assigned to each slot:
  • Determining an acoustic enhancement type slot corresponding to the application scene adding an acoustic reinforcement identifier to the acoustic reinforcement type slot as node information of the slot, and assigning a corresponding word to the acoustic enhancement type slot, wherein the acoustic enhancement indicator is used to indicate In the identification process, the acoustic recognition score of the acoustically enhanced groove is increased.
  • the slot corresponding to the application scenario is determined, and the corresponding at least one word is assigned to each slot:
  • Determining a language enhancement class slot corresponding to the application scenario adding a language enhancement identifier to the language enhancement class slot as node information of the slot, and assigning a corresponding word to the language enhancement class slot, wherein the language enhancement identifier is used to indicate In the recognition process, the language recognition score of the language enhancement class slot is increased.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Acoustics & Sound (AREA)
  • Human Computer Interaction (AREA)
  • Multimedia (AREA)
  • Artificial Intelligence (AREA)
  • Theoretical Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Telephonic Communication Services (AREA)
  • Machine Translation (AREA)

Abstract

一种语音识别语法树的构图方法、装置、终端设备及存储介质，该方法包括：确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词（11）；根据预定语法路径中的槽顺序和槽对应的字词，形成语法语句，并解析语法语句中字词的语音，确定对应的声学模型（12）；根据应用场景的预定语法路径中的槽构建语法树，其中，槽与语法树中的节点对应，且节点中存储该槽的槽索引，槽索引用于索引对应字词的声学模型存储位置（13）。该方法只需要确定与应用场景对应的槽，为每个槽赋予对应的至少一个字词槽中，通过构建与槽对应的语法树，在节点中存储该槽的槽索引，从而完成构图过程，能够大大节约内存空间和构图消耗时间。

Description

语音识别语法树的构图方法、装置、终端设备及存储介质
本专利申请要求于2015年07月29日提交的、申请号为201510455696.1、申请人为百度在线网络技术(北京)有限公司、发明名称为“语音识别语法树的构图方法及装置”的中国专利申请的优先权，该申请的全文以引用的方式并入本申请中。
技术领域
本发明实施例涉及语音识别技术领域，尤其涉及一种语音识别语法树的构图方法、装置、终端设备及存储介质。
背景技术
相对于联网语音识别的场景，同样有非常多的语音识别需求，需要在本地完成，例如用户使用语音输入进行拨打电话时，即可通过语音输入“打电话给张三”，移动终端接收到上述语音输入后，根据现有的语音识别技术即可匹配得到文字信息，从而完成通话操作。
本地语音识别也需要在识别之前对预识别的文本进行语法树构图，即建立用户输入语音的所有可能路径，在进行语音识别时，遍历整个构图空间，结合语音识别算法，查找与输入语音最佳的匹配路径，该路径会返回识别的最终结果。
目前，在构图时，大都采用权重有限状态机语音识别构图算法，该算法会将所有可能的路径穷举展开。例如，对于上述三条语句“打电话给张三”、“给张三打电话”和“拨打张三的电话”，在采用权重有限状态机语音识别构图算 法构图时，将会构建三条独立的路径，与语句“打电话给张三”、“给张三打电话”和“拨打张三的电话”分别对应。
由于构造的三条路径相互独立，且存在相同关键词“张三”或“电话”，需要重复存储其对应的声学模型和语音模型，因此，当数据量较大时，构图消耗的内存也较大，另外，为了使构图更加紧凑，必须进行多次遍历，寻找相同节点进行合并，导致构图消耗时间长。
发明内容
本发明实施例提供一种语音识别语法树的构图方法、装置、终端设备及存储介质，能够节约构图占用内存空间和构图消耗时间。
第一方面，本发明实施例提供了一种语音识别语法树的构图方法，包括：
确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词；
根据预定语法路径中的槽顺序和槽对应的字词，形成语法语句，并解析语法语句中字词的语音，确定对应的声学模型；
根据所述应用场景的预定语法路径中的槽构建语法树，其中，槽与语法树中的节点对应，且节点中存储该槽的槽索引，所述槽索引用于索引对应字词的声学模型存储位置。
第二方面，本发明实施例还提供一种语音识别语法树的构图装置，包括：
槽确定模块，用于确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词；
声学模型模块，用于根据预定语法路径中的槽顺序和槽对应的字词，形成语法语句，并解析语法语句中字词的语音，确定对应的声学模型；
语法树构建模块，用于根据所述应用场景的预定语法路径中的槽构建语法树，其中，槽与语法树中的节点对应，且节点中存储该槽的槽索引，所述槽索引用于索引对应字词的声学模型存储位置。
第三方面，本发明实施例还提供一种实现语音识别语法树的构图的终端设备，包括：
一个或者多个处理器；
存储器；
一个或者多个模块，所述一个或者多个模块存储在所述存储器中，当被所述一个或者多个处理器执行时，进行如下操作：
确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词；根据预定语法路径中的槽顺序和槽对应的字词，形成语法语句，并解析语法语句中字词的语音，确定对应的声学模型；根据所述应用场景的预定语法路径中的槽构建语法树，其中，槽与语法树中的节点对应，且节点中存储该槽的槽索引，所述槽索引用于索引对应字词的声学模型存储位置。
第四方面，本发明实施例还提供一种非易失性计算机存储介质，所述计算机存储介质存储有一个或者多个模块，当所述一个或者多个模块被一个执行语音识别语法树的构图方法的设备执行时，使得所述设备执行如下操作：
确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词；
根据预定语法路径中的槽顺序和槽对应的字词，形成语法语句，并解析语法语句中字词的语音，确定对应的声学模型；
根据所述应用场景的预定语法路径中的槽构建语法树，其中，槽与语法树中的节点对应，且节点中存储该槽的槽索引，所述槽索引用于索引对应字词的 声学模型存储位置。
本发明实施例，确定与应用场景对应的槽，为每个槽赋予对应的至少一个字词，基于槽构建语法树，而相同槽所对应字词的声学模型等信息无需反复存储，可通过槽索引至不同的语法路径中，所以完成构图过程时，能够节约内存空间和构图消耗时间。
附图说明
图1A为本发明实施例一提供的语音识别语法树的构图方法的流程示意图；
图1B为本发明实施例一提供的语音识别语法树的构图方法中的第一种语法树结构示意图；
图1C为本发明实施例一提供的语音识别语法树的构图方法中的第二种语法树结构示意图；
图2为本发明实施例二提供的语音识别语法树的构图装置的结构示意图；
图3为本发明实施例三提供的实现语音识别语法树的构图的终端设备的结构示意图。
具体实施方式
下面结合附图和实施例对本发明作进一步的详细说明。可以理解的是，此处所描述的具体实施例仅仅用于解释本发明，而非对本发明的限定。另外还需要说明的是，为了便于描述，附图中仅示出了与本发明相关的部分而非全部结构。
本发明实施例提供的语音识别语法树的构图方法的执行主体，可为本发明 实施例提供的语音识别语法树的构图装置，该装置可以集成于移动终端设备(例如，智能手机、平板电脑等)，也可以集成于服务器中，该语音识别语法树的构图装置可以采用硬件或软件实现。本发明实施例提供的语音识别语法树的构图方法尤为适用于本地语音识别，能够适应数量有限的语音指令的内容识别，但也可以适用于在线语音识别，相应增加预定语法路径和槽的数量即可，下面将结合实施例进行说明。
实施例一
图1A为本发明实施例一提供的语音识别语法树的构图方法的流程示意图，如图1A所示，具体包括：
S11、确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词；
其中，本实施例中的应用场景是用户控制终端操作的常用场景，典型的应用场景例如为，打电话、导航、发短信、以及搜索等。每个应用场景可能涉及至少一条控制指令，通常会涉及多条控制指令，且指令的实质内容往往相同。本发明实施例将以打电话应用场景为例进行说明。
例如，在打电话应用场景下，一般可涉及到的操作指令包括给张圆圆打电话、打电话给张圆圆、拨打张圆圆的电话等，则可根据人工经验或数据挖掘，确定该应用场景下的槽包括：＄SIL、＄to、＄action、＄de、＄name和＄phone，并为每个槽赋予与所述打电话应用场景相关联的至少一个字词：
＄SIL＝sil，该槽为特殊的起始槽，用于标注预定语法路径的起始点；
＄to＝给；
＄action＝打|拨打；
＄de＝的；
＄name＝张圆圆|张亚媛|朱大元|周小元；
＄phone＝电话；
其中，每个槽可用数组来表示，等号右边为与该槽所关联的字词信息，包括至少一个字词。
S12、根据预定语法路径中的槽顺序和槽对应的字词，形成语法语句，并解析语法语句中字词的语音，确定对应的声学模型；
其中，所述预定语法路径可预先设定，具体由上述步骤11中确定的槽按照一定语法顺序构成，按顺序读取槽中的字词，可形成语法语句。
具体的，针对上述打电话应用场景，假设预先定义如下三条语法路径：
(＄SIL＄to＄name＄action＄phone)
(＄SIL＄action＄phone＄to＄name)
(＄SIL＄action＄name＄de＄phone)
按照上述预定语法路径中的槽顺序读取槽中的字词，可对应形成多条语法语句，例如：
给张圆圆打电话
打电话给张圆圆
打张圆圆的电话
而后，对语法语句中的字词进行解析，得到对应的语音信息。例如，通过解析分别得到上述三条语法语句中各字词的语音信息，并确定对应的声学模型，以便于在语音识别中快速定位到对应的语法语句。例如确定的声学模型如下：
sil-g+ei g-ei+zh ei-zh+ang zh-ang+y ang-y+uang y-uang+y uang-y+uang  y-uang+d uang-d+a d-a+d a-d+ian d-ian+h ian-h+ua h-ua+sil；
sil-d+a d-a+d a-d+ian d-ian+h ian-h+ua h-ua+g ua-g+ei g-ei+zh ei-zh+ang zh-ang+y ang-y+uang y-uang+y uang-y+uang y-uang+sil；
sil-d+a d-a+zh a-zh+ang zh-ang+y ang-y+uang y-uang+y uang-y+uang y-uang+d uang-d+e d-e+d-e-d+ian d-ian+h ian-h+ua h-ua+sil。
S13、根据所述应用场景的预定语法路径中的槽构建语法树，其中，槽与语法树中的节点对应，且节点中存储该槽的槽索引，所述槽索引用于索引对应字词的声学模型存储位置。
例如，根据上述三条预定语法路径(＄SIL＄to＄name＄action＄phone)、(＄SIL＄action＄phone＄to＄name)和(＄SIL＄action＄name＄de＄phone)，可初步构建如图1B所示的语法树，每个节点中存储该槽的槽索引。＄SIL、＄to、＄action、＄de、＄name和＄phone即可以作为槽对应的槽索引，可以是指针或存储地址，指向该槽中字词的声学模型以及其他信息。
本实施例，确定与应用场景对应的槽，为每个槽赋予对应的至少一个字词，基于槽构建语法树，而相同槽所对应字词的声学模型等信息无需反复存储，可通过槽索引至不同的语法路径中，所以完成构图过程时，能够节约内存空间和构图消耗时间。
示例性的，在上述实施例的基础上，所述方法还包括：
按照预定语法路径中槽的顺序，确定父槽和子槽之间的槽联接，并存储所述槽联接的声学模型；
将所述槽联接的槽联接索引存储在语法树中子槽对应的节点中，所述槽联 接索引用于索引所述槽联接的声学模型的存储位置。
其中，所述槽联接表征了父槽和子槽之间的关联信息，在语法树构图时，以上述例子中的字词“打电话”为例，其中“打”的最后一个声学信息是d-a+*，“电话”的第一个声学信息是*-d+ian，*代表任意词的意思，但是单独的声学信息，无法确认*的内容，只有通过槽联接才能把他们展成：d-a+d(*被展成电话的声母)，a-d+ian(*被展成打的韵母)。
由于后文的发音会影响前文的发音，所以记录字词的声学上下文关系，有助于进行语音识别。所以对于“打电话”需构建“d-a+d”这样的声学模型。当槽中包括多个字词时，相邻槽的槽联接将包括多个声学模型。例如，＄to＄name的槽联接中将包括“给张圆圆”、“给李四”、“给王五”等多个声学模型。
如果多条预定语法路径中有重复的槽联接，则只需要将槽联接的声学模型存储一份即可，在构建语法树时，在子槽对应的节点中存储槽联接索引。这样可以节省重复存储槽联接声学模型的内存和构图消耗时间。
例如，对于上述预定语法路径(＄SIL＄to＄name＄action＄phone)中的父槽＄to和子槽＄name，和(＄SIL＄action＄phone＄to＄name)中的父槽＄to和子槽＄name，只需要存储预定语法路径(＄SIL＄to＄name＄action＄phone)中的父槽＄to和子槽＄name的槽联接即可，在存储(＄SIL＄action＄phone＄to＄name)中的父槽＄to和子槽＄name的槽联接时，由于＄to＄name的槽联接已经构建并存储，所以只需存储上述预定语法路径(＄SIL＄to＄name＄action＄phone)中的父槽＄to和子槽＄name的槽联接对应的槽联接索引即可。
在上述实施例的基础上，所述根据所述应用场景的预定语法路径中的槽构建语法树具体包括：
将所述应用场景的各预定语法路径的起始槽作为语法树的根节点；
将每条所述预定语法路径形成所述语法树的一条深度路径，预定语法路径中的槽与深度路径中的节点对应，且在节点中存储槽索引和与父槽之间的槽联接索引，在每条深度路径的尾节点存储结束标识符；
将各深度路径中位于相同层的相同槽对应的节点进行合并。
具体的，为进一步节省构树所占的内存空间，同样以上述实施例中打电话应用场景为例，如图1C所示，将所述各预定语法路径的SIL作为起始槽，形成语法树的根节点，且对于预定语法路径中相同层的相同槽对应的节点进行合并，例如，预定语法路径(＄SIL＄action＄phone＄to＄name)中的槽＄action和预定语法路径(＄SIL＄action＄name＄de＄phone)中的槽＄action，即可进行合并。或者，起始槽也可以为空槽，自动形成一空槽作为该应用场景下各预定语法路径的根节点。
在上述实施例的基础上，按照用户的说话习惯，可以总结出各预设语法路径中槽的特点，添加相应的标识，以便进一步节约构图资源。具体是添加如下至少一种标识，可以给槽添加一个标识，也可以同时添加多个：
添加资料标识(_CORE)，例如＄name_CORE，即，确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词包括：
确定与应用场景对应的资料类槽，为所述资料类槽添加资料标识作为槽的节点信息，且为资料类槽赋予设定资料库中的字词，其中，所述设定资料库至少包括：通信录姓名库、通信录电话号码库、公共电话库、或建筑物名称库。
其中，通信录姓名库、通信录电话号码库、公共电话库、或建筑物名称库均可以从移动终端中获取。
例如，当当前应用场景为打电话应用场景时，资料类槽包括通信录姓名库，具体可从终端的通信录中进行获取，例如包括：张圆圆、张亚媛、朱大元和周小元等。
添加循环标识(_LOOP)，例如＄number_LOOP，即，确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词包括：
确定与应用场景对应的循环类槽，为所述循环类槽添加循环标识作为槽的节点信息，且为所述循环类槽赋予对应的至少两个字词，其中，所述循环标识用于指示在识别过程中，循环采用循环类槽的声学模型进行字词识别。
例如，以打电话应用场景为例，可在该类槽中存储0-9中的数字，在语音识别时，用于循环识别用户输入的电话号码信息。可以通过在循环类槽之后设置静音槽＄SIL，当识别到出现静音时，则循环结束。
添加可选标识(_OPT)，即，确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词包括：
确定与应用场景对应的可选类槽，为所述可选类槽添加可选标识作为槽的节点信息，且为所述可选类槽赋予对应的可选字词，其中，所述可选标识用于指示在识别过程中，可选类槽的声学模型与同一深度路径中下一个槽的声学模型同时进行语音识别。
此类槽通常应用于简单的场合，即只在一条语法路径中只包含一个或两个可选的字词情况下使用。对此类槽进行构图时，会将一条预定语法路径展成两条或多条。例如，可将上述槽＄de标识为可选类槽，即在构建语法树时会构建如下两条语法语句“打张三的电话”和“打张三电话”对应的语法路径。在语音识别时，将同时支持语句“打张三的电话”和“打张三电话”，即，对于＄SIL＄action ＄hame＄de＄phone预定语法路径所对应语法树的深度路径中，可以将槽＄de和槽＄phone同时进行语音识别，即可支持两条语音识别。
添加跳转标识(_JMP&_TAG)，即，确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词包括：
确定与应用场景对应的跳转类槽，为所述跳转类槽添加跳转标识作为槽的节点信息，且为所述跳转类槽赋予对应的字词，其中，所述跳转标识用于指示在识别过程中，跳转类槽的识别之后跳转至指定槽进行语音识别。
例如，对于如下运算场景，将运算槽＄yunsuan和数字槽＄number设置跳转标识，即＄yunsuan_TAG1＄number_LOOP_JMP1；其中，＄yunsuan＝加|减|乘|除；＄number＝0|1|2|3|4|5|6|7|8|9；对于如下输入语句进行语音识别时，(＄SIL＄number_LOOP＄yunsuan_TAG1＄number_LOOP_JMP1＄SIL)，可以实现任意多的运算功能。
例如对于语句1+56-45/324*1000，其识别过程为，首先在数字槽＄number_LOOP识别出数字1，然后在运算槽＄yunsuan识别出+，然后跳转至数字槽＄number_LOOP循环识别出数字56，然后跳转至运算槽＄yunsuan识别出-，等等，依次类推，即可识别出上述计算公式。
其中，在某个应用场景下，人工比较容易获知与该场景相关的语境，可根据相关语句采用人工添加的方式对预定语法路径中的指定槽添加跳转标识。该类槽可用于识别较为复杂的语法语句。在某些语法语句下，该类槽等同于多个可选槽。例如，以下2个语法是等效的：＄SIL＄action＄name＄de_OPT＄phone；＄SIL＄action＄name_JMP2＄de＄phone_TAG2。
添加声学加强标识(_WGT)，例如＄kwd_WGT，即，确定与应用场景对应 的槽，且为每个槽赋予对应的至少一个字词包括：
确定与应用场景对应的声学加强类槽，为所述声学加强类槽添加声学加强标识作为槽的节点信息，且为声学加强类槽赋予对应的字词，其中，所述声学加强标识用于指示在识别过程中，增加声学加强类槽的声学识别得分。
此类槽一般应用于当语法树的语法路径非常多的情况(例如几百条)，用户可能更关注其中某条预定语法路径的检出率，则可以对该语法路径的槽都添加声学加强标识，则在进行语音识别时，该路径会由于得分高而更容易被检出，从而能够提高该路径的识别精度。
添加语言加强标识(_LM)，例如＄ime_LOOP_LM，即，确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词包括：
确定与应用场景对应的语言加强类槽，为所述语言加强类槽添加语言加强标识作为槽的节点信息，且为语言加强类槽赋予对应的字词，其中，所述语言加强标识用于指示在识别过程中，增加语言加强类槽的语言识别得分。
例如，对于建立的如下槽＄ime＝井|经|冈山；当用户输入语音“井冈山”，首先确定该语音对应的语法路径，即(＄SIL＄ime_LOOP＄SIL)，根据声学模型可识别出对应的字词“经冈山”和“井冈山”，由于他们的声学模型得分一样，所以将无法区分用户需要的“井冈山”。但是，通过引入语言模型得分，在语言模型中可知“井冈山”对应的语言模型得分要高于“经冈山”对应的语言模型得分，从而快速准确的匹配到用户需要有的结果。上述各实施例同样通过确定与应用场景对应的槽，为每个槽赋予对应的至少一个字词，基于槽构建语法树，而相同槽所对应字词的声学模型等信息无需反复存储，可通过槽索引至不同的语法路径中，所以完成构图过程时，能够节约内存空间和构图消耗 时间。
另外，通过对槽进行添加标识，可提高语音识别速率。
实施例二
图2为本发明实施例二提供的语音识别语法树的构图装置的结构示意图，如图2所示，具体包括：槽确定模块21、声学模型确定模块22和语法树构建模块23；
所述槽确定模块21用于确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词；
所述声学模型模块22用于根据预定语法路径中的槽顺序和槽对应的字词，形成语法语句，并解析语法语句中字词的语音，确定对应的声学模型；
所述语法树构建模块23用于根据所述应用场景的预定语法路径中的槽构建语法树，其中，槽与语法树中的节点对应，且节点中存储该槽的槽索引，所述槽索引用于索引对应字词的声学模型存储位置。
本实施例所述的语音识别构图装置用于执行上述各实施例所述的语音识别构图方法，其技术原理和产生的技术效果类似，这里不再累述。
示例性的，在上述实施例的基础上，所述装置，其特征在于，还包括：槽联接确定模块24和槽联接存储模块25；
所述槽联接确定模块24用于按照预定语法路径中槽的顺序，确定父槽和子槽之间的槽联接，并存储所述槽联接的声学模型；
所述槽联接存储模块25用于将所述槽联接的槽联接索引存储在语法树中 子槽对应的节点中，所述槽联接索引用于索引所述槽联接的声学模型的存储位置。
示例性的，在上述实施例的基础上，所述语法树构建模块23具体用于：
将所述应用场景的各预定语法路径的起始槽作为语法树的根节点；将每条所述预定语法路径形成所述语法树的一条深度路径，预定语法路径中的槽与深度路径中的节点对应，且在节点中存储槽索引和与父槽之间的槽联接索引，在每条深度路径的尾节点存储结束标识符；将各深度路径中位于相同层的相同槽对应的节点进行合并。
示例性的，在上述实施例的基础上，所述槽确定模块21具体用于：
确定与应用场景对应的资料类槽，为所述资料类槽添加资料标识作为槽的节点信息，且为资料类槽赋予设定资料库中的字词，其中，所述设定资料库至少包括：通信录姓名库、通信录电话号码库、公共电话库、或建筑物名称库。
示例性的，在上述实施例的基础上，所述槽确定模块21具体用于：
确定与应用场景对应的循环类槽，为所述循环类槽添加循环标识作为槽的节点信息，且为所述循环类槽赋予对应的至少两个字词，其中，所述循环标识用于指示在识别过程中，循环采用循环类槽的声学模型进行字词识别。
示例性的，在上述实施例的基础上，所述槽确定模块21具体用于：
确定与应用场景对应的可选类槽，为所述可选类槽添加可选标识作为槽的节点信息，且为所述可选类槽赋予对应的可选字词，其中，所述可选标识用于指示在识别过程中，可选类槽的声学模型与同一深度路径中下一个槽的声学模型同时进行语音识别。
示例性的，在上述实施例的基础上，所述槽确定模块21具体用于：
确定与应用场景对应的跳转类槽，为所述跳转类槽添加跳转标识作为槽的节点信息，且为所述跳转类槽赋予对应的字词，其中，所述跳转标识用于指示在识别过程中，跳转类槽的识别之后跳转至指定槽进行语音识别。
示例性的，在上述实施例的基础上，所述槽确定模块21具体用于：
确定与应用场景对应的声学加强类槽，为所述声学加强类槽添加声学加强标识作为槽的节点信息，且为声学加强类槽赋予对应的字词，其中，所述声学加强标识用于指示在识别过程中，增加声学加强类槽的声学识别得分。
示例性的，在上述实施例的基础上，所述槽确定模块21具体用于：
确定与应用场景对应的语言加强类槽，为所述语言加强类槽添加语言加强标识作为槽的节点信息，且为语言加强类槽赋予对应的字词，其中，所述语言加强标识用于指示在识别过程中，增加语言加强类槽的语言识别得分。
上述各实施例所述的语音识别构图装置同样用于执行上述各实施例所述的语音识别构图方法，其技术原理和产生的技术效果类似，这里不再累述。
实施例三
图3为本发明实施例三提供的一种实现语音识别语法树的构图的终端设备的硬件结构示意图，该终端设备包括一个或多个处理器31、存储器32，一个或者多个模块，所述一个或者多个模块(例如，附图2所示的语音识别构图装置中的槽联接确定模块21、声学模型确定模块22、语法树构建模块23、槽联接确定模块24和槽联接存储模块25)存储在所述存储器32中；图2中以一个处理器31为例；终端设备中的处理器31和存储器32可以通过总线或其他方式连接，图2中以通过总线连接为例。
当被所述一个或者多个处理器31执行时，进行如下操作：
确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词；根据预定语法路径中的槽顺序和槽对应的字词，形成语法语句，并解析语法语句中字词的语音，确定对应的声学模型；根据所述应用场景的预定语法路径中的槽构建语法树，其中，槽与语法树中的节点对应，且节点中存储该槽的槽索引，所述槽索引用于索引对应字词的声学模型存储位置。
上述终端设备可执行本发明实施例一所提供的方法，具备执行方法相应的功能模块和有益效果。
示例性的，所述处理器31按照预定语法路径中槽的顺序，确定父槽和子槽之间的槽联接，并存储所述槽联接的声学模型；将所述槽联接的槽联接索引存储在语法树中子槽对应的节点中，所述槽联接索引用于索引所述槽联接的声学模型的存储位置。
示例性的，所述处理器31将所述应用场景的各预定语法路径的起始槽作为语法树的根节点；将每条所述预定语法路径形成所述语法树的一条深度路径，预定语法路径中的槽与深度路径中的节点对应，且在节点中存储槽索引和与父槽之间的槽联接索引，在每条深度路径的尾节点存储结束标识符；将各深度路径中位于相同层的相同槽对应的节点进行合并。
示例性的，所述处理器31确定与应用场景对应的资料类槽，为所述资料类槽添加资料标识作为槽的节点信息，且为资料类槽赋予设定资料库中的字词，其中，所述设定资料库至少包括：通信录姓名库、通信录电话号码库、公共电话库、或建筑物名称库。
示例性的，所述处理器31确定与应用场景对应的循环类槽，为所述循环类 槽添加循环标识作为槽的节点信息，且为所述循环类槽赋予对应的至少两个字词，其中，所述循环标识用于指示在识别过程中，循环采用循环类槽的声学模型进行字词识别。
示例性的，所述处理器31确定与应用场景对应的可选类槽，为所述可选类槽添加可选标识作为槽的节点信息，且为所述可选类槽赋予对应的可选字词，其中，所述可选标识用于指示在识别过程中，可选类槽的声学模型与同一深度路径中下一个槽的声学模型同时进行语音识别。
示例性的，所述处理器31确定与应用场景对应的跳转类槽，为所述跳转类槽添加跳转标识作为槽的节点信息，且为所述跳转类槽赋予对应的字词，其中，所述跳转标识用于指示在识别过程中，跳转类槽的识别之后跳转至指定槽进行语音识别。
示例性的，所述处理器31确定与应用场景对应的声学加强类槽，为所述声学加强类槽添加声学加强标识作为槽的节点信息，且为声学加强类槽赋予对应的字词，其中，所述声学加强标识用于指示在识别过程中，增加声学加强类槽的声学识别得分。
示例性的，所述处理器31确定与应用场景对应的语言加强类槽，为所述语言加强类槽添加语言加强标识作为槽的节点信息，且为语言加强类槽赋予对应的字词，其中，所述语言加强标识用于指示在识别过程中，增加语言加强类槽的语言识别得分。
实施例四
本发明实施例还提供一种非易失性计算机存储介质，所述计算机存储介质存储有一个或者多个模块，当所述一个或者多个模块被一个执行语音识别语法 树的构图方法的设备执行时，使得所述设备执行如下操作：
确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词；
根据预定语法路径中的槽顺序和槽对应的字词，形成语法语句，并解析语法语句中字词的语音，确定对应的声学模型；
根据所述应用场景的预定语法路径中的槽构建语法树，其中，槽与语法树中的节点对应，且节点中存储该槽的槽索引，所述槽索引用于索引对应字词的声学模型存储位置。
上述存储介质中存储的模块被所述设备所执行时，该方法还优选包括：
按照预定语法路径中槽的顺序，确定父槽和子槽之间的槽联接，并存储所述槽联接的声学模型；
将所述槽联接的槽联接索引存储在语法树中子槽对应的节点中，所述槽联接索引用于索引所述槽联接的声学模型的存储位置。
上述存储介质中存储的模块被所述设备所执行时，根据所述应用场景的预定语法路径中的槽构建语法树优选为：
将所述应用场景的各预定语法路径的起始槽作为语法树的根节点；
将每条所述预定语法路径形成所述语法树的一条深度路径，预定语法路径中的槽与深度路径中的节点对应，且在节点中存储槽索引和与父槽之间的槽联接索引，在每条深度路径的尾节点存储结束标识符；
将各深度路径中位于相同层的相同槽对应的节点进行合并。
上述存储介质中存储的模块被所述设备所执行时，确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词优选为：
确定与应用场景对应的资料类槽，为所述资料类槽添加资料标识作为槽的节点信息，且为资料类槽赋予设定资料库中的字词，其中，所述设定资料库至 少包括：通信录姓名库、通信录电话号码库、公共电话库、或建筑物名称库。
上述存储介质中存储的模块被所述设备所执行时，确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词优选为：
确定与应用场景对应的循环类槽，为所述循环类槽添加循环标识作为槽的节点信息，且为所述循环类槽赋予对应的至少两个字词，其中，所述循环标识用于指示在识别过程中，循环采用循环类槽的声学模型进行字词识别。
上述存储介质中存储的模块被所述设备所执行时，确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词优选为：
确定与应用场景对应的可选类槽，为所述可选类槽添加可选标识作为槽的节点信息，且为所述可选类槽赋予对应的可选字词，其中，所述可选标识用于指示在识别过程中，可选类槽的声学模型与同一深度路径中下一个槽的声学模型同时进行语音识别。
上述存储介质中存储的模块被所述设备所执行时，确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词优选为：
确定与应用场景对应的跳转类槽，为所述跳转类槽添加跳转标识作为槽的节点信息，且为所述跳转类槽赋予对应的字词，其中，所述跳转标识用于指示在识别过程中，跳转类槽的识别之后跳转至指定槽进行语音识别。
上述存储介质中存储的模块被所述设备所执行时，确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词优选为：
确定与应用场景对应的声学加强类槽，为所述声学加强类槽添加声学加强标识作为槽的节点信息，且为声学加强类槽赋予对应的字词，其中，所述声学加强标识用于指示在识别过程中，增加声学加强类槽的声学识别得分。
上述存储介质中存储的模块被所述设备所执行时，确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词优选为：
确定与应用场景对应的语言加强类槽，为所述语言加强类槽添加语言加强标识作为槽的节点信息，且为语言加强类槽赋予对应的字词，其中，所述语言加强标识用于指示在识别过程中，增加语言加强类槽的语言识别得分。
注意，上述仅为本发明的较佳实施例及所运用技术原理。本领域技术人员会理解，本发明不限于这里所述的特定实施例，对本领域技术人员来说能够进行各种明显的变化、重新调整和替代而不会脱离本发明的保护范围。因此，虽然通过以上实施例对本发明进行了较为详细的说明，但是本发明不仅仅限于以上实施例，在不脱离本发明构思的情况下，还可以包括更多其他等效实施例，而本发明的范围由所附的权利要求范围决定。

Claims (20)

  1. 一种语音识别语法树的构图方法，其特征在于，包括：
    确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词；
    根据预定语法路径中的槽顺序和槽对应的字词，形成语法语句，并解析语法语句中字词的语音，确定对应的声学模型；
    根据所述应用场景的预定语法路径中的槽构建语法树，其中，槽与语法树中的节点对应，且节点中存储该槽的槽索引，所述槽索引用于索引对应字词的声学模型存储位置。
  2. 根据权利要求1所述的方法，其特征在于，还包括：
    按照预定语法路径中槽的顺序，确定父槽和子槽之间的槽联接，并存储所述槽联接的声学模型；
    将所述槽联接的槽联接索引存储在语法树中子槽对应的节点中，所述槽联接索引用于索引所述槽联接的声学模型的存储位置。
  3. 根据权利要求1或2所述的方法，其特征在于，根据所述应用场景的预定语法路径中的槽构建语法树包括：
    将所述应用场景的各预定语法路径的起始槽作为语法树的根节点；
    将每条所述预定语法路径形成所述语法树的一条深度路径，预定语法路径中的槽与深度路径中的节点对应，且在节点中存储槽索引和与父槽之间的槽联接索引，在每条深度路径的尾节点存储结束标识符；
    将各深度路径中位于相同层的相同槽对应的节点进行合并。
  4. 根据权利要求1～3任一项所述的方法，其特征在于，确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词包括：
    确定与应用场景对应的资料类槽，为所述资料类槽添加资料标识作为槽的 节点信息，且为资料类槽赋予设定资料库中的字词，其中，所述设定资料库至少包括：通信录姓名库、通信录电话号码库、公共电话库、或建筑物名称库。
  5. 根据权利要求1～4任一项所述的方法，其特征在于，确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词包括：
    确定与应用场景对应的循环类槽，为所述循环类槽添加循环标识作为槽的节点信息，且为所述循环类槽赋予对应的至少两个字词，其中，所述循环标识用于指示在识别过程中，循环采用循环类槽的声学模型进行字词识别。
  6. 根据权利要求1～5任一项所述的方法，其特征在于，确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词包括：
    确定与应用场景对应的可选类槽，为所述可选类槽添加可选标识作为槽的节点信息，且为所述可选类槽赋予对应的可选字词，其中，所述可选标识用于指示在识别过程中，可选类槽的声学模型与同一深度路径中下一个槽的声学模型同时进行语音识别。
  7. 根据权利要求1～6任一项所述的方法，其特征在于，确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词包括：
    确定与应用场景对应的跳转类槽，为所述跳转类槽添加跳转标识作为槽的节点信息，且为所述跳转类槽赋予对应的字词，其中，所述跳转标识用于指示在识别过程中，跳转类槽的识别之后跳转至指定槽进行语音识别。
  8. 根据权利要求1～7任一项所述的方法，其特征在于，确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词包括：
    确定与应用场景对应的声学加强类槽，为所述声学加强类槽添加声学加强标识作为槽的节点信息，且为声学加强类槽赋予对应的字词，其中，所述声学 加强标识用于指示在识别过程中，增加声学加强类槽的声学识别得分。
  9. 根据权利要求1～8任一项所述的方法，其特征在于，确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词包括：
    确定与应用场景对应的语言加强类槽，为所述语言加强类槽添加语言加强标识作为槽的节点信息，且为语言加强类槽赋予对应的字词，其中，所述语言加强标识用于指示在识别过程中，增加语言加强类槽的语言识别得分。
  10. 一种语音识别语法树的构图装置，其特征在于，包括：
    槽确定模块，用于确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词；
    声学模型确定模块，用于根据预定语法路径中的槽顺序和槽对应的字词，形成语法语句，并解析语法语句中字词的语音，确定对应的声学模型；
    语法树构建模块，用于根据所述应用场景的预定语法路径中的槽构建语法树，其中，槽与语法树中的节点对应，且节点中存储该槽的槽索引，所述槽索引用于索引对应字词的声学模型存储位置。
  11. 根据权利要求10所述的装置，其特征在于，还包括：
    槽联接确定模块，用于按照预定语法路径中槽的顺序，确定父槽和子槽之间的槽联接，并存储所述槽联接的声学模型；
    槽联接存储模块，用于将所述槽联接的槽联接索引存储在语法树中子槽对应的节点中，所述槽联接索引用于索引所述槽联接的声学模型的存储位置。
  12. 根据权利要求10或11所述的装置，其特征在于，所述语法树构建模块具体用于：
    将所述应用场景的各预定语法路径的起始槽作为语法树的根节点；将每条 所述预定语法路径形成所述语法树的一条深度路径，预定语法路径中的槽与深度路径中的节点对应，且在节点中存储槽索引和与父槽之间的槽联接索引，在每条深度路径的尾节点存储结束标识符；将各深度路径中位于相同层的相同槽对应的节点进行合并。
  13. 根据权利要求10～12任一项所述的装置，其特征在于，所述槽确定模块具体用于：
    确定与应用场景对应的资料类槽，为所述资料类槽添加资料标识作为槽的节点信息，且为资料类槽赋予设定资料库中的字词，其中，所述设定资料库至少包括：通信录姓名库、通信录电话号码库、公共电话库、或建筑物名称库。
  14. 根据权利要求10～13任一项所述的装置，其特征在于，所述槽确定模块具体用于：
    确定与应用场景对应的循环类槽，为所述循环类槽添加循环标识作为槽的节点信息，且为所述循环类槽赋予对应的至少两个字词，其中，所述循环标识用于指示在识别过程中，循环采用循环类槽的声学模型进行字词识别。
  15. 根据权利要求10～14任一项所述的装置，其特征在于，所述槽确定模块具体用于：
    确定与应用场景对应的可选类槽，为所述可选类槽添加可选标识作为槽的节点信息，且为所述可选类槽赋予对应的可选字词，其中，所述可选标识用于指示在识别过程中，可选类槽的声学模型与同一深度路径中下一个槽的声学模型同时进行语音识别。
  16. 根据权利要求10～15任一项所述的装置，其特征在于，所述槽确定模块具体用于：
    确定与应用场景对应的跳转类槽，为所述跳转类槽添加跳转标识作为槽的节点信息，且为所述跳转类槽赋予对应的字词，其中，所述跳转标识用于指示在识别过程中，跳转类槽的识别之后跳转至指定槽进行语音识别。
  17. 根据权利要求10～16任一项所述的装置，其特征在于，所述槽确定模块具体用于：
    确定与应用场景对应的声学加强类槽，为所述声学加强类槽添加声学加强标识作为槽的节点信息，且为声学加强类槽赋予对应的字词，其中，所述声学加强标识用于指示在识别过程中，增加声学加强类槽的声学识别得分。
  18. 根据权利要求10～17任一项所述的装置，其特征在于，所述槽确定模块具体用于：
    确定与应用场景对应的语言加强类槽，为所述语言加强类槽添加语言加强标识作为槽的节点信息，且为语言加强类槽赋予对应的字词，其中，所述语言加强标识用于指示在识别过程中，增加语言加强类槽的语言识别得分。
  19. 一种实现语音识别语法树的构图的终端设备，其特征在于，包括：
    一个或者多个处理器；
    存储器；
    一个或者多个模块，所述一个或者多个模块存储在所述存储器中，当被所述一个或者多个处理器执行时，进行如下操作：
    确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词；根据预定语法路径中的槽顺序和槽对应的字词，形成语法语句，并解析语法语句中字词的语音，确定对应的声学模型；根据所述应用场景的预定语法路径中的槽构建语法树，其中，槽与语法树中的节点对应，且节点中存储该槽的槽索引，所 述槽索引用于索引对应字词的声学模型存储位置。
  20. 一种非易失性计算机存储介质，所述计算机存储介质存储有一个或者多个模块，其特征在于，当所述一个或者多个模块被一个执行语音识别语法树的构图方法的设备执行时，使得所述设备执行如下操作：
    确定与应用场景对应的槽，且为每个槽赋予对应的至少一个字词；
    根据预定语法路径中的槽顺序和槽对应的字词，形成语法语句，并解析语法语句中字词的语音，确定对应的声学模型；
    根据所述应用场景的预定语法路径中的槽构建语法树，其中，槽与语法树中的节点对应，且节点中存储该槽的槽索引，所述槽索引用于索引对应字词的声学模型存储位置。
PCT/CN2015/096624 2015-07-29 2015-12-08 语音识别语法树的构图方法、装置、终端设备及存储介质 WO2017016126A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201510455696.1 2015-07-29
CN201510455696.1A CN105161095B (zh) 2015-07-29 2015-07-29 语音识别语法树的构图方法及装置

Publications (1)

Publication Number Publication Date
WO2017016126A1 true WO2017016126A1 (zh) 2017-02-02

Family

ID=54801928

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2015/096624 WO2017016126A1 (zh) 2015-07-29 2015-12-08 语音识别语法树的构图方法、装置、终端设备及存储介质

Country Status (2)

Country Link
CN (1) CN105161095B (zh)
WO (1) WO2017016126A1 (zh)

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105786971B (zh) * 2016-02-02 2019-06-11 宋继华 一种面向国际汉语教学的语法点识别方法
CN108922531B (zh) * 2018-07-26 2020-10-27 腾讯科技（北京）有限公司 槽位识别方法、装置、电子设备及存储介质
CN109087645B (zh) * 2018-10-24 2021-04-30 科大讯飞股份有限公司 一种解码网络生成方法、装置、设备及可读存储介质
CN110473551B (zh) * 2019-09-10 2022-07-08 北京百度网讯科技有限公司 一种语音识别方法、装置、电子设备及存储介质
CN112749550B (zh) * 2020-07-14 2023-02-03 腾讯科技（深圳）有限公司 数据存储方法、装置、计算机设备及存储介质
CN112466292B (zh) 2020-10-27 2023-08-04 北京百度网讯科技有限公司 语言模型的训练方法、装置和电子设备
CN112466291B (zh) * 2020-10-27 2023-05-05 北京百度网讯科技有限公司 语言模型的训练方法、装置和电子设备

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1140870A (zh) * 1995-04-07 1997-01-22 索尼公司 语言识别方法和装置及语言翻译系统
US5699456A (en) * 1994-01-21 1997-12-16 Lucent Technologies Inc. Large vocabulary connected speech recognition system and method of language representation using evolutional grammar to represent context free grammars
JPH10254481A (ja) * 1997-03-14 1998-09-25 Nippon Telegr & Teleph Corp <Ntt> 音声認識方法
CN1670728A (zh) * 2003-10-23 2005-09-21 微软公司 具有标记数据的完全形式词典及其构建和使用方法
CN101271689A (zh) * 2007-03-20 2008-09-24 国际商业机器公司 用数字化语音中呈现的词来索引数字化语音的方法和装置
CN102543071A (zh) * 2011-12-16 2012-07-04 安徽科大讯飞信息科技股份有限公司 用于移动设备的语音识别系统和方法

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN100473095C (zh) * 2004-01-20 2009-03-25 联想(北京)有限公司 一种实现语音交互应用场景方法
US7529657B2 (en) * 2004-09-24 2009-05-05 Microsoft Corporation Configurable parameters for grammar authoring for speech recognition and natural language understanding
GB0513820D0 (en) * 2005-07-06 2005-08-10 Ibm Distributed voice recognition system and method
CN102693237B (zh) * 2011-03-24 2014-09-10 中国科学院声学研究所 一种网页内容适配封装系统及方法
CN103544154A (zh) * 2012-07-11 2014-01-29 神州数码信息系统有限公司 一种数据格式转换的方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5699456A (en) * 1994-01-21 1997-12-16 Lucent Technologies Inc. Large vocabulary connected speech recognition system and method of language representation using evolutional grammar to represent context free grammars
CN1140870A (zh) * 1995-04-07 1997-01-22 索尼公司 语言识别方法和装置及语言翻译系统
JPH10254481A (ja) * 1997-03-14 1998-09-25 Nippon Telegr & Teleph Corp <Ntt> 音声認識方法
CN1670728A (zh) * 2003-10-23 2005-09-21 微软公司 具有标记数据的完全形式词典及其构建和使用方法
CN101271689A (zh) * 2007-03-20 2008-09-24 国际商业机器公司 用数字化语音中呈现的词来索引数字化语音的方法和装置
CN102543071A (zh) * 2011-12-16 2012-07-04 安徽科大讯飞信息科技股份有限公司 用于移动设备的语音识别系统和方法

Also Published As

Publication number Publication date
CN105161095B (zh) 2017-03-22
CN105161095A (zh) 2015-12-16

Similar Documents

Publication Publication Date Title
WO2017016126A1 (zh) 语音识别语法树的构图方法、装置、终端设备及存储介质
JP2021018797A (ja) 対話の交互方法、装置、コンピュータ可読記憶媒体、及びプログラム
WO2017166650A1 (zh) 语音识别方法及装置
KR101066741B1 (ko) 컴퓨터 시스템과 동적으로 상호작용하기 위한 컴퓨터 구현 방법, 시스템, 및 컴퓨터 판독가능 기록 매체
KR101042119B1 (ko) 음성 이해 시스템, 및 컴퓨터 판독가능 기록 매체
KR101963915B1 (ko) 강화된 대화 이해 아키텍처
KR101418163B1 (ko) 컨텍스트 정보를 이용한 음성 인식 복구
WO2015096564A1 (zh) 一种在线语音翻译方法和装置
CN109325091B (zh) 兴趣点属性信息的更新方法、装置、设备及介质
US10754885B2 (en) System and method for visually searching and debugging conversational agents of electronic devices
JP6726354B2 (ja) 訂正済みタームを使用する音響モデルトレーニング
US9594744B2 (en) Speech transcription including written text
WO2017166631A1 (zh) 语音信号处理方法、装置和电子设备
WO2014183373A1 (en) Systems and methods for voice identification
US20190042185A1 (en) Flexible voice-based information retrieval system for virtual assistant
TW201606750A (zh) 使用外國字文法的語音辨識
JP7158217B2 (ja) 音声認識方法、装置及びサーバ
US11605385B2 (en) Project issue tracking via automated voice recognition
JP2015176099A (ja) 対話システム構築支援装置、方法、及びプログラム
JP2014106523A (ja) 音声入力対応装置及び音声入力対応プログラム
JP2011232619A (ja) 音声認識装置および音声認識方法
US20220093103A1 (en) Method, system, and computer-readable recording medium for managing text transcript and memo for audio file
KR20200011198A (ko) 대화형 메시지 구현 방법, 장치 및 프로그램
WO2020156342A1 (zh) 语音识别方法、装置、电子设备及存储介质
KR20190115405A (ko) 검색 방법 및 이 방법을 적용하는 전자 장치

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: 15899476

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: 15899476

Country of ref document: EP

Kind code of ref document: A1