(19)
(11)EP 3 163 568 B1

(12)EUROPEAN PATENT SPECIFICATION

(45)Mention of the grant of the patent:
06.05.2020 Bulletin 2020/19

(21)Application number: 16188500.9

(22)Date of filing:  13.09.2016
(51)Int. Cl.: 
G10L 15/04  (2013.01)
G10L 15/22  (2006.01)

(54)

SPEECH ENDPOINTING

ENDPUNKTBESTIMMUNG VON SPRACHE

POINT FINAL DE LA PAROLE


(84)Designated Contracting States:
AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

(30)Priority: 19.10.2015 US 201562243463 P
27.10.2015 US 201514923637
29.06.2016 US 201615196663

(43)Date of publication of application:
03.05.2017 Bulletin 2017/18

(73)Proprietor: Google LLC
Mountain View, CA 94043 (US)

(72)Inventors:
  • TADPATRIKAR, Siddhi
    Mountain View, CA 94043 (US)
  • BUCHANAN, Michael
    Mountain View, CA 94043 (US)
  • GUPTA, Pravir Kumar
    Mountain View, CA 94043 (US)

(74)Representative: Watkin, Timothy Lawrence Harvey et al
Marks & Clerk LLP Fletcher House The Oxford Science Park Heatley Road
Oxford OX4 4GE
Oxford OX4 4GE (GB)


(56)References cited: : 
WO-A1-2015/073071
US-A1- 2015 206 544
US-A1- 2006 149 558
  
      
    Note: Within nine months from the publication of the mention of the grant of the European patent, any person may give notice to the European Patent Office of opposition to the European patent granted. Notice of opposition shall be filed in a written reasoned statement. It shall not be deemed to have been filed until the opposition fee has been paid. (Art. 99(1) European Patent Convention).


    Description

    TECHNICAL FIELD



    [0001] This disclosure generally relates to speech recognition, and one particular implementation relates to endpointing speech.

    BACKGROUND



    [0002] Natural language processing systems typically use endpointers to determine when a user has started and finished speaking. Some traditional endpointers evaluate the duration of pauses between words in determining when an utterance begins or ends. For instance, if a user says "what is <long pause> for dinner," a traditional endpointer may segment the voice input at the long pause, and may instruct the natural language processing system to attempt to process the incomplete phrase "what is," instead of the complete phrase "what is for dinner." If an endpointer designates an incorrect beginning or ending point for a voice input, the result of processing the voice input using the natural language processing system may be inaccurate or undesirable. US 2006/149558 A1 relates to an apparatus for collecting data from a plurality of diverse data sources, the diverse data sources generating input data selected from the group including text, audio, and graphics, the diverse data sources selected from the group including real-time and recorded, human and mechanically-generated audio, single-speaker and multispeaker.

    SUMMARY



    [0003] Different users may have different comfort levels with using voice input on their mobile devices. Some users may use the voice input features frequently and be able to form voice queries without hesitating during speaking. Other users may not use the voice input features as often or quickly formulate queries without hesitating. The more experienced user may benefit from a system that begins to process the user's voice queries after measuring a short pause in the user's speech. The less experienced user may benefit form a system that begins to process the user's voice queries after measuring a longer pause in the user's speech to ensure the less experienced user has finished speaking.

    [0004] To determine a pause length to apply to a particular user's voice queries, a system analyzes the particular user's previous voice queries. For a particular user, the system looks at voice query frequency, pause length between words of previous voice queries, completeness of previous voice queries, and length of previous voice queries. A user who more frequently speaks voice queries may benefit from a shorter pause length to apply to a user's voice queries compared to a user who does not speak voice queries as frequently. A user who has a short average pause length between words of previous voice queries may benefit from a shorter pause length than a user who has longer average pause lengths between words. A user who speaks complete voice queries more often may benefit from a shorter pause length than a user who speaks complete queries less frequently. A user who speaks longer voice queries may benefit from a shorter pause length than a user who speaks shorter voice queries. Once the system computes an appropriate pause length to apply to a particular user's voice queries, the system may generate an endpoint during the particular user's future utterances using the particular user's pause length.

    [0005] The invention is defined in the appended claims.

    [0006] Particular embodiments of the subject matter described in this specification can be implemented so as to realize one or more of the following advantages. A user may use the voice input capabilities of a computing device and speak at a pace that is comfortable for the user. An utterance may be endpointed at the intended end of the utterance, leading to more accurate or desirable natural language processing outputs, and to faster processing by the natural language processing system.

    [0007] The details of one or more embodiments of the subject matter described in this specification are set forth in the accompanying drawings and the description below. Other features, aspects, and advantages of the subject matter will become apparent from the description, the drawings, and the claims.

    BRIEF DESCRIPTION OF THE DRAWINGS



    [0008] 

    FIG. 1 is a diagram of example utterances and signals used in determining, for a particular user, whether a user has finished speaking a voice query.

    FIG. 2 is a diagram of an example system that classifies a particular user based on the particular user's experience with speech input.

    FIG. 3 is a diagram of an example process for classifying a particular user based on the particular user's experience with speech input and determining whether the particular user has finished speaking a voice query.

    FIG. 4 is a block diagram of computing devices on which the processes described herein, or portions thereof, may be implemented.



    [0009] Like reference numbers and designations in the various drawings indicate like elements.

    DETAILED DESCRIPTION



    [0010] FIG. 1 is a diagram 100 of example utterances and signals used in determining, for a particular user, whether a user has finished speaking a voice query. In general, diagram 100 illustrates signals 103-118 that are generated or detected by computing device 121 when the computing device 121 is processing an incoming audio input. The computing device 121 receives the audio data corresponding to utterance 124 through a microphone or other audio input device of the computing device 121, and generates a transcription of the utterance 124 depending on a user profile assigned to the user 127.

    [0011] The utterance timing 130 represents the timing of the various words of the user 127 speaking utterance 124 (in the Fig. 1, "Text Mom love you"). The user 127 speaks each word with an increasing pause length between each word. The number of dots between each word is proportional to the pause length between each word. Each dot may represent a particular period of time, such as one hundred milliseconds. The first word 133, "Text," is followed by pause 136 of three dots that may correspond to three hundred milliseconds. The second word 139, "Mom," is followed by pause 142 of eight dots that may correspond to eight hundred milliseconds. The third word 145, "love," is followed by pause 128 of twelve dots that may correspond to 1.2 seconds. The fourth word 151, "you," is followed by pause 154. In FIG. 1, pause 154 is twenty dots that may correspond to two seconds. Because word 151 is at the end of utterance 124, a pause after word 151 may be much longer than pause 154 because the user 127 has stopped speaking. These pauses and the other pauses described below represent a natural period of silence between two words, and not that the user actively stopped speaking.

    [0012] In some implementations, the computing device 121 may generate, without factoring in any characteristics of the user 127, the general endpoint signal 103 and the complete query signal 106. The complete query signal 106 represents an estimate performed by the computing device 121 that the generated transcription of the utterance 130 represents a complete utterance. The computing device 121 compares the generated transcription to one or more complete utterances that the user 127 and other users have previously spoken. The computing device 121 may compare the generated transcription to the complete utterances after a speech recognizer of computing device 121 has identified a new word. For example, after the user 127 speaks word 133, a speech recognizer of the computing device 121 generates the transcription "text." The computing device 121 compares "text" to other complete utterances and determines that "text" is not a complete utterance. After the user 127 speaks word 139, the speech recognizer generates the transcription "text mom" that the computing device 121 identifies as complete. A similar determination is made after word 151. After the user 127 speaks word 145, the speech recognizer generates the transcription "text mom love" that the computing device 121 identifies as incomplete.

    [0013] The general endpoint signal 103 represents an estimate performed by the computing device 121 that the user 127 has finished speaking. The computing device 121 may generate the general endpoint signal 103 based on the length of pauses between speech audio. The computing device 121 may generate the general endpoint signal 103 without generating a transcription of the utterance 124. For example, the computing device 121 may receive audio data corresponding to word 133. During pause 136, the computing device 121 measures the time as it elapses during pause 136. The pause 136 may only last for three hundred milliseconds. If the general endpoint threshold is longer than three hundred milliseconds, such as six hundred milliseconds, then the computing device 121 will not trigger the general endpointer. After the computing device 121 receives audio data corresponding to word 139, the computing devices 121 measures the time of pause 142. After six hundred milliseconds of pause 142 has elapsed, the computing device 121 triggers the general endpointer and the general endpoint signal 103 indicates that an endpoint has been reached. With the general endpoint signal 103 indicating an endpoint of the utterance 124 and the complete query signal 106 indicating that the utterance 124 is complete, the computing device generates transcription 157, "text mom," for utterance 124.

    [0014] In some implementations, the computing device 121 may factor in the characteristics of the user 127 when identifying an endpoint of the utterance 124. On one hand, a novice user may speak with longer pauses between words possibly because the novice user may be unfamiliar with what terms may be best to speak to the computing device 121. On the other hand, an expert user may speak with shorter pauses between words because the expert user may be more comfortable and familiar with the speech input technology of computing device 121. Accordingly, the computing device 121 may lengthen or shorten the amount of time before it identifies a pause depending on how the computing device 121 categorizes the user.

    [0015] The novice pause detector signal 109 illustrates the computing device 121 detecting a pause in audio data corresponding to utterance 124, where the detected pause length is longer than the pause length the corresponds to the general endpointer. For example, the computing device 121 may detect pauses with a length of one second when the user 127 is classified as a novice user. Applying this pause threshold to utterance 124, the computing device 121 will not detect novice length pauses during pauses 136 and 124 because those pauses are of length three hundred milliseconds and eight hundred milliseconds, respectively. The computing device 121 does detect novice length pauses during pauses 148 and 154. As shown in novice pause detector signal 109, the computing device 121 detects a pause of one second during pause 148 after the user 127 spoke word 145. The computing device 121 also detects a pause of one second during pause 154 after the user spoke word 151.

    [0016] The computing device 121 determines, based on the novice pause detector signal 109 and the complete query signal 106, a speech endpoint for the utterance 124 when the computing device classifies the user as a novice. When the computing device 121 detects a pause, such as the pause of the novice pause detector signal 109 during pause 148, the computing device 121 determines whether the utterance 124 is complete. During pause 148, the complete query signal 106 indicates that the utterance 124 is not complete. Even though the computing device 121 detected a novice length pause, the utterance 124 is not complete, so the computing device 121 continues processing the audio data of the utterance 124. During pause 154, the computing device 121 detects a novice length pause and the complete query signal 106 indicates that the utterance is complete and, therefore, generates an endpoint of the utterance 124 as indicated by the novice endpoint signal 112. When the user 127 is classified as a novice, the endpoint of the utterance 124 is after word 151, and the transcription 160 of the utterance 124 is "Text Mom love you."

    [0017] The expert pause detector signal 115 illustrates the computing device 121 detecting a pause in audio data corresponding to utterance 124, where the detected pause length is shorter than the pause length the corresponds to the general endpointer. For example, the computing device 121 may detect pauses with a length of three hundred milliseconds when the user 127 is classified as an expert user. Applying this pause threshold to utterance 124, the computing device 121 detects expert length pauses during pauses 136, 142, 148, and 154. Because none of the pauses are less than three hundred milliseconds, all of the pauses in utterance 124 include an expert length pause detection.

    [0018] The computing device 121 combines the expert pause detector signal 115 and the complete query signal 106 to determine a speech endpoint for the utterance 124 when the computing device classifies the user as an expert. When the computing device 121 detects a pause, such as the pause of the expert pause detector signal 115 during pause 136, the computing device 121 determines whether the utterance 124 is complete. During pause 136, the complete query signal 106 indicates that the utterance 124 is not complete. Even though the computing device 121 detected an expert length pause, the utterance 124 is not complete, so the computing device 121 continues processing the audio data of the utterance 124. During pause 142, the computing device 121 detects an expert length pause and the complete query signal 106 indicates that the utterance is complete and, therefore, generates an endpoint of the utterance 124 as indicated by the expert endpoint signal 118. When the user 127 is classified as an expert, the endpoint of the utterance 124 is after word 139, and the transcription 163 of the utterance 124 is "Text Mom."

    [0019] FIG. 2 is diagram of an example system 200 that classifies a particular user based on the particular user's experience with speech input. In some implementations, the system 200 may be included in a computing device that the particular user uses for speech input, such as computing device 121. In some implementations, the system may be included in a server that processes transcriptions of speech input.

    [0020] The system 200 includes voice queries 205. The voice query log 205 stores the previous voice queries that users provide to the system 200. The voice query log 205 may include search queries, for example, "cat videos," and command queries, for example, "call mom." The voice query log 205 may include for each stored voice query, a timestamp, data indicating the duration of each pause between words, and data indicating whether the voice query is complete or incomplete based on a comparison with other voice queries.

    [0021] Query log 210 illustrates the voice queries provided by the user Bob. The voice queries in query log 210 include three voice queries and each includes either a complete indicator "[C]" or an incomplete indicator "[I]." Each voice query includes a timestamp that notes the date and time that Bob spoke the voice query. Each voice query includes data indicating the pause intervals between the spoken words. For example, "cat videos" may include data to indicate that Bob paused two hundred milliseconds between "cat" and "video." "Call ... mom" may include data to indicate that Bob paused one second between "call" and "mom."

    [0022] Query log 215 illustrates the voice queries provided by the user Alice. The voice queries in query log 215 include five voice queries and each includes either a complete indicator "[C]" or an incomplete indicator "[I]." Each voice query includes a timestamp that notes the date and time that Alice spoke the voice query. Each voice query includes data indicating the pause intervals between the spoken words. For example, "Text Sally that I'll be ten minutes late" may include data to indicate that Alice paused one millisecond between "text" and "Sally," paused three hundred milliseconds between "Sally" and "that," and paused 1.5 seconds between "that" and "I'II," as well as pause intervals between the other words. "Call mom" may include data to indicate that Alice paused three milliseconds between "call" and "mom."

    [0023] The voice query processor 220 processes the voice queries received from the voice query log 205. The voice query processor 220 generates a voice query experience score for each user. The voice query experience score indicates a level of experience that a particular user has with voice queries. A higher voice query experience score indicates that the particular user may have more experience with speaking voice queries. For example, to generate the voice query experience score for Bob, the voice query processor 220 processes query log 210.

    [0024] The voice query processor 220 includes a query completeness processor 225. The query completeness processor 225 accesses, for each user, the completeness data for each voice query and adjusts the user's voice query experience score. If a particular user has more complete voice queries and fewer incomplete voice queries, then the query completeness processor 225 increases the voice query experience score. If a particular user has more incomplete voice queries and fewer complete voice queries, then the query completeness processor 225 decreases the voice query experience score. The query completeness processor 225 may compare a ratio of complete queries to incomplete queries to a completeness ratio threshold in determining whether to increase or decrease the voice query experience score. For example, Bob has one incomplete voice query and two complete queries. Based on that information, the query completeness processor 225 may decrease Bob's voice query experience score. Alice has no incomplete voice queries and five complete queries. Based on that information, the query completeness processor 225 may increase Alice's voice query experience score.

    [0025] The voice query processor 220 includes a query length processor 230. The query length processor 230 computes, for each user, a length of each voice query and adjusts the user's voice query experience score. If a particular user has more long voice queries and fewer short voice queries, then the voice query processor 220 increases the voice query experience score. If a particular user has more short voice queries and fewer long voice queries, then the voice query processor 220 decreases the voice query experience score. The query length processor 230 may use a threshold to determine whether a voice query is long or short. The query length processor 230 may compare a ratio of long queries to short queries to a length ratio threshold in determining whether to increase or decrease the voice query experience score. For example, Bob has one incomplete voice query and two complete queries. Based on that information, the query completeness processor 225 may decrease Bob's voice query experience score. Alice has no incomplete voice queries and five complete queries. Based on that information, the query completeness processor 225 may increase Alice's voice query experience score.

    [0026] The voice query processor 220 includes a pause interval processor 235. The pause interval processor 235 computes, for each user, an average pause length between words for the user's voice queries. The pause interval processor 235 may compare the average pause length for each user to a pause threshold to determine whether to increase or decrease the user's voice query experience score. An average pause length above the pause threshold decreases the voice query experience score. An average pause length below the pause threshold increases the voice query experience score. For example, Bob may have an average pause length of 1.2 seconds. Alice may have an average pause length of two hundred milliseconds. If the pause threshold is one second, then the pause interval processor 235 increases the voice query experience score for Alice and decreases the voice query experience score for Bob.

    [0027] The voice query processor 220 includes a query counter 240. The query counter 240 computes, for each user, a number of voice queries submitted and adjusts the voice query experience score. If a particular user provides many voice queries, then the voice query processor 220 increases the voice query experience score. If a particular user provides fewer voice queries, then the voice query processor 220 decreases the voice query experience score. The query length processor 230 may use a daily voice query threshold and compare the threshold to an average daily number of voice queries to determine whether a user submits many or few voice queries. For example, the daily voice query threshold is one query per day. Based on query log 210, Bob submitted less than one query per day. Therefore, the query counter 240 decreases Bob's voice query experience score. Alice has more than one query per day based on query log 215. Therefore, the query counter 240 increases Alice's voice query experience score.

    [0028] Each of the processors included in the voice query processor 220 may increase or decrease the user's voice query experience score an amount that is proportional to a difference between the threshold and the corresponding user's value. For example, Alice averages 2.5 voice queries per day, and Bob averages 0.125 queries per day. With a daily voice query threshold of one query per day, Alice is 1.5 over the threshold, and Bob is 0.875 below the threshold. The query counter 240 increases Alice's voice query experience score by some factor multiplied by 1.5 and decreases Bob's voice query experience score by the same factor multiplied by 0.875.

    [0029] The voice query processor 220 provides each user's voice query experience score to a user profiler 245. The user profiler 245 assigns a voice query experience profile to each user based the user's voice query experience score. The voice profiler 245 accesses the profile thresholds 250 to match each user's voice query experience score to a voice query experience profile. Each voice query experience profile may correspond to a range of voice query experience score. For example, an expert profile may correspond to a range of 80 to 100. A novice profile may correspond to a range of 0 to 20. Other profiles may exist between the voice query experience scores of 20 and 80. In the example in FIG. 2, Bob may have a voice query experience score of 18 and Alice may have a voice query experience score of 88. Therefore, Bob is classified as a novice, and Alice is classified as an expert. The user profiler 245 then stores the voice query experience profile for each user in profile storage 255. Profiles 260 illustrates the profiles of Alice and Bob stored in profile storage 255

    [0030] Each voice query experience profile corresponds to a different pause length threshold that the system 200 subsequently uses when generating an endpoint for a future utterance. As shown in FIG. 1 and described above, the expert profile corresponds to a pause length threshold of three hundred milliseconds that is used to endpoint an utterance. The novice profile corresponds to a pause length threshold of one second. The system 200 may define other profiles such as an intermediate profile with a pause length threshold of six hundred milliseconds. In some implementations, the system 200 may assign a pause length threshold to a user without assigning a profile to the user. The system may generate a voice query experience score and compute a pause length threshold that is inversely proportional to the voice query experience score.

    [0031] FIG. 3 is a diagram of an example process 300 for classifying a particular user based on the particular user's experience with speech input and determining whether the particular user has finished speaking a voice query. The process 300 may be performed by a computing device such as computing device 121 from FIG. 1 or computing device 200 from FIG. 2. The process 300 analyzes a user's previous voice queries to determine a pause threshold for use in determining when the user has finished speaking future queries.

    [0032] The computing device accesses voice query log data (310). The computing device determines, for a particular user, a pause threshold (320). The computing device receives, from the particular user, an utterance (330). The computing device determines that the particular user has stopped speaking for the pause threshold (340). The computing device processes the utterance as a voice query (350).

    [0033] FIG. 4 is a block diagram of computing devices 400 that may be used to implement the systems and methods described in this document, as either a client or as a server or plurality of servers. Computing device 400 is intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers.

    [0034] Computing device 400 includes a processor 402, memory 404, a storage device 406, a high-speed interface 408 connecting to memory 404 and high-speed expansion ports 410, and a low speed interface 412 connecting to low speed bus 414 and storage device 406. Each of the components 402, 404, 406, 408, 410, and 412, are interconnected using various busses, and may be mounted on a common motherboard or in other manners as appropriate. The processor 402 can process instructions for execution within the computing device 400, including instructions stored in the memory 404 or on the storage device 406 to display graphical information for a GUI on an external input/output device, such as display 416 coupled to high speed interface 408. In other implementations, multiple processors and/or multiple buses may be used, as appropriate, along with multiple memories and types of memory. Also, multiple computing devices 400 may be connected, with each device providing portions of the necessary operations, e.g., as a server bank, a group of blade servers, or a multiprocessor system.

    [0035] The computing device 400 may be implemented in a number of different forms, as shown in the figure. For example, it may be implemented as a standard server 420, or multiple times in a group of such servers. It may also be implemented as part of a rack server system 424. In addition, it may be implemented in a personal computer such as a laptop computer 422. Alternatively, components from computing device 400 may be combined with other components in a mobile device (not shown). Each of such devices may contain one or more of computing device 400 and an entire system may be made up of multiple computing devices 400 communicating with each other.

    [0036] Various implementations of the systems and methods described here can be realized in digital electronic circuitry, integrated circuitry, specially designed ASICs (application specific integrated circuits), computer hardware, firmware, software, and/or combinations of such implementations. These various implementations can include implementation in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

    [0037] The systems and techniques described here can be implemented in a computing system that includes a back end component, e.g., as a data server, or that includes a middleware component, e.g., an application server, or that includes a front end component, e.g., a client computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here, or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication, e.g., a communication network. Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), and the Internet.


    Claims

    1. A computer-implemented method comprising:

    accessing a collection of voice queries that were submitted by a particular user (310);

    determining a duration value based on durations of pauses between words of the voice queries in the collection of voice queries (320), wherein the duration value corresponds to a pause length threshold;

    classifying the voice queries in the collection of voice queries (320) as complete or incomplete, and modifying the duration value based on the classifications, wherein modifying the duration value based on the classifications comprises (a) based on classifications of the queries as being complete or incomplete, increasing or decreasing a voice query experience score for the user, and (b) assigning a voice query experience profile to the user based on the voice query experience score, wherein the voice query experience profile corresponds to the pause length threshold;

    (i) receiving an utterance in which the particular user speaks one or more words, then pauses for less than the duration value, then speaks one or more other words (330); and
    submitting a voice query that includes a transcription of the one or more words that make up an initial portion of the utterance and the one or more other words; and

    (ii) receiving an additional utterance in which the particular user speaks one or more additional words, then pauses for greater than the duration value, then speaks one or more other additional words; and

    submitting a voice query that includes the transcription of the one or more words that make up the initial portion of the additional utterance without the one or more other additional words.
     
    2. The method of claim 1, comprising:

    determining a voice query length based on the voice queries in the collection of voice queries; and

    further modifying the duration value based on the voice query length.


     
    3. The method of claim 2, wherein the voice query length is an average duration of the voice queries in the collection of voice queries.
     
    4. The method of claim 2, wherein the voice query length is an average number of words in the voice queries in the collection of voice queries.
     
    5. The method of any preceding claim, comprising:

    determining a quantity of voice queries based on quantities of the voice queries in the collection of voice queries that were spoken each day; and

    further modifying the duration value based on the quantity of voice queries.


     
    6. The method of claim 5, wherein the quantity of voice queries is an average number of the voice queries in the collection of voice queries that were spoken each day.
     
    7. The method of any preceding claim, comprising:
    rounding the duration value to a nearest duration among a set of predetermined durations.
     
    8. The method of any preceding claim, comprising:

    receiving an additional utterance in which another user speaks one or more words, then pauses for less than the duration value, then speaks one or more other additional words; and

    submitting an additional voice query that includes a transcription of the one or more additional words that make up an initial portion of the additional utterance.


     
    9. The method of any preceding claim, wherein the duration value is an average duration of pauses between words of the voice queries in the collection of voice queries.
     
    10. The method of any preceding claim, wherein the increasing or decreasing a voice query experience score for the user is based on a ratio of the complete queries to the incomplete queries.
     
    11. A system (200) comprising:
    one or more computers and one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising:

    accessing a collection of voice queries that were submitted by a particular user;

    determining a duration value based on durations of pauses between words of the voice queries in the collection of voice queries, wherein the duration value corresponds to a pause length threshold;

    classifying the voice queries in the collection of voice queries as complete or incomplete and modifying the duration value based on the classifications, wherein modifying the duration value based on the classifications comprises (a) based on classifications of the queries as being complete or incomplete, increasing or decreasing a voice query experience score for the user, and (b) assigning a voice query experience profile to the user based on the voice query experience score, wherein the voice query experience profile corresponds to the pause length threshold;

    (i) receiving an utterance in which the particular user speaks one or more words, then pauses for less than the duration value, then speaks one or more other words; and
    submitting a voice query that includes a transcription of the one or more words that make up an initial portion of the utterance and the one or more other words; and

    (ii) receiving an additional utterance in which the particular user speaks one or more additional words, then pauses for greater than the duration value, then speaks one or more other additional words; and
    submitting a voice query that includes the transcription of the one or more words that make up the initial portion of the additional utterance without the one or more other additional words.


     
    12. One or more storage devices storing instructions, wherein the instructions are operable, when executed by one or more computers, to cause the one or more computers to perform operations comprising the method of any of claims 1 to 10.
     
    13. A non-transitory computer-readable medium storing software comprising instructions, wherein the instructions are executable by one or more computers to cause one or more computers to perform the method of any of claims 1 to 10.
     


    Ansprüche

    1. Computerimplementiertes Verfahren, umfassend:
    Zugreifen auf eine Sammlung von Sprachanfragen, welche von einem bestimmten Benutzer (310) gestellt wurden:

    Bestimmen eines Dauerwerts, basierend auf Dauern von Pausen zwischen Worten der Sprachanfragen in der Sammlung von Sprachanfragen (320), wobei der Dauerwert einer Pausenlängenschwelle entspricht;

    Klassifizieren der Sprachanfragen in der Sammlung von Sprachanfragen (320) als vollständig oder unvollständig, und Modifizieren des Dauerwerts basierend auf den Klassifizierungen, wobei das Modifizieren des Dauerwerts basierend auf den Klassifizierungen umfasst: (a) basierend auf Klassifizierungen der Anfragen als vollständig oder unvollständig, Erhöhen oder Reduzieren einer Sprachanfrageerfahrungsbewertung für den Benutzer, und (b) Zuweisen eines Sprachanfrageerfahrungsprofils an den Benutzer basierend auf der Sprachanfrageerfahrungsbewertung, wobei das Sprachanfrageerfahrungsprofil der Pausenlängenschwelle entspricht;

    (i) Empfangen einer Äußerung, in welcher der bestimmte Benutzer ein oder mehrere Worte ausspricht, danach eine Pause macht, die kürzer als der Dauerwert ist, danach ein oder mehrere Worte (330) ausspricht; und
    Stellen einer Sprachanfrage, welche eine Transkription von einem oder mehreren Worten umfasst, welche einen anfänglichen Abschnitt der Äußerung und des einen oder mehreren Worte ausmachen; und

    (ii) Empfangen einer zusätzlichen Äußerung, in welcher der bestimmte Benutzer ein oder mehrere zusätzliche Worte ausspricht, danach eine Pause macht, die länger als der Dauerwert ist, danach ein oder mehrere weitere zusätzliche Worte ausspricht; und
    Stellen einer Sprachanfrage, welche die Transkription des einen oder der mehreren Worte umfasst, welche den anfänglichen Abschnitt der zusätzlichen Äußerung ohne das eine oder die zusätzlichen Worten ausmachen.


     
    2. Verfahren nach Anspruch 1, umfassend:

    Bestimmen einer Sprachanfragenlänge, basierend auf den Sprachanfragen in der Sammlung von Sprachanfragen; und

    weiteres Modifizieren des Dauerwerts basierend auf der Sprachanfragelänge.


     
    3. Verfahren nach Anspruch 2, wobei die Sprachanfragenlänge eine durchschnittliche Dauer der Sprachanfragen in der Sammlung von Sprachanfragen ist.
     
    4. Verfahren nach Anspruch 2, wobei die Sprachanfragenlänge eine durchschnittliche Anzahl von Worten in den Sprachanfragen in der Sammlung von Sprachanfragen ist.
     
    5. Verfahren nach einem der vorhergehenden Ansprüche, umfassend:

    Bestimmen einer Anzahl von Sprachanfragen basierend auf Anzahlen der Sprachanfragen in der Sammlung von Sprachanfragen, welche jeden Tag ausgesprochen werden; und

    weiteres Modifizieren des Dauerwerts basierend auf der Anzahl von Sprachanfragen.


     
    6. Verfahren nach Anspruch 5, wobei die Anzahl von Sprachanfragen eine durchschnittliche Anzahl von Sprachanfragen in der Sammlung von Sprachanfragen ist, welche jeden Tag ausgesprochen wurden.
     
    7. Verfahren nach einem der vorhergehenden Ansprüche, umfassend:
    Runden des Dauerwerts auf eine nächste Dauer in einem Satz von vorbestimmten Dauern.
     
    8. Verfahren nach einem der vorhergehenden Ansprüche, umfassend:

    Empfangen einer zusätzlichen Äußerung, in welcher ein anderer Benutzer ein oder mehrere Worte ausspricht, danach eine Pause macht, die kürzer als der Dauerwert ist, danach ein oder mehrere weitere zusätzliche Worte ausspricht; und

    Stellen einer zusätzlichen Sprachanfrage, welche eine Transkription des einen oder der mehreren zusätzlichen Worte umfasst, welche einen anfänglichen Abschnitt der zusätzlichen Äußerung ausmachen.


     
    9. Verfahren nach einem der vorhergehenden Ansprüche, wobei der Dauerwert eine durchschnittliche Dauer von Pausen zwischen Worten der Sprachanfragen in der Sammlung von Sprachanfragen ist.
     
    10. Verfahren nach einem der vorhergehenden Ansprüche, wobei das Erhöhen oder das Reduzieren einer Sprachanfrageerfahrungsbewertung für den Benutzer auf einem Verhältnis zwischen vollständigen und unvollständigen Anfragen basiert.
     
    11. System (200), umfassend:
    einen oder mehrere Computer und eine oder mehrere Speichervorrichtungen, welche Anweisungen speichern, welche, wenn durch den einen oder den mehreren Computern ausgeführt, betreibbar sind, um zu veranlassen, dass der eine oder die mehreren Computer Operationen ausführen, umfassend:
    Zugreifen auf eine Sammlung von Sprachanfragen, welche von einem bestimmten Benutzer gestellt wurden:

    +Bestimmen eines Dauerwerts, basierend auf Dauern von Pausen zwischen Worten der Sprachanfragen in der Sammlung von Sprachanfragen, wobei der Dauerwert einer Pausenlängenschwelle entspricht;

    Klassifizieren der Sprachanfragen in der Sammlung von Sprachanfragen als vollständig oder unvollständig, und Modifizieren des Dauerwerts basierend auf den Klassifizierungen, wobei das Modifizieren des Dauerwerts basierend auf den Klassifizierungen umfasst: (a) basierend auf Klassifizierungen der Anfragen als vollständig oder unvollständig, Erhöhen oder Reduzieren einer Sprachanfrageerfahrungsbewertung für den Benutzer, und (b) Zuweisen eines Sprachanfrageerfahrungsprofils an den Benutzer basierend auf der Sprachanfrageerfahrungsbewertung, wobei das Sprachanfrageerfahrungsprofil der Pausenlängenschwelle entspricht;

    (i) Empfangen einer Äußerung, in welcher der bestimmte Benutzer ein oder mehrere Worte ausspricht, danach eine Pause macht, die kürzer als der Dauerwert ist, danach ein oder mehrere Worte ausspricht; und
    Stellen einer Sprachanfrage, welche eine Transkription von einem oder mehreren Worten umfasst, welche einen anfänglichen Abschnitt der Äußerung und des einen oder mehreren Worte ausmachen; und

    (ii) Empfangen einer zusätzlichen Äußerung, in welcher der bestimmte Benutzer ein oder mehrere zusätzliche Worte ausspricht, danach eine Pause macht, die länger als der Dauerwert ist, danach ein oder mehrere weitere zusätzliche Worte ausspricht; und
    Stellen einer Sprachanfrage, welche die Transkription des einen oder der mehreren Worte umfasst, welche den anfänglichen Abschnitt der zusätzlichen Äußerung ohne das eine oder die zusätzlichen Worte ausmachen.


     
    12. Eine oder mehrere Speichervorrichtungen, welche Anweisungen speichern, wobei die Anweisungen, wenn durch einen oder mehreren Computern ausgeführt, steuerbar sind, um zu veranlassen, dass der eine oder die mehreren Computer Operationen ausführen, umfassend das Verfahren nach einem der Ansprüche 1 bis 10.
     
    13. Nichtflüchtiges computerlesbares Medium, welches eine Software speichert, die Anweisungen umfasst, wobei die Anweisungen von einem oder mehreren Computern ausführbar sind, um zu veranlassen, dass der eine oder die mehreren Computer das Verfahren nach einem der Ansprüche 1 bis 10 ausführen.
     


    Revendications

    1. Procédé mis en Ĺ“uvre par un ordinateur, comprenant :

    l'accès à une collection de requêtes vocales qui ont été soumises par un utilisateur particulier (310) ;

    la détermination d'une valeur de durée sur la base de durées de pauses entre des mots des requêtes vocales dans la collection de requêtes vocales (320), dans lequel la valeur de durée correspond à un seuil de longueur de pause ;

    la classification des requêtes vocales dans la collection de requêtes vocales (320) comme étant complètes ou incomplètes, et la modification de la valeur de durée sur la base des classifications, dans lequel la modification de la valeur de durée sur la base des classifications comprend (a) sur la base de classifications des requêtes comme étant complètes ou incomplètes, l'augmentation ou la diminution d'un score d'expérience de requêtes vocales pour l'utilisateur, et (b) l'attribution d'un profil d'expérience de requêtes vocales à l'utilisateur sur la base du score d'expérience de requêtes vocales, dans lequel le profil d'expérience de requêtes vocales correspond au seuil de longueur de pause ;

    (i) la réception d'une énonciation dans laquelle l'utilisateur particulier prononce un ou plusieurs mots, puis fait une pause d'une durée inférieure à la valeur de durée, puis prononce un ou plusieurs autres mots (330) ; et
    la soumission d'une requête vocale qui inclut une transcription des un ou plusieurs mots qui constituent une partie initiale de l'énonciation et des un ou plusieurs autres mots ; et

    (ii) la réception d'une énonciation additionnelle dans laquelle l'utilisateur particulier prononce un ou plusieurs mots additionnels, puis fait une pause d'une durée supérieure à la valeur de durée, puis prononce un ou plusieurs autres mots additionnels ; et
    la soumission d'une requête vocale qui inclut la transcription des un ou plusieurs mots qui constituent la partie initiale de l'énonciation additionnelle sans les un ou plusieurs autres mots additionnels.


     
    2. Procédé selon la revendication 1, comprenant :

    la détermination d'une longueur de requête vocale sur la base des requêtes vocales dans la collection de requêtes vocales ; et

    la modification supplémentaire de la valeur de durée sur la base de la longueur de requête vocale.


     
    3. Procédé selon la revendication 2, dans lequel la longueur de requête vocale est une durée moyenne des requêtes vocales dans la collection de requêtes vocales.
     
    4. Procédé selon la revendication 2, dans lequel la longueur de requête vocale est un nombre moyen de mots dans les requêtes vocales dans la collection de requêtes vocales.
     
    5. Procédé selon l'une quelconque des revendications qui précèdent, comprenant :

    la détermination d'une quantité de requêtes vocales sur la base de quantités des requêtes vocales dans la collection de requêtes vocales qui ont été prononcées chaque jour ; et

    la modification supplémentaire de la valeur de durée sur la base de la quantité de requêtes vocales.


     
    6. Procédé selon la revendication 5, dans lequel la quantité de requêtes vocales est un nombre moyen des requêtes vocales dans la collection de requêtes vocales qui ont été prononcées chaque jour.
     
    7. Procédé selon l'une quelconque des revendications qui précèdent, comprenant :
    l'arrondi de la valeur de durée à une durée la plus proche parmi un jeu de durées prédéterminées.
     
    8. Procédé selon l'une quelconque des revendications qui précèdent, comprenant :

    la réception d'une énonciation additionnelle selon laquelle un autre utilisateur prononce un ou plusieurs mots, puis fait une pause d'une durée inférieure à la valeur de durée, puis prononce un ou plusieurs autres mots additionnels ; et

    la soumission d'une requête vocale additionnelle qui inclut une transcription des un ou plusieurs mots additionnels qui constituent une partie initiale de l'énonciation additionnelle.


     
    9. Procédé selon l'une quelconque des revendications qui précèdent, dans lequel la valeur de durée est une durée moyenne de pauses entre des mots des requêtes vocales dans la collection de requêtes vocales.
     
    10. Procédé selon l'une quelconque des revendications qui précèdent, dans lequel l'augmentation ou la diminution d'un score d'expérience de requêtes vocales pour l'utilisateur est basée sur un rapport des requêtes complètes sur les requêtes incomplètes.
     
    11. Système (200) comprenant :
    un ou plusieurs ordinateurs et un ou plusieurs dispositifs de stockage qui stocke(nt) des instructions qui peuvent être rendues opérationnelles, lorsqu'elles sont exécutées par les un ou plusieurs ordinateurs, pour forcer les un ou plusieurs ordinateurs à réaliser les opérations comprenant :

    l'accès à une collection de requêtes vocales qui ont été soumises par un utilisateur particulier ;

    la détermination d'une valeur de durée sur la base de durées de pauses entre des mots des requêtes vocales dans la collection de requêtes vocales, dans lequel la valeur de durée correspond à un seuil de longueur de pause ;

    la classification des requêtes vocales dans la collection de requêtes vocales comme étant complètes ou incomplètes, et la modification de la valeur de durée sur la base des classifications, dans lequel la modification de la valeur de durée sur la base des classifications comprend (a) sur la base de classifications des requêtes comme étant complètes ou incomplètes, l'augmentation ou la diminution d'un score d'expérience de requêtes vocales pour l'utilisateur, et (b) l'attribution d'un profil d'expérience de requêtes vocales à l'utilisateur sur la base du score d'expérience de requêtes vocales, dans lequel le profil d'expérience de requêtes vocales correspond au seuil de longueur de pause ;

    (i) la réception d'une énonciation dans laquelle l'utilisateur particulier prononce un ou plusieurs mots, puis fait une pause d'une durée inférieure à la valeur de durée, puis prononce un ou plusieurs autres mots ; et
    la soumission d'une requête vocale qui inclut une transcription des un ou plusieurs mots qui constituent une partie initiale de l'énonciation et des un ou plusieurs autres mots ; et

    (ii) la réception d'une énonciation additionnelle dans laquelle l'utilisateur particulier prononce un ou plusieurs mots additionnels, puis fait une pause d'une durée supérieure à la valeur de durée, puis prononce un ou plusieurs autres mots additionnels ; et
    la soumission d'une requête vocale qui inclut la transcription des un ou plusieurs mots qui constituent la partie initiale de l'énonciation additionnelle sans les un ou plusieurs autres mots additionnels.


     
    12. Un ou plusieurs dispositifs de stockage stockant des instructions, dans lesquels les instructions peuvent être rendues opérationnelles, lorsqu'elles sont exécutées par un ou plusieurs ordinateurs, pour forcer les un ou plusieurs ordinateurs à réaliser des opérations comprenant le procédé selon l'une quelconque des revendications 1 à 10.
     
    13. Support non transitoire pouvant être lu par un ordinateur et stockant un logiciel qui comprend des instructions, dans lequel les instructions peuvent être exécutées par un ou plusieurs ordinateurs pour forcer les un ou plusieurs ordinateurs à réaliser le procédé selon l'une quelconque des revendications 1 à 10.
     




    Drawing















    REFERENCES CITED IN THE DESCRIPTION



    This list of references cited by the applicant is for the reader's convenience only. It does not form part of the European patent document. Even though great care has been taken in compiling the references, errors or omissions cannot be excluded and the EPO disclaims all liability in this regard.

    Patent documents cited in the description