JP5196114B2 - Speech recognition apparatus and program - Google Patents

Description

  The present invention relates to a technology for recognizing speech.

Techniques for searching for specific words (keywords) from voice signals have been proposed. For example, the speech recognition apparatus disclosed in Patent Document 1 is designated from a speech signal by collating an acoustic model corresponding to a word designated by a user (hereinafter referred to as “designated word”) with a speech signal recorded in advance. Search for a word.
Japanese Patent Laid-Open No. 2001-290496

  Since it is technically difficult to completely avoid misrecognition of speech recognition, in the technique of Patent Document 1, a word other than a designated word may be searched or a designated word may not be searched. If it is possible to search for a specific designated word from the sound recorded in a meeting uttered by multiple participants, it will be convenient for creating minutes, for example, but if a word other than the designated word is erroneously detected, the word will be If the designated word is not retrieved, a serious problem such as a lack of speech in the minutes can occur.

  Further, in the configuration of Patent Document 1, only a portion that completely matches one designated word in the audio signal is searched, and a separate word used in the same meaning as the designated word or a word related to the designated word is Not searched. Therefore, in order to detect all words related to a predetermined designated word, it is necessary to repeat the search after setting each word as the designated word. In view of the above circumstances, an object of the present invention is to solve the problem of efficiently recognizing a word related to a designated word while suppressing the possibility of erroneous recognition.

In order to solve the above problems, the speech recognition apparatus according to the present invention stores related word specifying means for specifying a plurality of related words for a word specified by a user, and a memory for storing an appearance probability for each of the plurality of words. Means, probability adjusting means for relatively increasing the appearance probability of each related word specified by the related word specifying means among the plurality of words with respect to the appearance probability of words other than the plurality of related words, and the voice represented by the audio signal A speech recognition means for specifying a word corresponding to the word based on the appearance probability after adjustment by the probability adjustment means, a selection means for selecting a related word from a plurality of words specified by the speech recognition means, and a speech signal for each speaker A character string of each related word by specifying a speaker identifying means that divides into utterance sections, and a utterance section in which each related word selected by the selecting means is uttered among a plurality of utterance sections classified by the speaker identifying means. For each speaker Display control means to be displayed on the display device, and the speech recognition means calculates the reliability (for example, reliability A1 in FIG. 1) of a specific result by the speech recognition means for each related word selected by the selection means. The speaker identification means calculates the reliability (for example, reliability A2 in FIG. 1) of the classification result by the speaker identification means for each utterance section, and the display control means calculates the character string of each related word. Aspect (size or display color (hue, brightness, saturation)) according to the reliability calculated by the speech recognition means for the related word and the reliability calculated by the speaker identification means for the utterance section in which the related word was uttered Or character type) on the display device .

  In the above configuration, since the speech signal recognition is performed after relatively increasing the appearance probability of each of the plurality of related words according to the designated word, the appearance probability of each word is maintained at the initial value. Compared to a configuration in which speech recognition is performed as it is, it is possible to efficiently recognize each related word while suppressing the possibility of erroneous recognition. Note that “increasing the appearance probability of each related word relative to the appearance probability of words other than a plurality of related words” means a process of increasing the appearance probability of each related word (for each word other than related words). The appearance probability is not changed) and the process of reducing the appearance probability of each word other than the related word (the appearance probability of each related word is not changed). In the present invention, whether or not a specified word (word specified by the user) is included in a plurality of related words specified by the related word specifying unit is unquestioned.

The speech recognition apparatus according to a preferred aspect of the present invention outputs, from the sound emitting device, the sound of the portion corresponding to the related word in the audio signal when the user specifies the related word displayed on the display device. A reproduction control means is provided. According to this aspect, since the sound of the part corresponding to each related word is reproduced, there is an advantage that the user can easily confirm the content of the utterance of the part corresponding to each related word. For example, the reproduction control unit outputs, from the sound emitting device, the sound of a portion starting from a time point a predetermined time before the time of the related word specified by the user in the audio signal.

The speech recognition apparatus according to the present invention is realized by hardware (electronic circuit) such as a DSP (Digital Signal Processor) dedicated to speech processing, and a general-purpose arithmetic processing device such as a CPU (Central Processing Unit). It is also realized through collaboration with the program. A program according to the present invention includes: a computer having a storage unit that stores an appearance probability for each of a plurality of words; a related word specifying process for specifying a plurality of related words for a word specified by a user; Probability adjustment processing that raises the appearance probability of each related word specified in the related word specification processing relative to the appearance probability of words other than a plurality of related words, and probability adjustment of the word corresponding to the voice represented by the audio signal Speech recognition processing that is specified based on the appearance probability after processing, selection processing that selects related words from a plurality of words that are specified in the speech recognition processing, and speaker identification processing that divides the speech signal into speech sections for each speaker And by identifying the utterance section in which each related word selected in the selection process is uttered from among the plurality of utterance sections divided by the speaker identification process, the character string of each related word is displayed on the display device for each speaker. The In the speech recognition process, the reliability of a specific result by the speech recognition process is calculated for each related word selected in the selection process, and the speaker identification process The reliability of the classification result by the person identification process is calculated for each utterance section, and in the display control process, the string of each related word is calculated with the reliability calculated by the speech recognition process for the related word and the related word is uttered. The displayed speech section is displayed on the display device in a manner corresponding to the reliability calculated in the speaker identification process . Even with the above program, the same operations and effects as the speech recognition apparatus according to the present invention are exhibited. The program of the present invention is provided to the user in a form stored in a computer-readable recording medium and installed in the computer, or is provided from the server device in the form of distribution via a communication network. To be installed.

  The present invention is also specified as a method for recognizing speech. The speech recognition method according to a specific aspect includes a related word specifying process for specifying a plurality of related words for a word specified by a user, and an appearance probability of each related word specified in the related word specifying process among the plurality of words And a speech recognition process for specifying a word corresponding to the speech represented by the speech signal based on the appearance probability after processing by the probability adjustment process. According to the above method, the same operation and effect as the speech recognition apparatus according to the present invention are exhibited.

  FIG. 1 is a block diagram showing a configuration of a speech recognition apparatus 100 according to an embodiment of the present invention. As shown in FIG. 1, the speech recognition apparatus 100 is a computer system that includes a control device 10 and a storage device 30. An input device 42, a display device 44, and a sound emitting device 46 are connected to the control device 10. The input device 42 is a device (keyboard or mouse) for a user to input an instruction to the voice recognition device 100. For example, the user inputs a desired word (keyword) KW by appropriately operating the input device 42. The display device 44 displays various images under the control of the control device 10. The sound emitting device 46 is a device (for example, a speaker or a headphone) that emits sound corresponding to a signal supplied from the control device 10.

  The storage device 30 stores a program executed by the control device 10 and various data used by the control device 10. A known recording medium such as a semiconductor storage device or a magnetic storage device is arbitrarily adopted as the storage device 30. As shown in FIG. 1, the storage device 30 stores a speech signal S, a co-occurrence database (DB) C, a recognition dictionary D, and a phoneme model group G. Each piece of information described above may be distributed and stored in separate storage devices.

  The sound signal S represents a waveform of a sound collected in advance using a sound collecting device (not shown). The voice represented by the voice signal S is, for example, voice recorded in a conference in which a plurality of participants speak at any time in a space such as a conference room.

  The co-occurrence database C is a database in which a plurality of different words are associated with each of a large number of words. One word is associated with a synonym of the word or a word that is semantically related (a word that is highly likely to appear in the same context).

  The recognition dictionary D is a database used for recognition of the audio signal S. FIG. 2 is a conceptual diagram schematically showing the contents of the recognition dictionary D. As shown in the figure, the recognition dictionary D includes a phoneme string DA, a character string DB, and an appearance probability P for each of a plurality (N) of words. The phoneme string DA is an array of phonemes constituting words. The character string DB is an array of characters (for example, kanji) when a word is written. The appearance probability P is a probability that a word appears. The appearance probability P is determined in advance for each word by statistically processing the frequency with which the word is used in general sentences such as newspaper articles.

  The phoneme model group G in FIG. 1 includes a plurality of phoneme models that model the acoustic characteristics of average speech for each phoneme. As the phoneme model, a known probability model represented by a hidden Markov model is arbitrarily adopted.

  The control device 10 executes a program stored in the storage device 30 to execute a plurality of elements (related word identification unit 12, probability adjustment unit 14, speech recognition unit 16, speaker identification unit 22, display control unit 24, reproduction It functions as the control unit 26). The function of each element realized by the control device 10 will be described in detail below. Note that each element of the control device 10 is also realized by an electronic circuit such as a DSP dedicated to voice processing. Further, the control device 10 may be distributed and mounted on a plurality of integrated circuits.

  The related word specifying unit 12 specifies a plurality of related words RW for the designated word KW. The plurality of related words RW specified by the related word specifying unit 12 include a word associated with the specified word KW in the co-occurrence database C and a specified word KW input from the input device 42 by the user.

  The probability adjusting unit 14 searches each related word RW (including the designated word KW) from a plurality of words registered in the recognition dictionary D, and increases the appearance probability P corresponding to the related word RW from the initial value. . For example, the probability adjustment unit 14 calculates the updated appearance probability P by multiplying or adding a predetermined coefficient to the initial appearance probability P. On the other hand, the appearance probability P of each word other than the related word RW is maintained at the initial value.

  The voice recognition unit 16 is means for recognizing the voice signal S stored in the storage device 30. More specifically, the voice recognition unit 16 determines a time series of words corresponding to the voice of the voice signal S (words other than the related word RW and the related word RW) and the time when each word was uttered as a recognition dictionary D. Based on the phoneme model group G, it specifies sequentially. Of the plurality of words registered in the recognition dictionary D, a word having a high evaluation value (score) SC is selected. The evaluation value SC is calculated as an addition value or a weighted sum of the acoustic evaluation value (acoustic score) AS and the linguistic evaluation value (language score) LS.

  The acoustic evaluation value AS is an index of correlation between a time series of acoustic feature quantities (for example, MFCC (Mel Frequency Cepstral Coefficients)) extracted from the speech signal S for each frame and each acoustic model of a plurality of words. (For example, the distance between the two). The acoustic model of each word is a probability model in which the phoneme models of each phoneme constituting the phoneme string DA of the word are selected from the phoneme model group G and combined. Therefore, the acoustic evaluation value AS is a higher numerical value for a word whose acoustic model approximates the time series of the feature amount of the speech signal S. On the other hand, the linguistic evaluation value LS is a numerical value corresponding to the appearance probability P set for each word in the recognition dictionary D after adjustment by the probability adjustment unit 14. In this embodiment, the appearance probability P is adopted as the linguistic evaluation value LS. Since the evaluation value SC is calculated based on the addition of the acoustic evaluation value AS and the linguistic evaluation value LS, the evaluation value SC increases as the acoustic evaluation value AS or the linguistic evaluation value LS increases.

  Since the probability adjustment unit 14 increases the appearance probability P of the related word RW, the evaluation value SC (linguistic evaluation value LS) of the related word RW increases, so the probability that the speech recognition unit 16 recognizes the related word RW is: It rises compared to the case where the word is not specified as the related word RW. That is, each related word RW is easily recognized by the voice recognition unit 16.

  As shown in FIG. 1, the voice recognition unit 16 includes a selection unit 162. The selection unit 162 selects each related word RW from the time series of a plurality of words recognized by the voice recognition unit 16 through the above processing. Furthermore, the speech recognition unit 16 calculates the accuracy (hereinafter referred to as “reliability”) A1 of the result of the speech recognition unit 16 recognizing the related word RW selected by the selection unit 162 for each related word RW. Since the higher the evaluation value SC, the higher the validity of the speech recognition result, the relative ratio of the evaluation value SC of the related word RW to the sum of the evaluation values SC calculated for a plurality of words in the speech recognition. (Evaluation value SC of related word RW / total sum of evaluation values SC) is calculated as reliability A1. The speech recognition unit 16 sequentially outputs the character string DB of the related word RW selected by the selection unit 162, the utterance time T1 of the related word RW, and the reliability A1 for each related word RW. The above is the content of the processing by the voice recognition unit 16.

  The speaker identifying unit 22 divides the audio signal S into a plurality of sections (hereinafter referred to as “speaking sections”) on the time axis for each speaker. For example, the speaker identification unit 22 classifies acoustic feature quantities (for example, MFCC) extracted for each frame of the audio signal S into a plurality of sets (clusters). That is, another cluster is set for each speaker. Furthermore, the speaker identification unit 22 classifies each of the plurality of frames of the speech signal S into clusters defined by a center vector having the most similar feature amount (the distance is minimum) among the plurality of clusters. By doing so, the audio signal S is divided into utterance intervals for each speaker (each cluster).

  Further, the speaker identification unit 22 calculates the accuracy (hereinafter referred to as “reliability”) A2 of the classification result by the speaker identification unit 22 for each utterance section. For example, the reliability A2 of the utterance section belonging to one cluster is calculated as the reciprocal of the average value of the distance between the feature quantity extracted from each frame in the utterance section and the center vector of the cluster. The speaker identifying unit 22 uses the identification code I uniquely assigned to the speaker (each cluster) in each utterance section, the time T2 of the start and end points of the utterance section, and the reliability A2 of the utterance section for each utterance section. Are output sequentially.

  The display control unit 24 causes the display device 44 to display the character string DB of each related word RW identified by the voice recognition unit 16. FIG. 3 is a schematic view illustrating a screen (hereinafter referred to as “search result display screen”) 442 displayed on the display device 44. As shown in the figure, the search result display screen 442 defines a region R (R1 to R3) for each speaker identified by the speaker identification unit 22. In addition, a speaker's identification code I (I1 to I3) corresponding to each region R is arranged in the vicinity of the region R, and the time with the start point of the audio signal S as a reference (0:00) is downward from above. It arranges at equal intervals toward.

  The display control unit 24 arranges the character string DB of each related word RW identified by the voice recognition unit 16 in the region R corresponding to the speaker of the related word RW. That is, the character string DB of each related word RW is displayed separately for each speaker. More specifically, the display control unit 24 utters an utterance section including the time T1 of each related word RW among the plurality of utterance sections specified at the time T2 specified by the speaker identification unit 22 (that is, the related word RW is uttered). And the character string DB of the related word RW is arranged at a position corresponding to the time T1 in the speaker's area R of the utterance section. For example, in FIG. 3, the related word RW “reduction” specified for the designated word KW “cost” is selected from the time “0:30” to “1:00”. The case where the voice is uttered in the section is illustrated. The display control unit 24 displays the character string DB of the related word RW in a different color for each region R (for each speaker).

  The display control unit 24 calculates the reliability A0 for each related word RW identified by the voice recognition unit 16. The reliability A0 is the sum of the reliability A1 specified by the speech recognition unit 16 for the related word RW and the reliability A2 specified by the speaker identification unit 22 for the utterance section including the time T1 of the related word RW (or Weighted sum). The display control unit 24 causes the display device 44 to display the character string DB of each related word RW in a manner corresponding to the reliability A0 of the related word RW. For example, the display control unit 24 displays the character string DB of the related word RW with a size corresponding to the reliability A0. In the example of FIG. 3, the reliability A0 of the related word RW “price” uttered in the section from the time “1:00” to “1:15” by the speaker of the identification code I2 is the utterance of the identification code I2. Higher than the reliability A0 of the related word RW “price” uttered in the section from the time “0:15” to “0:30” by each of the speaker and the speaker of the identification code I3 (therefore, the character string The size of DB is large).

  The user can specify any related word RW arranged on the search result display screen 442 from the input device 42. The playback control unit 26 in FIG. 1 sequentially selects a section (portion corresponding to the related word RW) after the time corresponding to the time T1 of the related word RW specified by the user in the audio signal S stored in the storage device 30. The sound is output to the sound emitting device 46. Therefore, the sound of the part including the related word RW is output from the sound emitting device 46. Note that a time point that is a predetermined time length before the time T1 of the related word RW may be set as a start point of playback by the playback control unit 26.

  As described above, in the present embodiment, since the appearance probability P of each related word RW among a plurality of words increases, the appearance probability P of each related word RW is used for voice recognition with the initial value. Compared to the above, the possibility of erroneous recognition of the audio signal S is reduced. Moreover, since a plurality of related words RW corresponding to the designated word KW are specified, compared with a configuration in which only the designated word KW is searched from the speech signal S, a wide range of words (related words reflecting the user's intention) There is also an advantage that (RW) is searched efficiently.

  Further, since the speech signal S is classified for each speaker and the character string DB of the related word RW is displayed separately for each speaker, there is an advantage that the user can easily grasp the speaker of the related word RW. is there. Further, since the character string DB of the related word RW is arranged at a position corresponding to the time T1 in the region R, the user can intuitively know the time when each speaker speaks the related word RW and the utterances of each speaker. There is also an advantage of being able to grasp. In addition, since the character string DB of the related word RW is displayed in a manner corresponding to the recognition reliability A0 (A1, A2) of each related word RW, the user intuitively determines the reliability A0 of each related word RW. In addition, it is possible to efficiently use the related word RW having a high reliability A0 in order.

<Modification>
Various modifications can be made to the above embodiment. An example of a specific modification is as follows. Two or more aspects may be arbitrarily selected from the following examples and combined.

(1) Modification 1
In the above embodiment, the configuration in which the voice recognition unit 16 includes the selection unit 162 is illustrated, but a configuration in which the display control unit 24 includes the selection unit 162 is also employed. For example, the voice recognition unit 16 sends the character string DB, the time T1, and the reliability A1 to the display control unit 24 for each of the words recognized from the voice signal S (words other than the related word RW and the related word RW). Output. The selection unit 162 of the display control unit 24 selects the related word RW from the words notified from the voice recognition unit 16 and causes the display device 44 to display the character string DB. According to the configuration of FIG. 1, since the related word RW is extracted from the result of speech recognition before the character string DB, time T1, and reliability A1 are specified, the character string DB and time T1 are related only to the related word RW. And the reliability A1 are sufficient (therefore, the processing amount is reduced).

(2) Modification 2
For convenience, the configuration in which the probability adjustment unit 14 is a separate element from the speech recognition unit 16 is illustrated, but the function of the probability adjustment unit 14 may be provided in the speech recognition unit 16. For example, when each of a plurality of words is sequentially selected and the evaluation value SC of the word is calculated, the speech recognition unit 16 increases the appearance probability P of the word if the selected word is the related word RW. On the other hand, while the evaluation value SC is calculated, if the selected word is other than the related word RW, the evaluation value SC is calculated while maintaining the appearance probability P of the word at the initial value.

(3) Modification 3
The configuration for selectively outputting (searching) each related word RW among the plurality of words recognized by the voice recognition unit 16 is not essential in the present invention. For example, a configuration in which the display control unit 24 sequentially outputs the character string DB of each word (words other than the related word RW and the related word RW) specified by the voice recognition unit 16 from the voice signal S from the display device 44 is also adopted. The Since the appearance probability P of each related word RW corresponding to the designated word KW rises from the initial value, even if the configuration is such that the character string DB of each word specified from the speech signal S is output, there is a possibility of erroneous recognition. The expected effect of efficiently recognizing each related word RW while suppressing is certainly achieved. As described above, the selection unit 162 (an element for searching for the related word RW from the speech signal S) that selects the related word RW from the plurality of words recognized by the voice recognition unit 16 is appropriately omitted. Further, the speaker identification unit 22 that classifies the voice signal S for each speaker may be omitted. In the configuration in which the speaker identifying unit 22 is omitted, the character strings DB of the words recognized by the speech recognizing unit 16 are displayed in time series (not distinguished for each speaker).

  Further, the reproduction control unit 26 is also omitted. However, according to the configuration of FIG. 1 including the reproduction control unit 26, the user can easily confirm the sound at a desired time (when the desired speaker in the sound signal S utters a specific word). There is. In addition, since the speaker identification part 22 in the above form only distinguishes the audio | voice signal S for every speaker (it does not specify each speaker), the user visually recognized the search result display screen 442. It is not possible to specify who the speaker in each region R is. However, according to the configuration including the reproduction control unit 26, the user can specifically identify the speaker of the related word RW in each region R by listening to the reproduction sound of the audio signal S.

(4) Modification 4
In the above-described embodiment, the configuration in which the speaker identifying unit 22 classifies the audio signal S for each speaker is exemplified, but a configuration in which the speaker identifying unit 22 executes until the speaker is specified is also suitable. For example, a feature model (for example, Gaussian mixture model) extracted from the uttered sound and the name of each utterer are stored in advance in the storage device 30 for each utterer. The speaker identification unit 22 identifies the name of the speaker in each utterance section of the speech signal S by comparing the feature amount extracted from the speech signal S with the model of the feature amount stored in the storage device 30. The name of the speaker is displayed in the vicinity of each region R on the search result display screen 442. According to the above configuration, the user can specify the speaker of each related word RW without listening to the reproduced sound of the audio signal S.

(5) Modification 5
In the above embodiment, the probability adjustment unit 14 increases the appearance probability P of each related word RW, but the probability adjustment unit 14 decreases the appearance probability P of words other than the related word RW (appearance of each related word RW). A configuration in which the probability P is maintained at an initial value is also adopted. However, since the number of words other than the related word RW is sufficiently larger than that of the related word RW, the appearance probability P of the words other than the related word RW is adjusted according to the form of adjusting the appearance probability P of the related word RW. As compared with the above, there is an advantage that the processing amount by the probability adjusting unit 14 is reduced.

(6) Modification 6
The configuration for calculating the reliability A0 (A1, A2) is not essential in the present invention. Therefore, the configuration for variably controlling the character string DB of each related word RW can be omitted in the present invention. In addition, a configuration in which the display control unit 24 controls the mode of the character string DB based only on the reliability A1 (a configuration in which the calculation of the reliability A2 is omitted), and a character string based on only the reliability A2 in the display control unit 24. A configuration for controlling the mode of DB (a configuration in which calculation of reliability A1 is omitted) is also employed.

It is a block diagram which shows the structure of the speech recognition apparatus which concerns on embodiment of this invention. It is a conceptual diagram which shows the structure of a recognition dictionary typically. It is a conceptual diagram which shows the content of the search result display screen.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 ... Voice recognition apparatus, 10 ... Control apparatus, 12 ... Related word specific | specification part, 14 ... Probability adjustment part, 16 ... Voice recognition part, 162 ... Selection part, 22 ... Speaker identification part, 24 ...... Display control unit, 26 ... Reproduction control unit, 30 ... Storage device, 42 ... Input device, 44 ... Display device, 46 ... Sound emitting device, KW ... Designated word, RW ... Related word, S: Speech signal, C: Co-occurrence database, D: Recognition dictionary, DA: Phoneme string, DB: Character string, P: Appearance probability, G: Phoneme model group.

Claims (4)

A related word specifying means for specifying a plurality of related words for a word specified by a user;
Storage means for storing an appearance probability for each of a plurality of words;
Probability adjusting means for relatively increasing the appearance probability of each related word specified by the related word specifying means among the plurality of words with respect to the appearance probability of words other than the plurality of related words;
Speech recognition means for specifying a word corresponding to the speech represented by the speech signal based on the appearance probability after adjustment by the probability adjustment means ;
Selecting means for selecting the related word from a plurality of words specified by the voice recognition means;
Speaker identification means for dividing the audio signal into utterance intervals for each speaker;
The character string of each related word is displayed for each speaker by specifying the utterance section in which each of the related words selected by the selecting means is uttered among the plurality of utterance sections divided by the speaker identifying means. Display control means for displaying on
The speech recognition means calculates the reliability of a specific result by the speech recognition means for each related word selected by the selection means,
The speaker identification means calculates the reliability of the result of the classification by the speaker identification means for each utterance section,
The display control means includes the reliability calculated by the speech recognition means for the related word for the character string of each related word and the reliability calculated by the speaker identification means for the utterance section in which the related word is uttered. A voice recognition device that displays on the display device in a manner according to the above . The sound according to claim 1 , further comprising: a reproduction control unit that outputs a sound of a portion corresponding to the related word in the sound signal from a sound emitting device when a user specifies the related word displayed on the display device. Recognition device. The reproduction control means outputs, from the sound emitting device, a sound of a portion starting from a time point a predetermined time before the time of the related word specified by the user in the audio signal.
The speech recognition apparatus according to claim 2 . A computer comprising storage means for storing the probability of occurrence of each of a plurality of words;
A related word specifying process for specifying a plurality of related words for a word specified by a user;
A probability adjustment process for relatively increasing the appearance probability of each related word specified in the related word specifying process among the plurality of words with respect to the appearance probability of words other than the plurality of related words;
A speech recognition process for specifying a word corresponding to the speech represented by the speech signal based on the appearance probability after the probability adjustment process ;
A selection process for selecting the related word from a plurality of words specified in the voice recognition process;
Speaker identification processing for dividing the voice signal into utterance sections for each speaker;
A device for displaying a character string of each related word for each speaker by identifying a utterance section in which each related word selected in the selection process is uttered from among a plurality of utterance sections divided in the speaker identification process A program for executing display control processing to be displayed on
In the voice recognition process, for each related word selected in the selection process, the reliability of a specific result by the voice recognition process is calculated,
In the speaker identification process, the reliability of the result of the classification by the speaker identification process is calculated for each utterance section,
In the display control process, the reliability of the string of each related word calculated by the speech recognition process for the related word and the reliability calculated by the speaker identification process for the utterance section in which the related word is uttered A program to be displayed on the display device in a manner according to the above .

Images