JP2010117528A - Vocal quality change decision device, vocal quality change decision method and vocal quality change decision program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To automatically determine vocal quality change of a speaker, without increasing a speech amount of the speaker. <P>SOLUTION: A vocal quality decision device 1 includes: a voice input unit 2 which accesses a reference data recording unit 7 for recording a reference feature value and a reference text, and which inputs voice of the speaker; a text input unit 3 for input of a text corresponding to input voice, a feature value calculating unit 4 for calculating a feature value of the input voice; a vocal quality decision unit 5 for determining vocal change of the input voice by comparing the feature value of the input voice corresponding to a part where the input text matches the reference text with the reference feature value; and a reference registration unit 6 for recording a text corresponding to the input voice in which the vocal quality decision unit 5 determines that the vocal change is within an allowable range, together with the corresponding feature value, as a new reference text and a new reference feature value, in the reference data recording unit 7. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present application relates to a voice quality change determination device, a voice quality change determination method, and a voice quality change determination program for determining a voice quality change of a speaker.

  When recording speech such as a speech synthesis database or recording / playback content, it is necessary to record a large amount of speech by a speaker with uniform voice quality. However, when recording is performed for a long time, the voice quality of the speaker changes over time due to fatigue. Further, when recording is performed over a plurality of days, the voice quality of the speaker changes according to the recording date. Conventionally, a method has been adopted in which the supervisor records while listening to the voice of the speaker. In this method, if the voice quality of the speaker changes, the supervisor will detect it and have the speaker re-record on the spot. It is difficult to equalize the voice quality by such a method that relies on the supervisor's sense of hearing. For this reason, a change in voice quality is revealed by the audition after recording, and a situation such as re-recording at a later date may occur. That is, there is a problem that cost costs and time costs are lost.

Therefore, a technique for determining a voice quality change of a speaker without depending on human hearing (see, for example, Patent Document 1). In this prior art, a specific voice is registered in advance, the voice is recognized by comparing the specific voice with the input voice, and the physical condition is determined according to the voice recognition rate.
JP 2006-230548 A

  However, in the above prior art, the voice quality change cannot be inspected unless the speaker makes a specific utterance. According to this method, since the speaker is forced to make a specific utterance for voice quality determination during the recording of the voice, the amount of utterance is unnecessarily increased. As a result, the speaker is tired and the voice quality of the speaker is changed.

  In view of the above problems, the present invention provides a voice quality change determination device, a voice quality change determination method, and a voice quality change determination program that enable automatic determination of a voice quality change of a speaker without increasing the amount of voice of the speaker. Objective.

  A voice quality change determination device disclosed in the present application is a voice quality change determination device that is accessible to a reference data recording unit that records a reference feature amount representing a reference voice quality and a reference text corresponding to the reference feature amount. A voice input unit that inputs a user's voice as input voice, a text input unit that inputs text corresponding to the input voice as input text, a feature amount calculation unit that calculates a feature amount of the input voice, and the input A voice quality that determines whether or not the voice quality change of the input voice is within a predetermined allowable range by comparing a feature quantity of the input voice corresponding to a portion where the text and the reference text match each other with a reference feature quantity A new reference text and an input text corresponding to the input speech determined by the determination unit and the voice quality determination unit that the voice quality change is determined to be within the allowable range. And a reference registration unit for recording the reference data recording section as a fine reference feature amount.

  The voice quality determination unit can determine the change in the voice quality of the input voice by comparing the feature quantity of the input voice with the reference feature quantity in a portion where the input text and the reference text match each other. The voice quality determination unit outputs the determination result.

  Further, the reference registration unit records the feature amount of the input speech and the corresponding text for which the voice quality change is determined to be within an allowable range by the voice quality determination unit as a new reference text and a reference feature amount. Therefore, additional registration of the reference feature amount and the reference text is automatically performed with the determination of the voice quality change of the input voice. That is, it is possible to add information serving as a reference for determining a voice quality change without increasing the utterance amount of the utterer. As a result, the reference data is expanded without unnecessarily increasing the utterance amount of the utterer, and the voice quality change of the utterer can be automatically determined. As a result, voice recording with stable voice quality is realized.

  The voice quality can be expressed by, for example, volume, speaking speed, voice pitch, intonation, timbre, or a combination thereof.

  In an embodiment of the present invention, the voice quality change determination device further has access to a text recording unit in which text to be input is pre-registered, and the text input unit accesses the text recording unit. The input text is read, the input text is output to the speaker, and the voice input unit can input the voice of the speaker who reads the output text.

  In an embodiment of the present invention, the voice quality change determination device further includes a language processing unit that performs language analysis on the input text and divides the input text into analysis units based on a result of language analysis, and the feature amount calculation unit includes: A feature amount corresponding to each analysis unit divided by the language processing unit is calculated, and the voice quality determination unit includes the feature amount and the reference of the input speech corresponding to the analysis unit in which the input text and the reference text match each other. By comparing the feature quantities, it can be determined whether or not the voice quality change of the input speech is within a predetermined allowable range.

  With the configuration described above, it is possible to calculate a feature amount and determine a voice quality change in an appropriate analysis unit based on language analysis. As a result, the determination accuracy can be further improved. The language analysis is a process of dividing text into meaningful units for language processing, such as morphological analysis. The analysis unit is a unit suitable for determining the voice quality change, and is not necessarily the same as the unit divided as a result of the language analysis.

  In an embodiment of the present invention, the reference data recording unit further includes a conversion unit that records the reference text as phonetic text and converts the input text into phonetic text, and the voice quality determination unit converts the phonetic text into phonetic text. The voice quality change of the input voice can be determined by comparing the feature quantity of the input voice corresponding to the portion where the converted input text and the reference text match each other and the reference feature quantity.

  As a result, the voice quality change is determined by comparing the feature quantity of the input speech with the reference feature quantity for the portion where the phonetic text matches, so the portion where the phonetic text does not match is excluded from the determination target. . Therefore, more accurate determination can be performed.

  The phonetic text may be any text that can be handled as information representing reading when the computer processes it.

  In an embodiment of the present invention, the reference registration unit adds an existing speech feature corresponding to a portion where the text and the reference text match to the reference data recording unit, and the existing reference feature corresponding to the portion. A quantity is replaced with a feature quantity of the input speech, or a feature quantity calculated using the existing reference feature quantity and the feature quantity of the input speech is added as a new reference feature quantity be able to.

  In an embodiment of the present invention, the reference registration unit includes an analysis unit in which the input text and the reference text match each other for the input speech for which the voice quality determination unit determines that the voice quality change is within an allowable range, and the analysis unit. The input text and the feature amount corresponding to the analysis unit adjacent to can be recorded in the reference data recording unit as a new reference text and reference feature amount.

  Thereby, the vicinity of the place where the input text and the reference text match can be newly registered as the reference text and the reference feature amount. Therefore, it is possible to appropriately select a feature amount of a portion with less change in voice quality compared to the existing reference speech and add it to a new reference feature amount.

  An audio recording device including the voice quality change determination device according to the embodiment, wherein the input of the input voice is determined by the voice quality determination unit of the voice quality change determination device to be within an allowable range. An audio recording apparatus including a recording unit that records audio as recorded data is also included in the embodiment of the present invention.

  A voice quality change determination program that causes a computer to function as the voice quality change determination device by causing the computer to execute voice quality change determination processing is also one embodiment of the present invention.

  A voice quality change determination method executed by a computer to realize the function of the voice quality change determination apparatus is also one embodiment of the present invention.

  According to the present disclosure, since it is possible to accurately and automatically determine the voice quality change of the speaker without relying on the human ear, the voice of the voice of the speaker is stable without increasing the amount of useless voice of the speaker. It becomes possible to record.

(First embodiment)
[Configuration of voice quality change determination device]
FIG. 1 is a functional block diagram illustrating a configuration of a voice quality change determination device according to the first embodiment. A voice quality change determination device 1 shown in FIG. 1 is a device that inputs a voice of a speaker, determines the voice quality of the speaker, and outputs a warning when the change in voice quality is not within an allowable range. Therefore, the voice quality change determination device 1 includes a voice input unit 2, a text input unit 3, a feature amount calculation unit 4, a voice quality determination unit 5, a reference registration unit 6, and a reference data recording unit 7.

  Although not shown, the voice quality change determination device 1 inputs output information for outputting a warning such as a speaker or a display, and input information such as a mouse, a keyboard, a tablet, a button, a scanner, or a camera. The input device may be provided.

  The voice quality change determination device 1 can be realized, for example, by installing a predetermined program in a general-purpose computer such as a personal computer or a server machine. In addition, the voice quality change determination device 1 may be formed by a computer incorporated in an electronic device such as an in-vehicle information terminal, a mobile phone, a PDA (Personal Digital Assistant), and a home appliance, for example. Alternatively, the voice quality change determination device 1 may be formed by a dedicated device incorporating a computer.

  Each function of the voice input unit 2, the text input unit 3, the feature amount calculation unit 4, the voice quality determination unit 5, and the reference registration unit 6 is realized by the CPU executing a predetermined program. Therefore, a program for realizing the above functions by a computer or a recording medium on which the program is recorded is also an embodiment of the present invention. The reference data recording unit 7 is voice data input with a stable voice quality, and is realized by a built-in storage device of a computer or a storage device accessible from the computer.

  The configuration of the voice quality change determination device 1 is not limited to the example illustrated in FIG. For example, the reference data recording unit 8 may be realized by a storage medium provided outside the voice quality change determination device 1. Hereinafter, each functional unit will be described in detail.

[Voice input unit 2]
The voice input unit 2 inputs the voice of the speaker from the outside, and sets the voice as a voice that can be processed by the voice quality change determination device 1. For example, the voice input unit 2 takes the voice of the speaker into the voice quality change determination device 1 via a recording microphone and passes it to the feature amount calculation unit 4. Alternatively, the voice input unit 2 may input the voice of the speaker by reading the voice data of the speaker recorded in advance on an external or built-in recording medium, or by receiving the voice data via a network. .

[Text input part 3]
The text input unit 3 inputs a text indicating the content of the voice of the speaker from the outside, and sets the input text to a state that can be processed by the voice quality change determination device 1. The text input unit 3 inputs, for example, the contents of a reference sentence or a reading document as input text. The input text format is preferably a phonetic text representing kanji reading or accent position information, but may be a normal text such as a kana-kanji mixed text.

  In addition, the text input unit 3 displays the text to the speaker via, for example, the GUI, and causes the voice input unit 2 to input the voice of the speaker who reads the displayed text. Can be identified. Alternatively, the text input unit 3 may accept input of text corresponding to the input voice from a speaker or another user via the GUI, or the input voice is recognized by recognizing the input voice of the voice input unit 2. Corresponding input text may be generated. Therefore, the sound quality change determination device 1 may further include, for example, a voice recognition unit that recognizes an input voice and generates an input text.

[Feature amount calculation unit 4]
The feature amount calculation unit 4 calculates a feature amount from the input voice and the input text, and outputs the feature amount to the voice quality determination unit 5 or the reference data recording unit 7 together with the corresponding input text. The feature amount in the present embodiment is an amount obtained by quantifying the acoustic and phonological features of speech. The feature amount can be obtained, for example, by calculating at least one of volume, speaking speed, voice pitch, intonation, timbre, or phonology. Specific examples of the feature amount are shown below, but the feature amount is not limited to these.

  For the volume, for example, an average amplitude value in a voiced section can be used as a feature amount. As the average amplitude value, an average absolute amplitude value, an RMS value, or the like can be adopted. For the speech speed, for example, the number of syllables or the number of beats (number of mora) per unit time (second) in a voiced section can be used as a feature amount. Regarding the pitch of the voice, for example, the pitch frequency (Hz) of the voiced section in which the speech waveform has periodicity among the voiced sections of the voice can be used as the feature amount. As for the inflection, for example, the change amount (dynamic range) of the pitch frequency, which is the pitch of the voice, can be used as the feature amount. For the timbre, for example, a formant frequency in a voiced section of speech can be used as a feature amount. For phonemes, for example, phoneme notation obtained by speech recognition of speech can be used as the feature amount.

[Switching means 8, 9]
In the example shown in FIG. 1, switching means 8 and 9 are provided for switching whether the feature quantity and text output destination from the feature quantity calculation unit 4 is the voice quality determination unit 5 or the reference data recording unit 7. The switching means 8 and 9 can connect the feature amount calculation unit 4 to B and D, for example, when the speaker registers the feature amount and text of the speech that is the reference for voice quality determination. Thereby, the feature amount and the text from the feature amount calculation unit 4 are output to the reference data recording unit 7. In other cases, the switching means 8 and 9 may connect the feature quantity calculation unit 4 to A and C. As a result, the feature amount and text of the input voice are output to the voice quality determination unit 5, and the voice quality change is determined.

  For example, the speaker determines the volume, speaking speed, voice pitch, inflection, timbre, etc. in the first recording of the voice, reads out the reference sentence as the reference voice, and sets the feature amount of the voice as the reference voice. It is assumed that the feature amount and the read-out document are registered as the reference text. In this case, the feature amount calculation unit 4 is connected to B and D only when the speaker reads out the sentence that becomes the reference text and registers the reference text and the reference feature amount. In other cases, the feature amount calculation unit 4 is connected to A and C. The switching means 8 and 9 can be controlled to connect. Here, the reference text registered for the first time may be one sentence, but it is preferable to prepare a plurality of sentences since the accuracy of the voice quality change determination of the speaker increases as the number of sentences increases.

  Note that the switching units 8 and 9 do not necessarily need to be physical switches, and may switch data output destinations by software processing, for example.

[Reference data recording unit 7]
The reference data recording unit 7 records the reference feature amount of the voice, which becomes the reference voice quality in the determination of the voice quality change, and the reference text in association with each other. The data format of the reference feature amount and the reference text is not particularly limited, but may be recorded in a table format, for example. The reference text is preferably recorded in a unit suitable for calculating a feature quantity and determining a voice quality change, for example, as a phrase unit.

[Voice quality determination unit 5]
The voice quality determination unit 5 determines the change in voice quality by comparing the feature amount of the input speech and the corresponding text output from the feature amount calculation unit 4 with the reference feature amount and the reference text read from the reference data recording unit 7. Execute. For example, the voice quality determination unit 5 can search a portion of the input text that matches the reference text recorded in the reference data recording unit 7 and compare the feature amount with the reference feature amount for the matching portion. . The voice quality determination unit 5 can calculate a value (voice quality change amount) indicating the degree of voice quality change by this comparison. As an example, the voice quality change amount can be calculated from the difference or ratio between the feature amount and the reference feature amount. When the feature amount includes a plurality of items such as volume, speech speed, voice pitch, intonation, timbre, phonology, etc., the voice quality change amount may be calculated for each item.

  The voice quality determination unit 5 determines whether or not the voice quality change amount exceeds the allowable range. This determination can be made, for example, by comparing the voice quality change amount with a preset threshold value. If the voice quality change amount exceeds the allowable range, a warning is output to the speaker. The warning can be communicated to the speaker via an output device such as a display or speaker, for example. Note that the determination result output by the voice quality determination unit 5 is not limited to a warning. For example, the voice quality determination unit 5 may output the calculated voice quality change amount as it is as a determination result.

[Standard registration unit 6]
The reference registration unit 6 records, in the reference data recording unit 7, the feature amount and text of the input voice that the voice quality determination unit 5 has determined that the voice change amount is within the allowable range as the reference feature amount and the reference text. At that time, a part of the feature amount and text may be extracted and recorded from the input speech determined that the speech change amount is within the allowable range.

  For example, when the input speech is input for each sentence and the voice quality change is determined for each sentence, the text and the feature amount of the entire sentence are recorded in the reference data recording unit 7 for the sentence determined to be within the allowable range. Alternatively, a portion of the sentence that matches the existing reference text and its peripheral portion may be recorded as the reference text and the reference feature amount.

  That is, the reference registration unit 6 can select a part to be added as the reference text based on the part where the input text matches the existing reference text. As a result, it is possible to add a portion of the text and feature amount suitable for the reference to the reference data recording unit with little change in voice quality.

[Operation example]
Next, an operation example of the voice quality change determination device 1 will be described. FIG. 2 is a flowchart showing an operation example of the voice quality determination unit 5 and the reference registration unit 6. The process illustrated in FIG. 2 is an example of a process in which the voice quality determination unit 5 determines a voice quality change from the feature amount and the reference feature amount of the portion where the text and the reference text match.

  First, the feature quantity and the input text of the input voice are input from the feature quantity calculation section 4 to the voice quality determination section 5 (Op1). Here, as an example, a case where an input text and an input feature amount are input in units of one sentence and sound quality change determination is performed for each sentence will be described. That is, in Op1, a set of input text of one sentence and a feature amount of input speech that reads out the one sentence is input. In this embodiment, the input text and the voice are input at the same time. However, the input text may be registered first, the registered input text may be displayed, and the voice may be input.

  Next, the voice quality determination unit 5 searches for a reference text that matches the text included in the input text of one sentence (one analysis unit) (Op2).

  As an example, if the speaker speaks the manuscript shown in Figure 3A, the input text will be “Check the weather yesterday and today.”, “Tell me your phone number.” And “There is no tomorrow. "Is input. For example, when one sentence “Please tell me the weather tomorrow” shown in FIG. 3B is recorded in the reference data recording unit 7 as the reference text, the text that matches the reference text in the above three sentences is , "Weather", "Tell me" and "Tomorrow".

  If there is no matching reference text (No in Op2), the voice quality determination unit 5 inputs the next set of input text and feature amount and continues the search (repeats the processing of Op1 and Op2). If there is no further set of input text and input feature quantity (No in Op7), the voice quality determination unit 5 ends the process.

  When there is a text that matches the reference text in the input text (Yes in Op2), the voice quality determination unit 5 calculates a voice quality change by comparing the matching feature quantity with the reference feature quantity (Op3). For example, the voice quality determination unit 5 can calculate the voice quality change amount from the difference or ratio between the feature value and the reference feature value. Here, a case will be described in which the feature amount includes values of a plurality of items such as the above-described volume, speech speed, voice pitch, intonation, timbre, and phonology. For example, when the number of items is n (n = natural number), voice quality change amounts K (1), K (2),... K (n) are calculated for each feature amount item (1,... N). Is done.

  Next, the voice quality determination unit 5 includes one or more voice quality change amounts K (1),..., K (n), and threshold values T (1),. T (n) is compared to determine whether or not there is a change in voice quality (Op4-1 to n). For example, when the voice quality change amount of at least one item of the one or more voice quality change amounts exceeds a threshold value (in the case of No in any of Op4-1 to n), the voice quality determination unit 5 determines the voice quality of the voice. It determines with having changed, and the information is output as a warning (Op6). In addition, the voice quality determination unit 5 does not change the voice quality of voice when all the voice quality change items do not exceed the threshold value or are equal to the threshold value (when all of Op4-1 to n are Yes), and appropriate voice The reference registration unit 6 registers the input text and its feature amount in the reference data recording unit 7 (Op5).

  Here, a specific example of determination by the voice quality determination unit 5 and registration by the reference registration unit 6 will be described. Here, an example will be described in which the example sentences 1) and 3) of the manuscript shown in FIG. 3A are determined as “no change in voice quality” and the example sentence 2) is determined as “with change in voice quality”. For the example sentence 2), the voice quality determination unit 5 notifies the speaker that the voice quality has changed via the display and the speaker. In this case, the voice can be re-recorded by displaying the example sentence 2) again to prompt the speaker to input the voice again.

  On the other hand, for the sentences of example sentences 1) and 3), since it is determined that there is no change in voice quality, the texts and feature quantities of these sentences are recorded in the reference data recording unit 7 as new reference texts and feature quantities. .

  4A and 4B are diagrams showing examples of reference text after new registration. In the example shown in FIG. 4A, the entire text of the sentence of example sentence 1) including “weather” matches the reference text and the entire text of the sentence of example sentence 3) including “Tomorrow” correspond as new reference texts. It is registered together with the feature amount.

  FIG. 4B shows an example in which not only the whole text of sentences in example sentences 1) and 3), but only the vicinity of the portion where the reference text matches is newly registered as the reference text. In the example shown in FIG. 4B, in the case of example sentence 1), only the phrases “today” and “check” before and after “weather” are registered as reference texts. The phrase “Yesterday” that is far from the matching part is not registered. In the example sentence 3), only the phrase “schedule” after “Ashitano” that matches the reference text is registered as a new reference text, and the phrase “None” that is far from the matching part is not registered.

  As described above, the reference registration unit 6 can select the text and the feature amount in the vicinity of the matching portion with a small change in voice quality, and can register the new reference text and the reference feature amount. As a result, it is possible to prevent the text and the feature amount of the portion away from the portion where the input text and the reference text match, that is, the portion where the voice quality is likely to change, from being added to the reference text and the reference feature amount. it can.

  Note that, in the above example, the reference registration unit 6 sets the clauses adjacent to the matching portion to be included in the reference text, but the range included in the reference text is not limited to this. For example, a predetermined number of words or a predetermined number of characters adjacent to the matching portion may be included in the reference text.

  As described above, by the operation of the voice quality change determination device 1 shown in FIG. 2, the voice text and the feature amount suitable for the reference voice are automatically added to the reference text and the reference feature amount among the input voices. Is done. For this reason, the reference text and the reference feature amount are expanded without increasing the amount of useless speech of the speaker. As a result, it becomes possible to more accurately and automatically determine changes in the voice quality (sound volume, speech speed, voice pitch, inflection, timbre, etc.) of the speaker.

  In the above example, the difference or ratio of n feature values is defined as n voice quality change amounts, and the presence or absence of voice quality change is determined by comparing the n voice quality change amounts with the n threshold values. The determination process is not limited to this. For example, n voice quality changes may be captured as an n-dimensional Euclidean space, and the presence or absence of voice quality changes may be determined by comparing the Euclidean distance with a predetermined threshold.

(Modification)
FIG. 5 is a diagram showing a modification of the configuration of the voice quality change determination device 1 in the embodiment. The voice quality change determination apparatus 1 shown in FIG. 5 inputs in advance text to be read by the speaker and displays it on the screen, and accepts as input speech the voice that the speaker who has read the displayed text reads the text. It is comprised as follows.

  Therefore, the voice quality change determination device 1 shown in FIG. 5 includes an input text output unit 17 that outputs the input text read by the text input unit 3 from the external text recording unit 16 to the speaker. Text to be read out by the speaker is recorded in the text recording unit 16 in advance. The text recording unit 16 may be provided inside the sound quality change determination device 1.

  The text input unit 3 passes the input text read from the text recording unit 16 to the input text output unit 17 and the feature amount calculation unit 4. The input text output unit 17 outputs the input text to the speaker. For example, the input text output unit 17 may display the input text on a display in units of sentences. At that time, a message prompting the speaker to read out the input text can be displayed at the same time. The speaker reads out the input text displayed on the display, and the voice input unit 2 inputs the voice of the speaker and passes it to the feature amount calculation unit 4. The input text and the input speech corresponding to the input text are input to the feature amount calculation unit 4.

  With the above configuration, the speaker can execute the voice quality change determination and automatically expand the reference data by simply reading the input text in accordance with the output of the input text output unit 17.

(Second Embodiment)
The second embodiment is a modification of the text input unit 3 in the first embodiment. FIG. 6 is a functional block diagram showing the configuration of the text input unit 3a in the present embodiment. A text input unit 3 a shown in FIG. 6 includes a language processing unit 11 and a conversion unit 12. The language processing unit 11 and the conversion unit 12 can access the language dictionary 13. The language dictionary 13 stores data necessary for the processing of the language processing unit 11 and the conversion unit 12. In the language dictionary 13, for example, notation, reading, part of speech, accent information, and the like of a plurality of words that can be included in the input text are recorded.

  The language processing unit 11 divides the input text into analysis units by performing language analysis using the language dictionary 13. For example, when the input text is a kana-kanji mixed text, the language processing unit 11 collates the input text with the language dictionary 13 and performs morphological analysis. Examples of the morphological analysis method include the Viterbi algorithm and the longest match method, but the morphological analysis method used in the present invention is not limited to a specific one. Language analysis is not limited to morphological analysis. For example, when the target language is English, word analysis for dividing the input text in units of words may be executed.

  The language processing unit 11 can divide the input text into element units having meaning in language processing by morphological analysis. For example, boundaries such as parts of speech, phrases, accent phrases or exhalation paragraphs (punctuation marks) are detected. The language processing unit 11 can set one of these boundaries as an analysis boundary. The input text is divided into analysis units by the analysis boundary.

  Here, it is preferable that the analysis boundary is determined so as to set a range suitable for calculating the feature amount and determining the voice quality change. For example, the boundary between clauses may be an analysis boundary. This is because if the boundary that sets a section longer than a clause is set as an analysis boundary, such as a breath paragraph boundary delimited by punctuation marks, the probability that the text matches the reference text will be reduced, and automatic determination of voice quality changes and reference voice quality This is because new registration may be difficult. On the other hand, if the analysis unit is shortened, the probability that the text matches the reference text will be greatly increased, and therefore automatic determination of voice quality changes tends to be easier. However, if a boundary that sets a section shorter than a phrase, such as a part of speech part, is an analysis boundary, comparison of the feature quantity and the reference feature quantity in a short syllable unit often lacks reliability of accuracy. Therefore, the phrase boundary, which is the intermediate unit between the exhalation paragraph boundary and the part-of-speech boundary, can be said to be a valid analysis boundary as the number of syllables. The accent phrase has a length composed of one or several phrases, and it is also preferable that the boundary of the accent phrase is an analysis boundary.

  The conversion unit 12 converts the input text into phonetic text using the language dictionary 13. The conversion unit 12 may function as a part of the language processing unit 11. The phonetic text is text representing the reading of the input text, and may be added with an accent or the like.

  For example, when the language processing unit 11 executes morphological analysis of the input text, it can acquire phonetic information such as reading and accent of each word from the language dictionary 13 and generate phonetic text of each word. .

  Hereinafter, specific examples of processing of the language processing unit 11 and the conversion unit 12 will be shown. FIG. 7A is an example of a manuscript read by the speaker. FIG. 7B shows an example in which the input text of the manuscript content shown in FIG. 7A is divided into analysis units by the text input unit 3a and converted into phonetic text. Here is an example of the text:

  When the text input unit 3a inputs a kana-kanji mixed text as shown in FIG. 7A, the language processing unit 11 collates the text with the language dictionary and performs morphological analysis. With this morphological analysis, kana-kanji mixed text is broken down into phrases. Each phrase is converted into katakana notation by the conversion unit 12, and further accent information and the like are added to be converted into phonetic text.

  For example, in the example sentence 1) shown in FIG. 7A, “in the middle of the meeting” is decomposed into the analysis units “meeting” and “in the middle” by the language processing unit 11, and the phonetic text “ka It is converted into “Iguino” and “Situ Dess” (see FIG. 7B). In the phonetic text shown in FIG. 7B, as an example, the boundary between clauses is represented by a space, and the position (accent nucleus) where the accent shifts from “high” to “low” is represented by “′”.

  As a result, both the reference text recorded in the reference data recording unit 7 from the feature amount calculation unit 4 and the text passed from the feature amount calculation unit 4 to the voice quality determination unit 5 are both phonogram text decomposed into analysis units. It becomes. As described above, if both the reference text and the input text are phonetic texts, when searching for a matching portion between the input text and the reference text, in addition to text character matching, matching of reading and accent can be considered. Therefore, it is possible to distinguish between differently read words with the same notation and text with the same notation.

  For example, example sentences 1) and 2) shown in FIG. 7A include “middle” of the same notation, but “middle” reading of example sentence 1) is “situ” and “middle” of example sentence 2). The reading is “Monaka” and is completely different. Comparing the notation of texts with the same notation but different readings, even if there is no change in the voice quality of the speaker, there is a large difference in the overall feature quantity, and the voice quality determination unit 5 erroneously determines that “there is a change in voice quality” Resulting in.

  Similarly, example sentences 3) and 4) shown in FIG. 7A are read-out manuscripts including “weather”, but “weather” in example sentence 3) is a head-high accent with an accent in the “te” part, example sentence 4) The "weather" is a flat accent (no accent). The text notation of both is the same, but the accents do not match. When comparing feature quantities whose accents do not match, even if there is no change in the voice quality of the speaker, there is a large difference in the pitch frequency in particular, and the voice quality determination unit erroneously determines that “voice quality has changed”.

  Therefore, in the configuration of the present embodiment, when searching for a matching portion between the text and the reference text, it is possible to search not only for matching the text but also for matching phonetic information such as reading and accent. That is, it is possible to search for a matching portion between the text including the reading and accent information and the reference text.

  Specifically, the voice quality determination unit 5 searches the input text for a phrase in which the reading and accent represented by the phonetic text match the reference text, and if there is a phrase in which these match, Compare the feature quantity of the phrase with the reference feature quantity. This makes it possible to more accurately determine voice quality changes.

  In addition, since the search for the matching portion of the text and the reference text and the comparison of the feature quantity between the text and the reference text are executed in units of phrases, it is possible to perform a more accurate search and feature quantity comparison. Furthermore, the addition of the reference text and the reference feature amount by the reference registration unit 6 can also be executed in phrase units, so that more appropriate text and feature amount can be added to the reference data recording unit as a reference. .

  In the above embodiment, the case where the text input unit 3a includes the language processing unit 11 and the conversion unit 12 has been described. However, the text input unit 3a includes any one of the language processing unit 11 and the conversion unit 12. It may be a configuration.

(Third embodiment)
The third embodiment is a modification of the reference registration unit 6 in the first or second embodiment. In the present embodiment, the reference registration unit 6b adds (1) an existing feature corresponding to the matching portion when adding to the reference data recording unit 7 a feature value of the input speech corresponding to the portion where the input text and the reference text match. Whether to replace the reference feature quantity of the input with the feature quantity of the input speech Alternatively, (2) the reference registration unit 6b executes a process of adding the feature amount calculated using the existing reference feature amount and the feature amount of the input voice to the reference data recording unit 7 as a reference feature amount.

  Below, the specific example of the process by the reference | standard registration part 6b of this embodiment is demonstrated. Here, as an example, a case where the input text includes example sentences 1) to 3) shown in FIG. 8A and the existing reference text and feature amount recorded in the reference data recording unit 7 have the contents shown in FIG. 8B. explain. In this example, in any of the example sentences 1) to 3), the portion (phrase) of “Ashitano” matches the existing reference text. Here, it is assumed that the voice quality determination unit 5 determines that the voice quality change amount does not exceed the allowable range (“no voice quality change”) for any of the example sentences 1) to 3). In this case, the reference registration unit 6b can execute, for example, the following three patterns of registration processing.

  As the first registration process, the reference registration unit 6b may replace the existing reference feature amount of the portion where the input text and the reference text match with the feature amount of the input voice. For example, the reference registration unit 6b discards the existing reference feature value “A1” of “Ashitano” and sets the feature values A2, A3, and A4 of “Ashitano” in the example sentences 1), 2), and 3). You can select and register one of them. FIG. 9A is a diagram illustrating an example of the data content of the reference data recording unit 7 when the existing basic feature “A1” is replaced with the feature “A2” of “Ashitano” in the example sentence 1). In this way, by replacing the existing reference feature quantity with the feature quantity of the input voice, the speaker's situation can be reflected in the reference feature quantity. That is, the reference feature amount can be updated according to the speaker's situation.

  As the second registration process, the reference registration unit 6b can register a feature quantity calculated using the existing reference feature quantity and the input voice feature quantity as a reference feature quantity. For example, a feature value obtained by averaging the existing reference feature value and the feature value of the input voice by weighted averaging can be used as the reference feature value. FIG. 9B shows the reference data recording unit 7 when the existing reference feature amount A1 and the weighted average of the feature amounts A2, A3, and A4 of the example sentences 1), 2), and 3) are recorded as new reference feature amounts. It is a figure which shows an example of the data content of. In FIG. 9B, α, β, γ, and ε each represent a weighting coefficient. Thereby, the reference feature value can be brought closer to a more appropriate value.

  Note that the feature amount calculated using the existing reference feature amount and the feature amount of the input speech is not limited to the weighted average in the above example. For example, other representative values such as an average, an intermediate value (a value between the maximum value and the minimum value), a mode value, and the like may be calculated.

  As the third registration process, the reference registration unit 6b may continue to use the reference feature amount of the portion where the input text and the reference text match without being updated. By omitting the update process in this way, the load on the computer can be reduced. FIG. 9C is a diagram illustrating data contents of the reference data recording unit 7 when the reference feature amount A1 is not updated.

(Fourth embodiment)
The fourth embodiment relates to an audio recording device including the voice quality change determination device according to any one of the first to third embodiments. FIG. 10 is a functional block diagram showing the configuration of the audio recording device 10 in the present embodiment. The voice recording device 10 shown in FIG. 10 further has a function of recording input voice as recorded data. The audio recording device 10 has a configuration in which a recording unit 14 is added to the voice quality change determination device 1 shown in FIG. In FIG. 10, the same functional blocks as those in FIG.

  The recording unit 14 records the input voice input by the voice input unit 2 in the recorded data recording unit 15. At that time, the recording unit 14 can record only the voice that the voice quality determination unit 5 determines that the voice quality change is within the allowable range.

  As a result, the voice whose voice quality change exceeds the allowable range is automatically excluded from the recorded data. As a result, sound recording with stable sound quality is realized.

  As mentioned above, embodiment of this invention was illustrated as 1st-4th embodiment. According to the above embodiment, the speaker can read out the read-out manuscript, and the voice quality is automatically determined and the data used as a reference for the voice quality is expanded. And the burden on the speaker is reduced. That is, in the above-described embodiment, it is sufficient that the utterance of the speaker as the reference voice is at least about one sentence, and thereafter, the reference sound quality is automatically newly registered as the speaker continues to read the document. In particular, of the input voice read out by the speaker, the whole or a part of the voice determined to have no voice quality change is newly registered as the reference voice quality (text, feature amount). Thereby, the reference voice mass naturally increases. As a result, the text matching rate between the read-out original and the reference voice quality increases, and the voice quality determination accuracy is dramatically improved.

  In addition, the applicable range of this invention is not limited to said embodiment. Further, combinations of at least two of the first to fourth embodiments are also included in the embodiments of the present invention.

  Regarding the above first to fourth embodiments, the following additional notes are further disclosed.

(Appendix 1)
A voice quality change determination device accessible to a reference data recording unit for recording a reference feature amount representing a reference voice quality and a reference text corresponding to the reference feature amount,
A voice input unit for inputting a voice of a speaker as an input voice;
A text input unit for inputting text corresponding to the input voice as input text;
A feature amount calculation unit for calculating a feature amount of the input speech;
It is determined whether or not a change in voice quality of the input speech is within a predetermined allowable range by comparing a feature amount of the input speech corresponding to a portion where the input text and the reference text match each other with a reference feature amount. A voice quality determination unit,
A reference for recording, in the reference data recording unit, a new reference text and a reference feature amount together with the corresponding feature amount, the input text corresponding to the input speech for which the voice quality determination unit determines that the voice quality change is within an allowable range. A voice quality change determination device comprising a registration unit.

(Appendix 2)
The voice quality change determination device is further accessible to a text recording unit in which a text to be voice input is registered in advance,
The text input unit accesses the text recording unit, reads the input text, outputs the input text to the speaker,
The voice quality change determination apparatus according to claim 1, wherein the voice input unit inputs a voice of a speaker who reads the input text that has been output.

(Appendix 3)
Further comprising a language processing unit that linguistically analyzes the input text and divides the input text into analysis units based on a result of language analysis;
The feature amount calculation unit calculates a feature amount corresponding to each analysis unit divided by the language processing unit,
The voice quality determination unit compares a feature quantity of the input speech corresponding to an analysis unit in which the input text and the reference text match each other with a reference feature quantity, so that a change in voice quality of the input speech is within a predetermined allowable range. The voice quality change determination device according to supplementary note 1 or 2, which determines whether or not

(Appendix 4)
The reference data recording unit records the reference text as a phonetic text,
A conversion unit for converting the input text into phonetic text;
The voice quality determination unit compares a feature quantity of the input speech corresponding to a portion where the input text converted to the phonetic text and the reference text match each other and a reference feature quantity, thereby changing the voice quality change of the input speech. The voice quality change determination device according to any one of appendices 1 to 3, wherein the determination is performed.

(Appendix 5)
The reference registration unit adds an existing reference feature amount corresponding to the portion of the input speech when adding the feature amount of the input speech corresponding to the portion where the text and the reference text match to the reference data recording unit. Any one of appendices 1 to 4, wherein the feature quantity is replaced with a feature quantity, or a feature quantity calculated using the existing reference feature quantity and the feature quantity of the input speech is added as a new reference feature quantity. The voice quality change determination device described in 1.

(Appendix 6)
The reference registration unit corresponds to an analysis unit in which the input text and the reference text match each other and an analysis unit adjacent to the analysis unit for the input speech that the voice quality determination unit determines that the voice quality change is within an allowable range. The voice quality change determination device according to attachment 3, wherein the input text and the feature amount to be recorded are recorded in the reference data recording unit as a new reference text and a reference feature amount.

(Appendix 7)
An audio recording device including the voice quality change determination device according to any one of appendices 1 to 6,
A voice recording apparatus comprising: a recording unit that records, as recorded data, an input voice in which the change in voice quality is determined to be within an allowable range by the voice quality determination unit of the voice quality change determination device among the input voices.

(Appendix 8)
A voice quality change determination program for causing a computer accessible to a reference data recording unit that records a reference feature quantity indicating a reference voice quality and a reference text corresponding to the reference feature quantity to execute a voice quality change determination process,
Voice input processing for inputting the voice of the speaker as input voice,
Text input processing for inputting the text corresponding to the input voice as input text;
A feature amount calculation process for calculating the feature amount of the input speech;
It is determined whether or not a change in voice quality of the input speech is within a predetermined allowable range by comparing a feature amount of the input speech corresponding to a portion where the input text and the reference text match each other and a reference feature amount. Voice quality determination processing,
A reference for recording, in the reference data recording unit, a new reference text and a reference feature amount, together with the corresponding feature amount, the text corresponding to the input speech whose voice quality change is determined to be within an allowable range in the voice quality determination process. A voice quality change determination program that causes a computer to execute registration processing.

(Appendix 9)
A voice quality change determination method in which a computer executes voice quality change determination,
The computer can access a reference data recording unit that records a reference feature amount indicating a voice quality as a reference and a reference text corresponding to the reference feature amount,
A voice input step in which the computer receives a voice of a speaker as an input voice;
A text input step in which the computer acquires text corresponding to the input speech as input text;
A feature amount calculating step in which the computer calculates a feature amount of the input speech;
A voice quality determination step of determining a voice quality change of the input voice by comparing the feature quantity of the input voice and a reference feature quantity corresponding to a portion where the input text and the reference text coincide with each other;
The computer uses the whole or a part of the text corresponding to the input speech determined that the voice quality change is within an allowable range in the voice quality determination step as the new reference text and the reference feature quantity together with the corresponding feature quantity. A voice quality change determination method including a reference registration step of recording in a reference data recording unit.

1 is a functional block diagram showing the configuration of a voice quality change determination device according to a first embodiment. A flowchart showing an operation example of the voice quality determination unit and the reference registration unit Figure showing an example of a manuscript read by a speaker The figure which shows an example of the reference | standard text recorded on the reference | standard data recording part 7 Figure showing an example of the reference text after new registration Figure showing an example of the reference text after new registration The figure which shows the modification of a structure of the voice quality change determination apparatus in 1st Embodiment. Functional block diagram showing a configuration of a text input unit in the second embodiment Figure showing an example of a manuscript read by a speaker The figure which shows an example of the text after the input text of the content of the original shown to FIG. 7A was converted into the phonetic text Figure showing an example of a sentence read out by a speaker The figure which shows an example of the reference | standard text recorded on a reference | standard data recording part, and a reference | standard feature-value The figure which shows an example of the data content in the reference | standard data recording part after the process by the reference | standard registration part in 3rd Embodiment The figure which shows an example of the data content in the reference | standard data recording part after the process by the reference | standard registration part in 3rd Embodiment The figure which shows an example of the data content in the reference | standard data recording part after the process by the reference | standard registration part in 3rd Embodiment Functional block diagram showing the configuration of an audio recording apparatus according to the fourth embodiment

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Voice quality change determination apparatus 2 Voice input part 3 Text input part 4 Feature-value calculation part 5 Voice quality determination part 6 Reference registration part 7 Reference data recording part 10 Voice recording device 11 Language processing part 12 Conversion part 13 Language dictionary 14 Recording part 15 Recording Data recording part

Claims (7)

A voice quality change determination device accessible to a reference data recording unit for recording a reference feature amount representing a reference voice quality and a reference text corresponding to the reference feature amount,
A voice input unit for inputting a voice of a speaker as an input voice;
A text input unit for inputting text corresponding to the input voice as input text;
A feature amount calculation unit for calculating a feature amount of the input speech;
It is determined whether or not a change in voice quality of the input speech is within a predetermined allowable range by comparing a feature amount of the input speech corresponding to a portion where the input text and the reference text match each other with a reference feature amount. A voice quality determination unit,
A reference for recording, in the reference data recording unit, a new reference text and a reference feature amount together with the corresponding feature amount, the input text corresponding to the input speech for which the voice quality determination unit determines that the voice quality change is within an allowable range. A voice quality change determination device comprising a registration unit. The voice quality change determination device is further accessible to a text recording unit in which a text to be voice input is registered in advance,
The text input unit accesses the text recording unit, reads the input text, outputs the input text to the speaker,
The voice quality change determination device according to claim 1, wherein the voice input unit inputs a voice of a speaker who reads out the output input text. Further comprising a language processing unit that linguistically analyzes the input text and divides the input text into analysis units based on a result of language analysis;
The feature amount calculation unit calculates a feature amount corresponding to each analysis unit divided by the language processing unit,
The voice quality determination unit compares a feature quantity of the input speech corresponding to an analysis unit in which the input text and the reference text match each other with a reference feature quantity, so that a change in voice quality of the input speech is within a predetermined allowable range. The voice quality change determination device according to claim 1, wherein the voice quality change determination device determines whether or not The reference data recording unit records the reference text as a phonetic text,
A conversion unit for converting the input text into phonetic text;
The voice quality determination unit compares a feature quantity of the input speech corresponding to a portion where the input text converted to the phonetic text and the reference text match each other and a reference feature quantity, thereby changing the voice quality change of the input speech. The voice quality change determination apparatus according to claim 1, wherein the determination is performed.   The reference registration unit adds an existing reference feature amount corresponding to the portion of the input speech when adding the feature amount of the input speech corresponding to the portion where the text and the reference text match to the reference data recording unit. Any one of claims 1 to 4, wherein the feature quantity is replaced with a feature quantity, or a feature quantity calculated using the existing reference feature quantity and the feature quantity of the input speech is added as a new reference feature quantity. The voice quality change determination device according to the item. A voice quality change determination program for causing a computer accessible to a reference data recording unit that records a reference feature quantity indicating a reference voice quality and a reference text corresponding to the reference feature quantity to execute a voice quality change determination process,
Voice input processing for inputting the voice of the speaker as input voice,
Text input processing for inputting the text corresponding to the input voice as input text;
A feature amount calculation process for calculating the feature amount of the input speech;
It is determined whether or not a change in voice quality of the input speech is within a predetermined allowable range by comparing a feature amount of the input speech corresponding to a portion where the input text and the reference text match each other and a reference feature amount. Voice quality determination processing,
A reference for recording, in the reference data recording unit, a new reference text and a reference feature amount, together with the corresponding feature amount, the text corresponding to the input speech whose voice quality change is determined to be within an allowable range in the voice quality determination process. A voice quality change determination program that causes a computer to execute registration processing. A voice quality change determination method in which a computer executes voice quality change determination,
The computer can access a reference data recording unit that records a reference feature amount indicating a voice quality as a reference and a reference text corresponding to the reference feature amount,
A voice input step in which the computer receives a voice of a speaker as an input voice;
A text input step in which the computer acquires text corresponding to the input speech as input text;
A feature amount calculating step in which the computer calculates a feature amount of the input speech;
A voice quality determination step of determining a voice quality change of the input voice by comparing the feature quantity of the input voice and a reference feature quantity corresponding to a portion where the input text and the reference text coincide with each other;
The computer uses the whole or a part of the text corresponding to the input speech determined that the voice quality change is within an allowable range in the voice quality determination step as the new reference text and the reference feature quantity together with the corresponding feature quantity. A voice quality change determination method including a reference registration step of recording in a reference data recording unit.

Images