JP4587854B2 - Emotion analysis device, emotion analysis program, program storage medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an emotion analyzer which can quantitatively evaluate the perplexed condition of an utterer. <P>SOLUTION: The emotion analyzer 1 is provided with: a means which extracts a sound pressure level value from digital voice data obtained by sampling the uttered voice of a person; a means which divides the time series of the digital voice data into measuring intervals; a means which computes an average value of the sound pressure for every measuring interval; a means that recognizes the condition, in which the sound pressure level value continues for equal to or longer than a prescribed time and becomes equal to or less than a reference value, as a pause and obtains the percentage of the time being occupied by the pause for every measuring interval; a means which recognizes an uttered condition and a silenced condition in accordance with the sound pressure level value and obtains the number of appearances per unit time of the uttered condition divided by the silenced condition as the uttering speed for every measuring interval; and an evaluation result outputting means which appropriately outputs the evaluation point that indicates the degree of perplexity corresponding to the total point of the points corresponding to the time percentage between the point corresponding to the average sound pressure level value for every measuring interval and the point corresponding to the uttering speed. <P>COPYRIGHT: (C)2007,JPO&amp;INPIT

Description

  The present invention relates to an emotion analysis apparatus that analyzes an emotion of a speaker based on an audio signal when a person speaks.

A technique for analyzing emotions based on spoken speech is described in Patent Document 1 below, for example. Emotion analysis is, for example, the purpose of analyzing call voices between customers and operators in a call center, quickly detecting customer emotions and operator swaying, etc. It is possible to adapt to.
Special table 2003-508805 gazette

  Conventional emotion analysis is mainly aimed at extracting emotions of emotions. However, it is desirable for call center operators to respond calmly to customers, and they are often trained not to express emotions. For this reason, even if the operator feels some kind of “anxiety”, the emotion cannot be hit with the customer as emotions. If the state of anxiety persists, the operator will eventually have to respond vaguely to the customer, and eventually the conversation will fail. This could furious customers and lose trust in the company.

  The present inventors tried to extract “anxiety”, that is, “confused”, which is an extremely delicate emotion compared to emotions, with high accuracy. The voices of telephone conversations between the operator and the customer at the call center prepared for this attempt are repeatedly given to multiple people who are excellent in customer interaction technology, such as the person in charge at the call center department in the company. The audio recording was auditioned and the mental state of the operator was pointed out. On the other hand, the sound feature information (sound pressure level, frequency, etc.) extracted from the sampling data of the recorded sound was analyzed in detail. As a result, it was found that there is a specific change in the predetermined feature information extracted from the sampling data in a state where the operator seems to be confused. The present invention has been made based on this finding, and an object of the present invention is to provide an emotion analysis apparatus that can accurately extract the emotion “confused” from the speech and can quantitatively evaluate the degree of the confusion.

The present invention for achieving the above object is an emotion analysis apparatus for quantitatively analyzing a confused state of a speaker,
A sound pressure extraction means for extracting a sound pressure level value from digital sound data obtained by sampling a person's speech collected from a microphone;
A measurement period setting means for dividing the time series of the digital audio data into measurement periods at predetermined time intervals;
Average sound pressure acquisition means for calculating an average value of sound pressure level values extracted by the sound pressure extraction means as an average sound pressure level value for each measurement period;
Recognize a state where the sound pressure level value extracted by the sound pressure extraction means continues for a predetermined time or more and stays below the reference value as an interval, and obtain an inter-rate ratio for determining the time ratio for each measurement period Means,
The speech speed that recognizes the speech state and the silence state according to the sound pressure level value extracted by the sound pressure extraction means, and obtains the number of occurrences per unit time of the speech state divided by the silence state for each measurement period as the speech speed Acquisition means;
For each measurement period, obtain a score according to the average sound pressure level value and a score according to the speech rate, and a score according to the time ratio, and based on an evaluation score according to the total score value of each score An evaluation result output means for appropriately outputting the evaluation result;
It has.

Further, sound pressure extraction means for extracting a sound pressure level value from digital sound data obtained by sampling a person's speech collected from a microphone,
A measurement period setting means for dividing the time series of the digital audio data into measurement periods at predetermined time intervals;
Average sound pressure acquisition means for obtaining an average value of sound pressure level values B extracted by the sound pressure extraction means for each measurement period;
A state in which the state in which the sound pressure level value B from the sound pressure extracting means is equal to or less than the predetermined reference sound pressure level value BS1 continues for a predetermined time t or longer is recognized as an interval, and the total time during the measurement period D, a period ratio acquisition means for determining a ratio E during the measurement period by a formula of E = D / (C + D), where C is the total time in a state other than the interval,
A state in which the sound pressure level value B from the sound pressure extraction means is equal to or higher than a predetermined reference sound pressure level value BS2 is defined as an utterance state, and the number of occurrences of the utterance state per unit time detected within the measurement period is defined as an utterance speed G. The required speech speed acquisition means;
In the time series of voice data, the first measurement period or the part in the measurement period of the predetermined number of times from the first time is used as the reference voice, and the average sound pressure level BS, the ratio ES between them, and the speech rate GS in the reference voice are acquired. A reference value acquisition means for
For each measurement period, depending on the score according to the ratio between the average value of the sound pressure level B and the BS, the score according to the ratio between the ratio E and ES, and the ratio between the speech rates G and GS An evaluation result output means for providing a stepwise evaluation score according to the range of the total score value of each score as a degree of confusion and appropriately outputting an evaluation result based on the evaluation score. ,
It is more preferable if it is an emotion analysis apparatus provided with.

The present invention extends to a computer program installed in a computer, and the program is stored in the computer.
An audio data acquisition step of acquiring digital audio data obtained by A / D converting an analog audio signal;
A measurement period setting step for dividing the digital audio data time series into measurement periods at predetermined time intervals;
For each measurement period, an average sound pressure acquisition step for calculating an average value of sound pressure level values extracted by the sound pressure extraction means as an average sound pressure level value;
Recognize a state where the sound pressure level value extracted by the sound pressure extraction means continues for a predetermined time or more and stays below the reference value as an interval, and obtain an inter-rate ratio for determining the time ratio for each measurement period Steps,
The speech speed that recognizes the speech state and the silence state according to the sound pressure level value extracted by the sound pressure extraction means, and obtains the number of occurrences per unit time of the speech state divided by the silence state for each measurement period as the speech speed An acquisition step;
For each measurement period, obtain a score according to the average sound pressure level value and a score according to the speech rate, and a score according to the time ratio, and an evaluation according to the total score value of each score as the degree of confusion An evaluation result output step of assigning points and outputting the evaluation results based on the evaluation points as appropriate;
It is an emotion analysis program that executes A program storage medium in which the emotion analysis program is recorded is also included in the scope of the present invention.

  According to the emotion analysis apparatus of the present invention, it is possible to accurately extract a “confused” state, which is an extremely delicate emotion compared to emotions, and to quantitatively evaluate the degree of the confusion.

=== About quantitative evaluation of emotions ===
The feelings of the person who is speaking and the appropriateness of the utterance can be judged by listening to a person with academic experience or an expert who listens to the utterance state of the person who has been acoustically output. However, unless the judgment results are actually output as numerical values, it is impossible to automatically detect and appropriately deal with the feelings of operators and customers in call center operations, and to objectively evaluate operator skills. .

  As is well known, sound feature information is extracted from an input sound signal, and for example, time transition of fundamental frequency (pitch) and sound pressure level, frequency analysis result per unit time obtained by Fourier transform of sound signal (feature) There is a speech analyzer for visualizing the utterance state by graphing parameters) (for example, CSL Computerized Speech Lab Model 4500 manufactured by Kay Elemetrics Corp.).

  Therefore, the present inventors display a graph relating to the sound characteristics of the speech analysis device while the sound output when the call center operator talks to the customer over the telephone is acoustically output, and scholars and experts can provide this information. The correlation between the emotion and the temporal displacement state of the predetermined feature information was obtained, and the appropriateness of the emotion and utterance was quantified. The emotion analysis apparatus of the present invention digitizes and outputs various emotions and appropriateness of utterances included in the uttered speech based on the correlation found by the above method.

=== Configuration of Emotion Analysis Device ===
The emotion analysis apparatus according to the present invention is realized by, for example, using a personal computer on which an audio card is mounted as hardware, and executing an emotion analysis application program (hereinafter, emotion analysis software) installed in the computer. .

  FIG. 1 shows a functional block configuration of an emotion analysis apparatus according to this embodiment of the present invention. The audio signal processing unit 10 corresponds to a sound card mounted on a computer in the above hardware configuration. The audio signal processing unit 10 performs A / D conversion on the voice signal of the speaker input from the microphone 11 and outputs digital voice data. Specifically, it is converted into digital audio data sampled at a sampling frequency of 8 kHz, a quantization number of 16 bits, and 1 channel (monaural).

  The voice analysis unit 20 corresponds to digital voice data processing by emotion analysis software, and the sound characteristic information such as the fundamental frequency (pitch) and the sound pressure level from the digital voice data sequentially transferred from the audio signal processing unit 10. A function 21 for extracting, and a function 22 for acquiring, as speech analysis parameters, feature information and speech information obtained based on the feature information, secondary duration such as interval, interval ratio, feature parameters, etc. Contains. In the present embodiment, weighting and limiting conditions for each voice analysis parameter are set as initial setting parameters, and when the initial setting parameters are input via the operation input unit 41 such as a keyboard or a mouse in the user interface 40, The parameters are stored in the external storage 50 of the computer as a file (initial setting file) 52 of a predetermined format.

  The voice analysis unit 20 extracts / acquires a voice analysis parameter based on a specific voice analysis parameter and a corresponding initial setting parameter, and based on the voice analysis parameter, various speech states (emotion: anger, confusion, etc. A function 23 for extracting appropriateness of speech: speech speed, time ratio, clarity, pronunciation characteristics of endings, and the like, and a function 24 for obtaining the degree of the extracted speech state and the evaluation result. The evaluation result output from the voice analysis unit 20 is converted into data (text, numerical values, etc.) in an appropriate format by the analysis result output unit 30, for example, an output device (display device, printing device, etc.) of the user interface 40. 42 and the external storage 50 are output as an evaluation result file 53. Of course, the evaluation result can also be output to an appropriate device or in an appropriate format, such as outputting an audio signal or transferring data corresponding to the evaluation result to another computer.

  In addition to the digital audio data sampled in real time by the audio signal processing unit 10, the audio analysis unit 20 also processes a recorded audio data file 51 in a predetermined format such as a WAV format recorded in the external storage 50. Can be targeted. Also provided is a function 25 for creating and recording speech data collected from the microphone 11 in the audio data file 50.

=== Speech analysis parameter ===
The voice analysis unit 20 extracts feature information from the time displacement of the digital voice data collected every sampling period. In this embodiment, the time series of digital audio data is divided into periods (for example, several seconds) sufficiently longer than the sampling period as measurement periods, and the pitch A and sound pressure level B are calculated from the time series of digital audio data in each measurement period. , Speech duration C, duration D between, overall ratio E between, feature parameter F, speech rate G are acquired as speech analysis parameters. FIG. 2 shows the concept of the measurement period. The determination is made uniform by overlapping the end point of one measurement period and the start point of the next measurement period in time series. Further, in this embodiment, at the beginning of the utterance, it is assumed that the speaker's emotion is stable, and in the above time series in the digital audio data, the first or initial predetermined measurement period, or the initial predetermined number of times. The amount equivalent to the measurement period is used as the reference voice. And the value of each said analysis parameter acquired from a reference | standard voice is employ | adopted as a reference value. Each speech analysis parameter will be described below.

<Pitch (A)>
The fundamental frequency (Hz) of voice and the first overtone of voice. Let AS be the average pitch of the reference speech. An average value (average pitch) of pitches in a measurement period or a predetermined period is A.

<Sound pressure level (B)>
The average sound pressure level (reference sound pressure level) of the reference sound is defined as BS. The unit is dB (decibel), the reference sound pressure is 20 μPa, and the sound pressure level value is indicated by a multiple of the reference sound pressure level BS. Let B be the average value (average sound pressure level) of the sound pressure levels during the measurement period or a predetermined period.

<Duration (C)>
The duration of a series of utterances and the duration of the voice group are calculated from the sound pressure threshold. It is obtained from the temporal displacement of the sound pressure level extracted from the digital audio data. After each word, syllable, and sentence conversation, if the sound pressure level is below a threshold for a predetermined time (eg, 0.3 seconds), it is determined that there is no sound. The threshold value of the sound pressure level can be set by user input.

<Duration between (D)>
The time when you are not speaking in a conversation. That is, a state in which the sound pressure level is not more than the threshold value of the sound pressure level input by the user and continues for the predetermined time or more is defined as “between”, and the duration between them is obtained.

<Overall ratio (E)>
The ratio reference value between them is defined as ES. From the duration (D) between the conversation duration (C) and the duration (D), the overall ratio is calculated by E = D / (C + D).

<Characteristic parameter (F)>
This is an analysis of the frequency per unit time, which is calculated by Fourier transforming the digital audio data of the speaker. The feature parameter reference value is FS. As is well known, the frequency contained in a human voice extends from 60 Hz to 10,000 thousand Hz. In addition, by analyzing the frequency of human voice, it is possible to read the gender, age, height, occupational awareness, physical condition, etc. of the speaker. When determining the appropriateness of a call center operator, it is possible to determine whether or not the voice is “passing through” by using frequency analysis. The “passing voice” is based on the premise that an abdominal utterance is made, and when the abdomen utterance is made, the frequency of the voice is gathered around 2500 Hz to 3000 Hz. In this embodiment, since the sampling frequency is 8 kHz, the upper limit of the frequency that can be analyzed is 4000 Hz.

<Speaking speed (G)>
Number of word elements per unit time. Assume that the utterance speed reference value is GS. The speaking speed is defined as a “speech element” from the time point when the sound pressure level becomes equal to or lower than the sound pressure level threshold to the next boundary. The number of utterance elements per unit time is used as the utterance speed.

=== Quantification of confusion ===
The emotion analysis apparatus of the present invention has a great feature in that it can quantitatively analyze an extremely delicate emotion called “confused”. FIG. 3 is a graph showing fluctuations in the sound pressure level during speech. This graph 60 shows a variation 61 in the sound pressure level in the normal (initial) state and a variation 62 in the sound pressure level when the expert determines that the state is in a confused state. The “confused state” is lower than the “normal state” by about 1/9 of the sound pressure level. In addition, it can be seen from this graph that the speaking rate is decreasing and the ratio between the speaking rate is increasing. The emotion analysis apparatus 1 quantitatively analyzes these changes and outputs the degree of confusion numerically.

  In the present embodiment, a predetermined score is assigned to the values of each voice analysis parameter of the sound pressure level B, the speech rate E, and the overall ratio G between the measurement periods according to the range of the values. And the evaluation result according to the total score given about each audio | voice analysis parameter is output. FIG. 4 shows the correspondence between the value of each voice analysis parameter and the score, and the correspondence between the total score and the degree of confusion in this example. For each of the above-described speech analysis parameters employed for extracting the confusion state, the speech analysis unit 20 points in a stepwise manner in accordance with how much the value in the measurement period is relative to the initial value. And the total value of the scores for each speech analysis parameter is calculated. Then, the degree of confusion is evaluated in five stages according to the range of the total points. The emotion analysis device 1 accepts the correspondence between the value of each analysis parameter and the point and the correspondence between the total score and the evaluation point by the user input as the above-mentioned initial setting parameters.

=== Evaluation of other emotions and utterance states ===
In addition to the above-mentioned confusion, the emotion analysis apparatus of the present embodiment has anger, appropriateness of speech rate, appropriateness between utterances, clarity of speech, pronunciation characteristics of endings, appropriateness of vocabulary, and appropriateness of inflection. The degree is output numerically.

<Anger>
In an utterance state including an angry emotion, an increase in pitch A, an increase in sound pressure level B, and an increase in utterance speed G are observed. FIG. 5 exemplifies the correspondence between the scores according to the numerical ranges of these speech analysis parameters and the correspondence between the graded evaluation points according to the anger score which is the sum of the scores.

<Appropriate speech rate>
In this embodiment, it is evaluated whether or not the speech speed G is within a certain range. FIG. 6 shows the correspondence between the utterance speed value and the evaluation level of the appropriateness of the utterance speed in this example.

<Adequacy during utterance>
In the present embodiment, the appropriateness is evaluated based on the overall ratio E and the speech rate G. FIG. 7 shows the relationship between the range of the total rate E between utterances, the score and the correspondence, the correspondence between the range of the utterance speed G and the score, and the appropriateness between the total points of these speech analysis parameters. An example of correspondence with evaluation points is shown.

<Clarity of utterance>
It was found that when the utterance was “clear”, the basic pitch of the voice increased and the sound pressure increased compared to “unclear”. In the present embodiment, the clarity of the utterance is evaluated based on the vertical width of the pitch, the vertical width of the sound pressure level, and the distribution state of the characteristic parameters. FIG. 8 shows the correspondence between the range of the vertical width A1 of the pitch and the score, the correspondence between the range of the vertical width B1 of the sound pressure level and the score, the range of the characteristic parameter F value, the score, and the correspondence. The correspondence between the total score of these speech analysis parameters and the evaluation score of speech clarity was shown. Here, the pitch width A1 = (maximum pitch−minimum pitch) and the sound pressure level height B1 = (maximum sound pressure level−minimum sound pressure level), and the pitch height and sound of the reference sound. The vertical width of the pressure level is A1S and B1S, respectively.

<Characteristics of ending pronunciation>
In this embodiment, the time from the immediately preceding 2 seconds to the state in between is defined as “End of word”, the digital voice data at the end of the word is analyzed, and the pitch A, sound pressure level B, and feature parameter F at the end of the word are analyzed. Based on the values, the speaker's pronunciation characteristics are classified into five characteristics: “no problem”, “childhood”, “office work”, “dark”, and “others”. FIG. 9 exemplifies the correspondence between the voice analysis parameter value and the score when corresponding to the features of each utterance.

  If the pronunciation is “no problem (normal)”, it is within the appropriate range, and if it is “childish”, a decrease in sound pressure level and vertical movement of the pitch are recognized. In “office work”, the sound pressure level is constant and the pitch does not move up and down. “Dark” has a narrow frequency distribution in the feature parameter. In other words, it was not possible to make a double utterance, and it was not a “passing voice”. “Other” is a case that does not correspond to any of the above features. That is, the feature of the utterance having the highest score according to the value of each voice analysis parameter is the evaluation result. As a specific evaluation procedure, a score corresponding to the value of each voice analysis parameter is set as a set for each pronunciation feature. Then, the score corresponding to each voice analysis parameter is summed for each feature of each pronunciation, and the feature of the pronunciation having the highest total score is used as the evaluation result. For example, it is assumed that the average sound pressure level A at the ending, the average pitch B at the ending, and the average feature parameter value F at the ending are 80%, 50%, and 20% as the ratios to the respective reference values AS, BS, and FS, respectively. . In “Kindergarten”, the numerical values of sound pressure and pitch fall within the numerical range shown in the correspondence relationship between the scores, and 90 points, which is a total of 40 points for sound pressure and 50 points for pitch, are scored for “Kindergarten”. It becomes.

  When the same calculation is performed for “office work”, “dark”, and “no problem”, the total points are 0, 120, and 50, and the evaluation result is “dark”. In addition, when the total score is 0 to 70 points for all the pronunciation features, it is “others”. In this example, the evaluation points of 0, 1, 2, 3, and 4 are output for the pronunciation characteristics of “no problem”, “childhood”, “office work”, “dark”, and “others”, respectively.

<Rationality>
The appropriateness of vocabulary is evaluated based on the same parameters as the voice analysis parameters employed when evaluating the characteristics of the utterance. In the above utterance feature, the ending part is analyzed, while the morale analyzes all the conversation parts except “between”. Then, the vertical movement change of the pitch, the appropriateness of the sound pressure level, and the presence / absence of abdominal utterance by frequency analysis are comprehensively determined and output as evaluation points. FIG. 10 shows the correspondence between the value and the score and the correspondence between the total score and the evaluation score for each of the speech analysis parameters of pitch A), sound pressure level B, and feature parameter F.

<Adequacy of intonation>
The appropriateness of inflection is evaluated by the vertical width of the pitch. FIG. 11 shows the range of the value of the vertical width A1 of the pitch, the score, and the corresponding relationship, and the corresponding relationship between the score and the evaluation score for the appropriateness of the inflection in this example. The vertical width A1 of the pitch is (maximum pitch−minimum pitch).

=== User interface ===
The emotion analysis device includes a setting parameter input function that accepts evaluation criteria for emotions and utterance states by user input, and a result presentation function that outputs evaluation results of various emotions and utterance states based on the input setting parameters and digital voice data, Is provided by the GUI. FIG. 12 shows an outline of a screen for inputting basic initial setting parameters such as extracting feature information from digital audio data. The screen 70 is provided with a plurality of text box groups 71 for inputting various setting parameters necessary for extracting a pitch, a text box group 72 for inputting other setting parameters, and the like. Yes. In this example, in order to extract the pitch, a sampling frequency, a window function designation, a frame length applied to the window function, a frame period, and the like are set. In addition, as other setting parameters, a threshold of a sound pressure level that should be determined as silence when measuring the speech speed, a threshold of a sound pressure level that should be determined as silence when detecting a gap, and a threshold for determining as a gap Designated inputs such as the threshold value of the continuation period, the length of the measurement period (period), the overlap period in the period, or the frame length of the Fourier transform (FFT) when calculating the feature parameter are accepted.

  FIG. 13 is a schematic screen for receiving various setting parameters for a specific emotion or speech state. Here, a screen 80 for inputting various parameters related to “confusing” is shown. On this screen 80, for each voice analysis parameter of the sound pressure level, the speech rate, and the ratio between them, which is used to extract the confusion, the threshold and weight, that is, the range and score of the voice analysis parameter value, A text box group (81-83) for accepting designation of the correspondence, and an input field 84 for designating the correspondence between the total score and the evaluation result.

=== Application example in a call center ===
As an application form of the emotion analysis apparatus of the present embodiment, an application example to call center business is actually given. The call voice from the microphone of the telephone used by the operator is input to the audio signal processing unit of the emotion analysis device. The emotion analysis apparatus starts processing the digital voice data from the point in time when it responds to an incoming call from the customer's phone. And the evaluation point of an operator's confusion state is output sequentially during the telephone call with a customer.

  If the output point of the evaluation point is a computer used by the person in charge of the call center, the person in charge can monitor the confusion state of each operator at any time using the evaluation point displayed and output on his computer. If a severe confusion is recognized for a certain operator, a configuration in which the telephone line is switched to the own telephone can be considered. Thereby, before the operator gets stuck in the conversation with the customer, the experienced person in charge can respond to the customer, and the communication can be continued smoothly without making the customer feel uncomfortable. Alternatively, when the confusion evaluation score exceeds a predetermined value (for example, 4 or more), that fact is notified to the computer of the person in charge, or the telephone line is automatically switched to another operator or the like. Good.

It is a functional block diagram of the emotion analysis apparatus in the Example of this invention. It is a conceptual diagram of the measurement period of the audio | voice data in the said emotion analyzer. It is a fluctuation graph of the sound pressure level in a normal state and a confused state of a speaker. It is the schematic of the scoring method when evaluating a puzzled state. It is the schematic of the scoring method when evaluating anger. It is the schematic of the scoring method when evaluating the appropriateness of speech speed. It is the schematic of the scoring method when evaluating the accuracy between utterances. It is the schematic of the scoring method when evaluating the clarity of an utterance. It is the schematic of the scoring method when evaluating the feature of the utterance of an ending. It is the schematic of the scoring method when evaluating the appropriateness of language. It is the schematic of the scoring method when evaluating the appropriateness of intonation. It is a screen schematic diagram for inputting general setting parameters in the GUI of the emotion analysis apparatus. It is the screen schematic diagram for inputting the setting parameter regarding confusion in the GUI of the emotion analysis apparatus.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Emotion analyzer 10 Audio signal processing part 11 Microphone 20 Voice analysis part 30 Analysis result output part 40 User interface

Claims (4)

An emotion analysis device that quantitatively analyzes the confused state of a speaker,
A sound pressure extraction means for extracting a sound pressure level value from digital sound data obtained by sampling a person's speech collected from a microphone;
A measurement period setting means for dividing the time series of the digital audio data into measurement periods at predetermined time intervals;
Average sound pressure acquisition means for calculating an average value of sound pressure level values extracted by the sound pressure extraction means as an average sound pressure level value for each measurement period;
Recognize a state where the sound pressure level value extracted by the sound pressure extraction means continues for a predetermined time or more and stays below the reference value as an interval, and obtain an inter-rate ratio for determining the time ratio for each measurement period Means,
The speech speed that recognizes the speech state and the silence state according to the sound pressure level value extracted by the sound pressure extraction means, and obtains the number of occurrences per unit time of the speech state divided by the silence state for each measurement period as the speech speed Acquisition means;
For each measurement period, obtain a score according to the average sound pressure level value and a score according to the speech rate, and a score according to the time ratio, and an evaluation according to the total score value of each score as the degree of confusion An evaluation result output means for assigning points and outputting the evaluation results based on the evaluation points as appropriate;
Emotion analysis device with An emotion analysis device that quantitatively analyzes the confused state of a speaker,
A sound pressure extraction means for extracting a sound pressure level value from digital sound data obtained by sampling a person's speech collected from a microphone;
A measurement period setting means for dividing the time series of the digital audio data into measurement periods at predetermined time intervals;
Average sound pressure acquisition means for obtaining an average value B of sound pressure level values extracted by the sound pressure extraction means for each measurement period;
A state in which the state in which the sound pressure level value from the sound pressure extracting means is equal to or less than the predetermined reference sound pressure level value BS1 continues for a predetermined time t or more is recognized as an interval, and the total time D during the measurement period , The total time of the state other than between is C, and the ratio E during the measurement period is calculated by the formula E = D / (C + D),
A state where the sound pressure level value from the sound pressure extracting means is equal to or higher than a predetermined reference sound pressure level value BS2 is defined as an utterance state, and the number of occurrences of the utterance state per unit time detected within the measurement period is obtained as the utterance speed G. Speaking speed acquisition means;
In the time series of voice data, the first measurement period or the part in the measurement period of the predetermined number of times from the first time is used as the reference voice, and the average sound pressure level BS, the ratio ES between them, and the speech rate GS in the reference voice are acquired. A reference value acquisition means for
For each measurement period, the number according to the ratio between the average sound pressure level value B and BS, the number according to the ratio between the ratio E and ES, and the ratio between the speech rate G and GS according to the value. An evaluation result output means that obtains a score, assigns a stepwise evaluation score according to the range of the total score value of each score as a degree of confusion, and appropriately outputs an evaluation result based on the evaluation score;
Emotion analysis device with A computer program installed on a computer,
An audio data acquisition step of acquiring digital audio data obtained by A / D converting an analog audio signal;
A measurement period setting step for dividing the digital audio data time series into measurement periods at predetermined time intervals;
For each measurement period, an average sound pressure acquisition step for calculating an average value of sound pressure level values extracted by the sound pressure extraction means as an average sound pressure level value;
Recognize a state where the sound pressure level value extracted by the sound pressure extraction means continues for a predetermined time or more and stays below the reference value as an interval, and obtain an inter-rate ratio for determining the time ratio for each measurement period Steps,
The speech speed that recognizes the speech state and the silence state according to the sound pressure level value extracted by the sound pressure extraction means, and obtains the number of occurrences per unit time of the speech state divided by the silence state for each measurement period as the speech speed An acquisition step;
For each measurement period, obtain a score according to the average sound pressure level value and a score according to the speech rate, and a score according to the time ratio, and an evaluation according to the total score value of each score as the degree of confusion An evaluation result output step of assigning points and outputting the evaluation results based on the evaluation points as appropriate;
Emotion analysis program that runs. A program storage medium in which the emotion analysis program according to claim 3 is recorded.

Images