JPH05289691A - Speech speed measuring instrument - Google Patents

Abstract

PURPOSE:To measure the speech speed of a voice signal whose vocalization contents are unkown in real time. CONSTITUTION:A dynamic feature quantity analyzing circuit 12 extracts LPC cepstrum coefficients Ci-Ck from an input voice at a frame period and finds a dynamic feature quantity Di(t) from an equation Di(t)=SIGMAi1<k>((SIGMAff0<f0> fXCi)/(SIGMAf-f0<f0>f<2>)<2>, where f0 is the number of frames. A circuit 13 counts maximum values of this Di(t) at intervals of one minute and regards the result as a speech speed counted value, which is corrected by a circuit 15 by referring to the relation between the previously found actual number of phonemes and the number of the maximum points in a memory 14 and outputted as the speech speed.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention positively utilizes the dynamic feature amount of a voice signal corresponding to the movement of a human mouth, and the voice rate of a voice signal whose utterance content is unknown, for example, The present invention relates to a speech speed measuring device capable of automatically estimating the number of mora per unit time without human intervention.

[0002]

2. Description of the Related Art As a method for automatically estimating a speech rate, a method using a Fourier expansion coefficient of an envelope of a voice signal power (Furui, Voice Study Group, S81-62, 1981).
December). In this method, the syllable is applied to the word speech which is uttered in isolation, and the effect when applied to the naturally uttered sentence speech and conversation speech is unknown.

Further, conventionally, in the case of accurately measuring the speech rate, a human being listens to the uttered voice and converts it into a character to obtain it. Alternatively, by using a voice signal waveform, a sound spectrogram, or the like, a human being counts the number corresponding to the phoneme or the like by inspection and converts it into a vocalization rate. With such a method, real-time measurement was impossible.

On the other hand, it is a well-known fact that the dynamic feature amount of a voice signal is used for phonological segmentation and the like, but the feature amount corresponding to the movement of the human mouth is positively used to change the "utterance rate". There is no example applied to automatic estimation so far.
An object of the present invention is to measure the speech rate of a speech signal whose speech content is completely unknown as described above, in real time by calculation from only the physical feature amount of the speech signal without human intervention. To provide a device.

[0005]

According to the present invention, the time change amount of the voice spectrum characteristic is obtained as the dynamic feature amount of the voice signal, and the maximum point of the change pattern of the time change amount is the predetermined time. The speech rate of the voice signal is estimated based on the counted result. When a human utters a voice, the articulation mechanism required for utterance is moved fast when speaking quickly, and vice versa when speaking slowly. Therefore, when measuring the speech rate from a voice signal, a physical quantity that corresponds well to the movement of the mechanism used for vocalization,
For example, the amount corresponding to the movement of the mouth may be measured. Therefore, in the present invention, attention is paid to, for example, the movement of the spectrum envelope characteristic of the voice that corresponds well to the movement of the utterance mechanism, and the utterance speed of the voice signal is measured by using this dynamic feature amount.

[0006]

FIG. 1A shows an embodiment of the present invention. First, the dynamic feature amount of the voice is calculated by the dynamic feature amount analysis circuit 12 from the voice signal from the input terminal 11. Next, the speech rate coefficient counting circuit 13 obtains the number of maximum points per unit time in the calculated time-varying characteristic pattern of the dynamic feature quantity as a "speech rate coefficient". With that speech rate coefficient,
The speech rate is estimated by referring to the relationship between the speech rate coefficient and the speech rate stored in advance in the relational memory 14 in the speech rate calculation circuit 15.

The dynamic feature amount calculated by the dynamic feature amount analysis circuit 12 is such that it represents the time variation characteristic in the time or frequency domain of the voice signal feature amount corresponding to the movement of the mouth. Good. In one example, a dynamic feature amount that corresponds to a temporal change in the spectral envelope characteristic of the audio signal is used. That is, the LPC cepstrum coefficients (C ₁ to C) from the audio signal at a constant frame period (for example, 10 ms).
_k ) is extracted, and the dynamic feature amount D _i (t) is obtained from these time series by the following equation.

D _i (t) = Σ _{i = 1} ^k [{Σ _{f = −f0} ^f0 f × C _i } / (Σ _{f = −f0} ^f0 f ² )] ² (1) where f is the number of frames Here, for example, 3, k is the order of the LPC cepstrum coefficient. FIG. 1B shows an example of the analysis result of D _i (t). This example is a short sentence voice that one man speaks at three speeds of "slow", "normal" and "fast". From the figure, it can be seen that the temporal characteristics of D _i (t) change with the change of the speech rate. That is, it can be seen that the number of peaks in the observation time window increases as the speaking speed becomes faster.

In the above, as an example of the speech rate coefficient, D _i
Although the number of maximum points of the change of (t) per unit time is shown, the rate of change of D _i (t) per unit time, the change area per unit time, and the like can also be used. The unit time may be, for example, 2 to 6 seconds, or 1 to 6 minutes, and 1 to 6 minutes, 2 to 6 seconds may be averaged. As described above, according to the device of the present invention, the utterance speed of a voice signal whose utterance content is unknown can be measured with a simple device configuration without human intervention, and therefore the following application examples are possible.

(1) This device is used for a communication path (telephone line, etc.)
The amount of transmission information on the communication path can be roughly estimated just by connecting to the, and it is very useful in planning the use of the transmission system. (2) Since the speech rate can be automatically estimated, the user can obtain a comfortable response from the machine side when constructing various man / machine interfaces (for example, a voice response system using voice recognition or synthesis). Since it can be adaptively controlled according to the utterance tempo, etc., it can be used very effectively in such a system. An example thereof is shown in FIG. Input terminal 11
The input voice from is supplied to the voice recognition circuit 21 and is also supplied to the speech rate measuring device 22 according to the present invention. A response sentence is created by the response sentence creating circuit 23 according to the content recognized by the voice recognition circuit 21, and the response sentence is voice-synthesized by the voice synthesis circuit 24 and output to the output terminal 25. The synthesized speech utterance speed control circuit 26 controls the utterance speed of the synthesized sound in accordance with the utterance speed estimated at 22, and the estimated utterance speed and the speech recognition circuit 2
The response time is controlled by the response time control circuit 27 in response to the end of recognition of 1.

(3) The speech rate can be estimated prior to the actual phoneme recognition processing by applying it to the pre-processing of the speech recognition apparatus, which is considered to be useful for improving the recognition performance.
For example, as shown in FIG. 3, the voice signal from the input terminal 11 is supplied to the speech rate measuring device 22 of the present invention and the phoneme recognition circuit 28. Speech rate measuring device 22
The phoneme segment position determination circuit 29 determines the phoneme segment position based on the speech rate measured in (1) and the dynamic feature amount from the dynamic feature amount analysis circuit 12. Using the determined phoneme segment position, the phoneme recognition circuit 28 refers to the phoneme feature dictionary memory 31 to recognize the phoneme, and from the recognition result, refers to the word, phrase, and sentence feature dictionary memory 32 to refer to the word, The circuit 33 recognizes the clauses and sentences.

(4) The transmission line delay is relatively large in satellite telephones and other international telephones, but the influence of this delay on conversation varies depending on the state of conversation. In other words, when trying to talk in a hurry, and therefore the speech speed is high, the transmission line delay has a large adverse effect, but in a slow conversation, the effect is relatively small. Also, the influence varies depending on the opponent's response speed, how to insert a hammer, etc. Therefore, as shown in FIG.
The voice from the input terminal 11 is supplied to the speech rate measuring device 22 and the conversation time pattern analysis circuit 34, and the circuit 35 extracts the time feature amount of the conversation pattern from the analyzed conversation time pattern,
From this, the speech rate, and the transmission line delay amount from the terminal 36, the circuit 37 estimates the subjective delay quality of the transmission line delay amount for the conversation voice by referring to, for example, a correspondence table created in advance.

[0013]

FIG. 5 shows an example of experimental results for showing the effect of the present invention. This example is a result obtained by estimating the number of phonemes per minute from the maximum point of D _i (t) described above using short sentence voices uttered by five men and women and comparing the results with the correct answer. It can be seen that the correlation coefficient between them is 0.89, which is a high correlation. By obtaining the correction coefficient in advance and storing it together with the relational expression in the relational expression memory 14, the relationship between the two can be plotted on a diagonal line, and the speech rate can be easily estimated.

As described above, according to the present invention, it is possible to relatively accurately estimate the utterance speed with a relatively simple structure even for a voice signal whose utterance content is completely unknown in real time.

[Brief description of drawings]

FIG. 1A is a block diagram showing an embodiment of the device of the present invention,
FIG. 3B is a diagram showing an example of a dynamic feature amount of voice used in the device of the present invention.

FIG. 2 is a block diagram showing a voice response system to which the device of the present invention is applied.

FIG. 3 is a block diagram showing a voice recognition device to which the device of the present invention is applied.

FIG. 4 is a block diagram showing a delay quality objective measuring device to which the device of the present invention is applied.

FIG. 5 is a diagram showing an example of a speech rate measurement result according to the embodiment of the device of the present invention.

Claims (1)

[Claims] 1. A circuit for obtaining a time change amount of a voice spectrum characteristic as a dynamic feature amount of an audio signal, a circuit for counting a maximum point of a change pattern of the obtained time change amount for a predetermined time, and a counting result thereof. A speech rate measuring device comprising: a circuit for estimating a speech rate of the voice signal based on the above.