JP2005077970A - Device and method for speech quality objective evaluation - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To precisely estimate the subjective quality of the speech of a real telephone call, by taking into account the influence of the speaking state of a speaker of a side (receiving side) of speech quality evaluation on the quality evaluation. <P>SOLUTION: A distortion measurement part 12 compares a speech signal at a point A obtained from a speech DB 11 with a deteriorated speech obtained by passing the former speech signal through a system 2 to be evaluated, to quantize the extent of the distortion of the deteriorated speech as a time series. A two-way speech section detection part 13 detects a bilateral speech section by comparing the speech signal at the point A obtained from the speech DB 11 with a speech signal at a point B. A weighting part 14 weights and averages the extent of distortion time series that the distortion quantity measurement part 12 outputs, while making the weight on the extent of quality deterioration in the two-way speech section which is smaller than in an independent speech section. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a speech quality evaluation technique, and more particularly to a speech quality objective in which subjective quality is estimated from measurement of physical features of a speech signal without performing a subjective assessment test in which a human listens to speech and evaluates the quality. The present invention relates to an evaluation apparatus and a voice quality objective evaluation method.

  FIG. 7 shows a block configuration of a conventional voice quality objective evaluation apparatus. In FIG. 7, reference numeral 2 denotes an evaluation target system that is an evaluation target of how much the voice quality deteriorates due to transmission, and 4 denotes a voice quality objective evaluation apparatus. In the voice quality objective evaluation device 4, reference numeral 41 denotes a voice database (DB) in which an evaluation sound source is held, reference numeral 42 denotes a distortion amount measurement unit that measures the distortion amount generated in the evaluation target system 2, and reference numeral 43 denotes a distortion amount measurement unit 42. It is an average process part which carries out the time average of the measured distortion time series. The evaluation target system 2 is, for example, a fixed telephone system, a mobile phone system, an IP telephone system, or the like.

  The conventional speech quality objective evaluation apparatus 4 uses the input signal from the speech DB 41 holding the evaluation sound source to the evaluation target system 2 and the output signal from the evaluation target system 2 (hereinafter referred to as degraded speech) to measure the amount of distortion. The unit 42 calculates a distortion time series based on an audio quality objective evaluation algorithm (for example, an audio quality objective evaluation method defined in ITU-T recommendation P.862).

  ITU-T recommendation P.I. In the speech quality objective evaluation method defined in 862, specifically, the frequency spectrum analysis of the evaluation sound source and the degraded speech through the evaluation target system 2 is performed, the difference between these is obtained, and weighting based on human auditory characteristics is performed. By doing so, the perceived amount of distortion is quantified.

  At this time, only a one-way audio signal in a two-way call is used as a test signal. The average processing unit 43 obtains a final objective evaluation value by time averaging the distortion time series as the output of the distortion amount measuring unit 42.

In general, two-way calls (calls at point A and point B) are as shown in FIG.
(1): Both silent sections,
(2) -A: Point A single utterance section,
(2) -B: Point B single utterance section,
(3): Both utterance intervals,
Are classified into four states.

  In the conventional method, for example, when attention is paid to the point A, the average processing unit 43 shown in FIG. Distinguishing between speech and silent sections), and changing the weighting for the amount of distortion of the speech signal between them, the degradation quantified consistent with the subjective quality evaluation characteristics.

Note that the speech quality objective evaluation algorithm used in the distortion amount measurement unit 42 shown in FIG. 7 is described in Non-Patent Document 1, for example. This algorithm is the ITU-T recommendation P.I. 862 is adopted.
Perceptual evaluation of speech quality (PESQ) -a new method for speech quality assessment of telephone networks and codecsRix, AW; Beerends, JG; Hollier, MP; Hekstra, AP; Acoustics, speech, and Signal Processing, 2001.Proceedings. '01). 2001 IEEE International Conference on, Volume: 2,7-11 May 2001 Page (s): 749-752 vol.2.

  However, in fact, even in the point A utterance section, the sensitivity to the voice quality of the speaker at the point A felt by the speaker at the point B is different between the point A single utterance section and the two-side utterance section. That is, in the case of the two-speaking section, since the speaker at the point B is speaking, even if the speaker's voice at the point A is distorted, it is masked by the voice of the speaker at the point B and As a result, the deterioration of the subjective quality is reduced compared with the point A single utterance section even with the same amount of distortion. In the conventional method, since this effect is not taken into consideration, the estimated subjective quality is evaluated more severely than the subjective quality felt by the speaker at the point B in an actual call, and there is a problem in quality estimation accuracy.

  SUMMARY OF THE INVENTION An object of the present invention is to provide an objective speech quality objective evaluation device and speech quality objective that can accurately estimate the subjective quality of speech in an actual call in consideration of the influence of the speech state of the speaker evaluating the quality on the quality assessment. To provide an evaluation method.

  In order to solve the above-mentioned problem, the present invention relates to a point B story when evaluating a voice quality at a point A (that is, evaluating a voice quality felt by a speaker at the point B) in a call between the points A and B. The main feature is that the speech distortion amount of the speaker at the point A is weighted based on the utterance state of the speaker.

  In other words, the present invention provides a speech quality objective evaluation device that estimates the subjective quality of a speech signal (the quality that a person feels when listening to the signal) from the measurement results of the physical features of the speech signal. The feature is that the weighting for the amount of degradation is reduced compared to the single utterance interval. Further, in consideration of the fact that the bidirectional speech rate is correlated with the end-to-end transmission delay time, the bidirectional speech rate of the voice signal used for the evaluation is determined from the transmission delay time of the evaluation target system.

  In the conventional method, the distortion amount is quantified based on only the result of analyzing the voice signal of the speaker at the point A, and the subjective quality is estimated. This is a difference from the present invention. In the present invention, since the voice quality of the speaker at the point A felt by the speaker at the point B is evaluated in consideration of the utterance state of the speaker at the point B, the accuracy is higher than that in the conventional method not considering this. Good subjective quality estimation is possible.

  According to the voice quality objective evaluation device of the present invention, it is possible to objectively evaluate the voice quality in consideration of the influence of the speech state of the speaker on the voice quality evaluation side on the quality evaluation. It is possible to accurately estimate the subjective quality of. Further, by determining the both speech rates of the voice signal used for the evaluation in accordance with the transmission delay time and using the voice signal of the both speech rates as the evaluation sound source, it is possible to estimate the subjective quality with higher accuracy.

  A block configuration of the first embodiment of the present invention is shown in FIG. In FIG. 1, reference numeral 1 denotes an audio quality objective evaluation apparatus according to the present invention. As described above, the evaluation target system 2 is assumed to be, for example, an IP telephone system, a fixed telephone system, a mobile phone system, or the like.

  The voice quality objective evaluation device 1 compares the voice DB 11 holding the voice signals of both the points A and B, the voice signal of the point A obtained from the voice DB 11 and the deteriorated voice obtained by passing this through the evaluation target system 2. The distortion amount measuring unit 12 that quantifies the distortion amount of the deteriorated speech as a time series, and the two-way utterance detecting the two-speaking section by comparing the voice signal at the point A and the voice signal at the point B obtained from the voice DB 11 Based on information obtained from the section detection unit 13 and the both-speaking section detection unit 13, a weighting unit 14 that weights the distortion time series output from the distortion measurement unit 12 is provided.

  The voice DB 11 stores voice signals of speakers at each point. Specific audio signals include ITU-T recommendation P.I. 59 pseudo-conversational speech can be used. This pseudo audio signal is composed of a 2-channel audio signal as shown in FIG.

  FIG. 2 is a diagram for explaining the characteristics of a two-way call. The utterance section at the point A or the point B is indicated by the hatched portion. In FIG. 2, the section (1) is a silent section for both, the section (2) -A is a point A single utterance section, the section (2) -B is a point B single utterance section, and the section (3) is a double utterance. It is a section.

  For the calculation of the strain amount in the strain amount measuring unit 12, for example, ITU-T recommendation P.I. The algorithm defined in 862 is applied. The two-speaking section detecting unit 13 detects the two-speaking section by comparing the powers of the two channel signals of the point A voice signal and the point B voice signal as shown in FIG. 2 and provides the information to the weighting unit 14. To do.

  Based on the information obtained from the both-side utterance section detection unit 13, the weighting unit 14 includes the point A utterance section ((2) -A in FIG. 2) and the two-part utterance section (((2) in FIG. 2)). 3)) is distinguished, and the distortion time series as the output of the distortion measurement unit 12 is weighted averaged.

  For example, when the point A single utterance interval set used for evaluation is Ωs, both utterance interval sets are Ωd, both silent interval sets are Ωe, and the distortion time series is D (t) (t: time), for example, objective evaluation value Y Is determined as follows.

  The specific weighting coefficient α is optimized in advance so that the correlation between the subjective evaluation value (learning data) obtained by the subjective evaluation experiment and the objective evaluation value Y is the highest.

  FIG. 3 is a diagram showing an example of an audio quality objective evaluation process flow according to the first embodiment of the present invention. First, the distortion amount measuring unit 12 quantifies the distortion amount of the deteriorated speech as a time series by comparing the speech signal of the point A obtained from the speech DB 11 and the deteriorated speech obtained by passing this through the evaluation target system 2. (Step S1).

  The both-speaking section detecting unit 13 detects the both-speaking section by comparing the voice signal at the point A and the voice signal at the point B obtained from the voice DB 11 (step S2). Then, the weighting unit 14 weights and averages the distortion amount time series output from the distortion amount measurement unit 12 based on the information obtained from the two-utterance section detection unit 13 according to the above formula, and calculates the objective evaluation value Y (step) S3).

  FIG. 4 shows a block configuration of the second embodiment of the present invention. The voice quality objective evaluation device 3 includes a test signal database (DB) 31 that holds a test signal used for measuring the transmission delay time of the evaluation target system 2 between the points A and B, and the time and point at which the test signal is transmitted from the point A. A delay time measuring unit 32 that measures the transmission delay time by comparing the time when it was received in B, a two-way speech rate table 34 that holds correspondence information between the transmission delay time and the two-way speech rate, and a delay time measuring unit 32 The bilateral speech rate determination unit 33 that determines the bilateral speech rate by referring to the bilateral speech rate table 34 from the transmission delay time that is the output of the voice, and the audio signal generation unit that generates a 2-channel audio signal that realizes the bilateral speech rate 35. Distortion amount measuring unit 3 for quantifying the amount of distortion of degraded speech as a time series by comparing the speech signal of point A with the degraded speech obtained by passing the speech signal through the evaluation target system. The information obtained from the both-speaking section detecting unit 37 and the both-speaking section detecting unit 37 for detecting the both-speaking section by comparing the speech signal at the point A and the speech signal at the point B obtained from the speech signal generating unit 35. Is provided with a weighting unit 38 that weights the distortion amount time series output from the distortion amount measurement unit 36.

  In the first embodiment, on the assumption that the two-way utterance rate is constant, a method is used in which a voice signal for realizing this is stored in advance in the voice DB 11 shown in FIG. However, in general, the longer the transmission delay time between points A and B, the more difficult it is to have a conversation and the more likely that a conversation collision will occur. That is, the longer the transmission delay time, the higher the both-side utterance rate.

  Therefore, in this embodiment, the relationship between the transmission delay time and the two-way speech rate is prepared in advance as a table (two-way speech rate table 34), and an appropriate two-way speech rate is determined based on the result of measuring the transmission delay time. .

  FIG. 5 is a diagram illustrating a data configuration example of the two-way utterance rate table 34. For example, as shown in FIG. 5, the two-way utterance rate table 34 stores correspondence information between the transmission delay time (msec) and the two-way utterance rate (%). This two-way utterance rate table 34 is created by conducting a conversation experiment using the transmission delay time as a parameter, analyzing the two-way utterance rate of the conversation voice at this time, examining the correspondence between the two, and tabulating the results. can do.

  In order to obtain a voice signal that realizes the two-way utterance rate determined by the two-way utterance rate determining unit 33, the voice signal generating unit 35, for example, ITU-T Recommendation A pseudo two-channel audio signal is generated according to an algorithm based on 59. In the generation of this voice signal, the state transition probabilities between the four states of speaker A single utterance, speaker B single utterance, bilateral utterance, and bilateral silence are determined so as to satisfy the bilateral utterance rate, and correspond to each state The pseudo audio signal is transmitted to ITU-T recommendation P.22. Using the generation method defined in 50, a signal is generated so that speech feature quantities such as a long-time average spectrum, instantaneous amplitude distribution, and pitch frequency have an average characteristic.

  Instead of providing the voice signal generation unit 35, it is also possible to prepare a two-channel voice signal having various two-way speech rates in advance as a voice database and select and use a voice signal having an appropriate two-way voice rate.

  The subsequent operation is in accordance with the first embodiment. That is, the both-speaking section detecting unit 37 detects the both-speaking section by comparing the point A speech signal obtained from the speech signal generating unit 35 and the point-B speech signal, and the weighting unit 38 detects the both-speaking section detection. Based on the information obtained from the unit 37, the strain amount time series output from the strain amount measuring unit 36 is weighted.

  FIG. 6 is a diagram showing an example of an audio quality objective evaluation process flow according to the second embodiment of the present invention. First, the delay time measuring unit 32 measures the transmission delay time by comparing the time when the test signal is transmitted from the point A and the time when the test signal is received at the point B (step S11).

  The two-way utterance rate determining unit 33 determines the two-way utterance rate by referring to the two-way utterance rate table 34 from the measured transmission delay time (step S12). Based on the result, the voice signal generation unit 35 generates a two-channel voice signal (the voice signal at the point A and the voice signal at the point B) that realizes the determined bilateral speech rate (step S13).

  The distortion amount measuring unit 36 quantifies the distortion amount of the deteriorated sound as a time series by comparing the sound signal of the point A with the deteriorated sound obtained by passing the sound signal through the evaluation target system (step S14). The both-speaking section detecting unit 37 detects the both-speaking section by comparing the sound signal at the point A and the sound signal at the point B obtained from the sound signal generating unit 35 (step S15). The weighting unit 38 weights and averages the distortion amount time series output from the distortion amount measuring unit 36 based on the information obtained from the two-speaking section detecting unit 37 according to the above-described formula, thereby calculating the objective evaluation value Y (step S16). ).

It is a figure which shows the block configuration of the 1st Example of this invention. It is a figure explaining the characteristic of a two-way call. It is a figure which shows an example of an audio | voice quality objective evaluation processing flow. It is a figure which shows the block configuration of the 2nd Example of this invention. It is a figure which shows the data structural example of a both-side utterance rate table. It is a figure which shows an example of an audio | voice quality objective evaluation processing flow. It is a figure which shows the block structure of the conventional audio | voice quality objective quality evaluation apparatus.

Explanation of symbols

1,3,4 Speech quality objective evaluation system 2 Evaluation target system 11,41 Voice database (DB)
12, 36, 42 Distortion amount measurement unit 13, 37 Both utterance section detection unit 14, 38 Weighting unit 31 Test signal database (DB)
32 Delay time measurement unit 33 Both-side utterance rate determination unit 34 Both-side utterance rate table 35 Audio signal generation unit 43 Average processing unit

Claims (4)

An audio quality objective evaluation device for objectively evaluating audio quality from measurement results of physical features of audio signals through an evaluation target system from a first point to a second point,
A distortion amount measuring means for measuring the distortion amount of the deteriorated voice by comparing the voice signal of the first point with the deteriorated voice obtained by passing the voice signal through the evaluation target system, and quantifying the distortion amount as a time series;
A both-speaking section detecting means for detecting a both-speaking section by comparing the sound signal of the first point and the sound signal of the second point;
Weighting means for calculating a weighted average of distortion amount time series output from the distortion amount measuring means using a weighting value obtained by reducing the weight for the distortion amount of the detected both utterance intervals compared to a single utterance interval; A voice quality objective evaluation device characterized by comprising: An audio quality objective evaluation device for objectively evaluating audio quality from measurement results of physical features of audio signals through an evaluation target system from a first point to a second point,
A delay time measuring means for measuring a transmission delay time of an evaluation target system from the first point to the second point;
Based on the measured transmission delay time, a both-side speech rate determining means for determining a two-way speech rate from correspondence information between a predetermined transmission delay time and a two-way speech rate in a call;
An audio signal generating / selecting means for generating an audio signal for realizing the determined bilateral speech rate as an evaluation sound source or selecting from an audio database prepared in advance;
Distortion amount measuring means for measuring the distortion amount of the deteriorated speech by comparing the sound signal of the first point of the evaluation sound source with the deteriorated speech obtained by passing this through the evaluation target system, and quantifying it as a distortion time series When,
A both-speaking section detecting means for detecting a both-speaking section by comparing a voice signal of the first point and a second point of the evaluation sound source;
Weighting means for calculating a weighted average of distortion amount time series output from the distortion amount measuring means using a weighting value obtained by reducing the weight for the distortion amount of the detected both utterance intervals compared to a single utterance interval; A voice quality objective evaluation device characterized by comprising: An audio quality objective evaluation method for objectively evaluating audio quality from measurement results of physical features of audio signals through an evaluation target system from a first location to a second location,
A distortion amount measuring step of measuring a distortion amount of the deteriorated voice by comparing the voice signal of the first point and the deteriorated voice obtained by passing this through the evaluation target system, and quantifying the distortion amount as a time series;
A both-speaking section detecting step of detecting a both-speaking section by comparing the sound signal of the first point and the sound signal of the second point;
A weighting step of calculating a weighted average of the distortion amount time series using a weighting value obtained by reducing a weight for the distortion amount of the detected both utterance sections as compared to a single utterance section. Quality objective evaluation method. An audio quality objective evaluation method for objectively evaluating audio quality from measurement results of physical features of audio signals through an evaluation target system from a first location to a second location,
A delay time measuring step of measuring a transmission delay time of an evaluation target system from the first point to the second point;
A two-way utterance rate determining step for determining a two-way utterance rate from correspondence information between a predetermined transmission delay time and a two-way utterance rate in a call based on the measured transmission delay time;
An audio signal generation / selection step of generating an audio signal that realizes the determined bilateral speech rate as an evaluation sound source or selecting from an audio database prepared in advance;
A distortion amount measuring step of measuring the distortion amount of the deteriorated speech by comparing the sound signal of the first point of the evaluation sound source with the deteriorated speech obtained by passing this through the evaluation target system, and quantifying it as a distortion amount time series. When,
A both-speaking section detecting step of detecting a both-speaking section by comparing the sound signal of the first point of the evaluation sound source with the sound signal of the second point;
A weighting step of calculating a weighted average of the distortion amount time series using a weighting value obtained by reducing a weight for the distortion amount of the detected both utterance sections as compared to a single utterance section. Quality objective evaluation method.

Images