JP4514149B2 - Speech quality estimation apparatus and speech quality estimation method - Google Patents

Description

  The present invention relates to a speech quality estimation technique for estimating speech quality experienced by a user from physical features of speech without using the user.

As a conventional speech quality estimation technique, there is E-model disclosed in Non-Patent Document 1. According to this E-model, a quality parameter (sound codec, delay amount, volume, echo amount, etc.) is input and a communication quality index called an R value is output.
There are various voice bands for which quality is to be estimated, and typical voice communication service bands include a telephone band (300 to 3400 Hz) and a wide band (50 to 7000 Hz). The former telephone band is widely used as a telephone service on a telephone service network or an IP network, and the latter broadband is used as a video conference service or a high quality telephone service on a telephone service network or an IP network. Yes. The E-model is used when estimating the voice quality of the telephone band.

"The E-model, a computational model for use in transmission planning", ITU-T Recommendation G.107, 2003

  The E-mode 1 has been developed targeting the voice of the telephone band, and sufficient accuracy cannot be obtained when estimating the quality of the wideband voice in which the low frequency band and the high frequency band are extended from the telephone band. Specifically, since the maximum value of the communication quality index value in E-model is 93, a communication quality index value exceeding 93 is output in broadband communication that realizes clearer and higher quality voice communication. It cannot be said that the quality of high-quality broadband voice communication is sufficiently estimated.

  Designing the quality of networks and terminals so that users can be satisfied is important in providing voice communication services. However, since the target of E-model is voice in the telephone band, when E-model is used for network and terminal quality design, quality design can only be performed up to the highest quality in the telephone band, and higher quality broadband. There was a problem that the quality of voice communication service could not be designed.

  An object of the present invention is to derive a speech quality estimation value that can provide sufficient estimation accuracy for wider-band speech communication using speech quality estimation technology developed for a telephone band such as E-model. There is.

The present invention is a speech quality estimation device for estimating the quality of a target speech in a target speech frequency band, and estimates the quality of the target speech in a reference speech frequency band different from the target speech frequency band and outputs a quality index value Quality estimation means, a database for storing speech evaluation values for each voice frequency band in advance, a reading means for acquiring the voice evaluation values for each of the target voice frequency band and the reference voice frequency band from the database, Calculation means for calculating a correction value for correcting the quality index value to a value in the target audio frequency band based on the audio evaluation value of the target audio frequency band and the audio evaluation value of the reference audio frequency band; and the quality index Correction means for correcting the value by the correction value.
Further, in one configuration example of the speech quality estimation apparatus of the present invention, the database stores the speech evaluation value obtained in advance by a subjective quality evaluation experiment for evaluating a relationship between each speech frequency band and speech quality for each speech frequency band. To remember.
In the configuration example of the speech quality estimation apparatus according to the present invention, the database may store the speech evaluation value obtained in advance by objective quality measurement for measuring the relationship between each speech frequency band and speech quality for each speech frequency band. It is something to remember.
Moreover, one configuration example of the speech quality estimation apparatus of the present invention further includes a target speech frequency for obtaining the target speech frequency band based on a speech signal output from a communication device when the target speech frequency band is unknown. It has a band determination means.

  Also, the speech quality estimation method of the present invention includes a quality estimation procedure for estimating the quality of the target speech in a reference speech frequency band different from the target speech frequency band and outputting a quality index value; Based on the relationship between the frequency band and the voice quality, an evaluation value acquisition procedure for obtaining a voice evaluation value of each of the target voice frequency band and the reference voice frequency band, a voice evaluation value of the target voice frequency band, and the reference voice A correction value calculation procedure for calculating a correction value for correcting the quality index value to a value in the target audio frequency band based on a voice evaluation value in a frequency band; and a correction procedure for correcting the quality index value by the correction value; It is what has.

  According to the present invention, the correction value is calculated based on the voice evaluation value of the target voice frequency band and the voice evaluation value of the reference voice frequency band, and the quality index value of the target voice in the reference voice frequency band is corrected by the correction value. Thus, the quality index value of the target voice in the reference voice frequency band can be extended to the quality index value in the target voice frequency band. As a result, in the present invention, the relationship between each audio frequency band and audio quality is registered in advance in the database as an audio evaluation value, so that when using conventional quality estimation means such as E-model, the reference audio A quality index value corresponding to the subjective quality of the target voice in the target voice frequency band wider than the frequency band can be estimated with high accuracy.

  Further, in the present invention, by providing the target voice frequency band determining means, the quality index value corresponding to the subjective quality of the target voice in the target voice frequency band is estimated even when the target voice frequency band is unknown. Can do.

[First Embodiment]
Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing a configuration of a speech quality estimation apparatus according to the first embodiment of the present invention. The speech quality estimation apparatus of the present embodiment estimates the quality of the target speech Ain in the reference speech frequency band and outputs the quality index value R, and the speech subjective evaluation value for each speech frequency band. Subjective quality database 2 stored in advance, MOS data reading unit 3 for acquiring subjective speech evaluation values of the target speech frequency band and the reference speech frequency band to be subjected to quality estimation from the subjective quality database 2, and a MOS data reading unit 3 calculates a correction value for correcting the quality index value R to a value in the target audio frequency band from the interval scale conversion unit 4 that converts the acquired speech subjective evaluation value into the interval scale, and the evaluation value converted into the interval scale. A calculation unit 5 and a scaling unit 6 that corrects the quality index value R with a correction value are provided.

  The overall quality estimation unit 1 receives a target speech Ain that is a target of quality estimation from a communication device (not shown). The total quality estimation unit 1 estimates the total quality of the target voice Ain in the reference voice frequency band (which is a telephone band in the present embodiment, 300 to 3400 Hz), and outputs a quality index value R. The quality index value R corresponds to the subjective quality of the target voice Ain in the reference voice frequency band. As an example of the total quality estimation unit 1, there is an E-model disclosed in Non-Patent Document 1, for example. In the case of E-model, first, quality parameters (speech codec, delay amount, volume, echo amount, etc.) of the target speech Ain are obtained, and a quality index value R is obtained from the quality parameters.

  In the subjective quality database 2, speech subjective evaluation values for each sound frequency band are registered in advance. As an example of the speech subjective evaluation value, there is a mean opinion score (MOS). The MOS is an evaluation value evaluated by a subject on a five-point scale by listening to an audio signal sample. In the present embodiment, audio signal samples that are different in bandwidth but have no other deterioration are prepared for each audio frequency band. For example, documents “A. Takahashi, A. Kurashima, and H. Yoshino,“ Subjective quality ” The MOS for each audio signal sample obtained by the global test proposed in “index for compatibly evaluating narrowband and wideband speech”, MESAQIN2005, June, 2005 ”is registered in the subjective quality database 2. A configuration example of the subjective quality database 2 is shown in FIG. In the subjective quality database 2, the lower limit frequency and the upper limit frequency of the voice frequency band and the MOS of the voice frequency band are registered in association with each other.

  For example, the MOS data reading unit 3 accepts designation of a lower limit frequency Fl (for example, 50 Hz) and an upper limit frequency Fh (for example, 7000 Hz) of the target speech frequency band that is a target of quality design from a user of the speech quality estimation apparatus. Then, the MOS data reading unit 3 acquires from the subjective quality database 2 speech subjective evaluation values of the designated target speech frequency band (50 to 7000 Hz) and a predetermined reference speech frequency band (300 to 3400 Hz). Hereinafter, the speech subjective evaluation value in the target speech frequency band is MOSt, and the speech subjective evaluation value in the reference speech frequency band is MOSb. In the example of FIG. 2, MOSt is 4.3 and MOSb is 3.5.

  Since MOS is a quantity on a psychological scale, on the MOS scale, the relationship between the ratio of scale value differences and quality is not constant. For example, the psychological scale is saturated such that the quality is very good near the maximum scale value 5 and the quality is very bad near the minimum scale value 1, and the quality does not change much even if the scale value changes. It has the nature of Therefore, the interval scale conversion unit 4 converts the MOSt and MOSb acquired by the MOS data reading unit 3 into an interval scale in which the relationship between the scale value difference ratio and the quality is constant. Hereinafter, the evaluation values after the conversion of MOSt and MOSb are Rt and Rb, respectively. As an example of the interval scale conversion unit 4, one using a conversion formula that is proposed in Appendix I of Non-Patent Document 1 and converts a MOS into a quality index value R can be considered.

The calculation unit 5 calculates an expansion coefficient λ, which is a correction value of the quality index value R, from the evaluation values Rt and Rb converted by the interval scale conversion unit 4 according to the following equation (1).
λ = Rt−Rb (1)

Finally, the scaling unit 6 corrects the quality index value R output from the total quality estimation unit 1 with the expansion coefficient λ as shown in the equation (2), so that the total quality of the target speech Ain in the target speech frequency band is corrected. An index value R ′ is calculated. This total quality index value R ′ corresponds to the subjective quality of the target voice Ain in the target voice frequency band.
R ′ = R + λ (2)

  As described above, in the present embodiment, the extension coefficient λ is calculated based on the evaluation value Rt of the target voice frequency band and the evaluation value Rb of the reference voice frequency band, and the target voice in the reference voice frequency band is calculated based on the extension coefficient λ. The quality index value R can be expanded to the quality index value R ′ in the target audio frequency band by correcting the quality index value R.

  In the present embodiment, a subjective quality database 2 is used which stores a speech subjective evaluation value obtained by a subjective quality evaluation experiment for evaluating the relationship between each speech frequency band and speech quality. Instead, a database that stores voice evaluation values obtained by objective quality measurement for measuring the relationship between each voice frequency band and voice quality may be used for each voice frequency band.

[Second Embodiment]
Next, a second embodiment of the present invention will be described. FIG. 3 is a block diagram showing the configuration of the speech quality estimation apparatus according to the second embodiment of the present invention. The same components as those in FIG. The target voice Ain that is input to the overall quality estimation unit 1 described in the first embodiment is output from the voice communication device 11. In the present embodiment, the voice frequency used by the voice communication device 11 is used. This is an example when the bandwidth is unknown.

  In the present embodiment, a test signal Stest is input to the voice communication device 11 at the time of measuring the target voice frequency band of the voice communication device 11 in advance, and the voice output from the voice communication device 11 according to the test signal Test. A target voice frequency bandwidth determination unit 7 for obtaining a target voice frequency band is provided based on the communication device output voice Atest. As the test signal Test at this time, a signal having all frequency band components such as white noise and pink noise can be used. In order to determine the target voice frequency band, for example, the frequency lower than the average power of the pass band by xdB (x is an arbitrary value) may be set as the lower limit frequency Fl and the upper limit frequency Fh of the target voice frequency band.

  A determination example of the target audio frequency band will be described with reference to FIG. 4A is a diagram showing the frequency characteristics of the test signal Stest, and FIG. 4B is a diagram showing the frequency characteristics of the voice communication apparatus output voice Atest. In the present embodiment, white noise having flat characteristics in the entire frequency band is used as the test signal Test as shown in FIG. When white noise is input to the voice communication device 11, only the signal in the pass band passes through the voice communication device 11, so that the band-limited voice communication device output voice Atest is output as shown in FIG. Here, assuming that the frequency that is 3 dB lower than the average power of the pass band is the lower limit frequency Fl and the upper limit frequency Fh of the target voice frequency band, it can be determined that the lower limit frequency Fl is 50 Hz and the upper limit frequency Fh is 7000 Hz. .

The lower limit frequency Fl and the upper limit frequency Fh of the target voice frequency band obtained by the target voice frequency bandwidth determination unit 7 are notified to the overall quality estimation unit 1 through the reception unit 8.
After the measurement of the target voice frequency band is completed, the target voice frequency bandwidth determination unit 7 stops the input of the test signal Test to the voice communication device 11 and returns the voice communication device 11 to a normal state. Thereafter, the target voice Ain is input from the voice communication device 11 to the total quality estimation unit 1. The operations of the overall quality estimation unit 1, the subjective quality database 2, the MOS data reading unit 3, the interval scale conversion unit 4, the calculation unit 5 and the scaling unit 6 are the same as those in the first embodiment.

  As described above, in the present embodiment, by providing the target audio frequency bandwidth determination unit 7, the quality of the target audio in the target audio frequency band is estimated even when the target audio frequency band is unknown. Can do.

[Third Embodiment]
Next, a third embodiment of the present invention will be described. FIG. 5 is a block diagram showing the configuration of a speech quality estimation apparatus according to the third embodiment of the present invention. The same components as those in FIG. In this embodiment, an interval scale quality database 9 and a quality data reading unit 10 are provided instead of the subjective quality database 2, the MOS data reading unit 3, and the interval scale conversion unit 4 in the first embodiment.

  In the interval scale quality database 9, an evaluation value obtained by converting a speech subjective evaluation value for each audio frequency band into an interval scale in advance is registered. Similar to the first embodiment, an example of the speech subjective evaluation value is MOS. In order to convert the MOS into the interval scale, the conversion formula described in the interval scale conversion unit 4 of the first embodiment may be used. A configuration example of the interval scale quality database 9 is shown in FIG. In the interval scale quality database 9, the lower limit frequency and the upper limit frequency of the voice frequency band and the interval scale value of the voice frequency band are registered in association with each other.

  The quality data reading unit 10 accepts designation of the lower limit frequency Fl (for example, 50 Hz) and the upper limit frequency Fh (for example, 7000 Hz) of the target audio frequency band from the user, and the specified target audio frequency band and a predetermined reference audio frequency band The interval scale values Rt and Rb of (300 to 3400 Hz) are acquired from the interval scale quality database 9 and output. In the example of FIG. 6, the interval scale value Rt of the target audio frequency band is 100, and the interval scale value Rb of the reference audio frequency band is 80.

The operations of the overall quality estimation unit 1, the calculation unit 5, and the scaling unit 6 are the same as those in the first embodiment.
As described above, in this embodiment, the interval scale conversion unit 4 can be omitted by converting the speech subjective evaluation value into the interval measure in advance.

  The present invention can be applied to a speech quality estimation technique.

It is a block diagram which shows the structure of the audio | voice quality estimation apparatus used as the 1st Embodiment of this invention. It is a figure which shows the structural example of the subjective quality database in the 1st Embodiment of this invention. It is a block diagram which shows the structure of the audio | voice quality estimation apparatus used as the 2nd Embodiment of this invention. It is a figure which shows the frequency characteristic of the audio | voice communication apparatus output audio | voice output from an audio | voice communication apparatus according to a test signal and a test signal. It is a block diagram which shows the structure of the audio | voice quality estimation apparatus used as the 3rd Embodiment of this invention. It is a figure which shows the structural example of the space | interval scale quality database in the 3rd Embodiment of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Total quality estimation part, 2 ... Subjective quality database, 3 ... MOS data reading part, 4 ... Space | interval scale conversion part, 5 ... Calculation part, 6 ... Scaling part, 7 ... Target audio | voice frequency bandwidth determination part, 8 ... Reception , 9 ... interval scale quality database, 10 ... quality data reading unit, 11 ... voice communication device.

Claims (5)

A speech quality estimation device for estimating the quality of a target speech in a target speech frequency band,
Quality estimation means for estimating the quality of the target voice in a reference voice frequency band different from the target voice frequency band and outputting a quality index value;
A database that pre-stores voice evaluation values for each voice frequency band;
Reading means for acquiring the voice evaluation value of each of the target voice frequency band and the reference voice frequency band from the database;
Calculation means for calculating a correction value for correcting the quality index value to a value in the target voice frequency band based on the voice evaluation value of the target voice frequency band and the voice evaluation value of the reference voice frequency band;
A speech quality estimation apparatus comprising: a correction unit that corrects the quality index value by the correction value. The speech quality estimation apparatus according to claim 1, wherein
The speech quality estimation apparatus characterized in that the database stores, for each speech frequency band, the speech evaluation value obtained in advance by a subjective quality evaluation experiment for evaluating a relationship between each speech frequency band and speech quality. . The speech quality estimation apparatus according to claim 1, wherein
The speech quality estimation apparatus, wherein the database stores, for each speech frequency band, the speech evaluation value obtained in advance by objective quality measurement for measuring a relationship between each speech frequency band and speech quality. The speech quality estimation apparatus according to claim 1, wherein
The speech quality estimation apparatus further comprising target speech frequency band determining means for obtaining the target speech frequency band based on a speech signal output from a communication device when the target speech frequency band is unknown. A speech quality estimation method for estimating a target speech quality in a target speech frequency band,
A quality estimation procedure for estimating the quality of the target voice in a reference voice frequency band different from the target voice frequency band and outputting a quality index value;
Based on the relationship between each voice frequency band and voice quality obtained in advance, an evaluation value acquisition procedure for obtaining a voice evaluation value of each of the target voice frequency band and the reference voice frequency band;
A correction value calculation procedure for calculating a correction value for correcting the quality index value to a value in the target audio frequency band based on the audio evaluation value of the target audio frequency band and the audio evaluation value of the reference audio frequency band;
And a correction procedure for correcting the quality index value by the correction value.

Images