JP2002297199A - Method and device for discriminating synthesized voice and voice synthesizer - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method for discriminating whether or not an inputted aural signal is a synthesized aural signal. SOLUTION: The voice synthesizer 12 outputs an aural signal 18 which is a synthesis aural signal generated by performing voice synthesis according to a character string inputted thereto and added with discrimination information for discriminating that the voice is the synthesized aural signal, and a synthesized voice discrimination device 20 arranged on the receiving side of the aural signal detects the presence or absence of the discrimination information from the input aural signal, and judges that the input aural signal is the synthesized signal 18 when detecting this discrimination information.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a synthesized speech discriminating method and apparatus for discriminating a synthesized speech by a speech synthesizer from an actual uttered speech by a human, and a speech synthesis apparatus.

[0002]

2. Description of the Related Art The so-called speech rule synthesis technology for synthesizing speech from text as a character string has been remarkably advanced. Recently, as shown in FIG. 1, utterances of announcers and narrators are recorded, and the recorded generated voice 1 is converted into data so that it can be used by a computer. Information) is automatically acquired by the learning processing unit 3 using a statistical learning algorithm to create a control information database 4, and synthesized from the input text 5 by the speech rule synthesis processing unit 6 using the control information database 4. Techniques for generating speech are being established.

[0003] For such a technique, for example, X. Ha
ng et al “Recent Improvementson Microsoft's Trai
nable Text-To-Speech System-Whistler ”. IEEE Inte
rnational Conference on Acoustics, Speech, and Sig
nal Processing. pp.959-962, 1997, etc.

[0004] The synthesized speech obtained based on such a method well retains the characteristics of the recorded speaker such as the recorded announcer and narrator. In other words, the actual uttered voice of the recorded speaker and the synthesized voice produced by the speech rule synthesis are very similar in voice quality and inflection characteristics. Cannot be distinguished. In particular, when the voice is deteriorated due to band limitation, such as when the voice is transmitted through a transmission system such as a telephone line, it is difficult to distinguish the actually generated voice from the synthesized voice.

Further, a technique has been studied in which a user or a non-user's uttered voice is taken into a personal computer or the like by using a voice rule synthesis technique to generate a synthesized voice that imitates an actual uttered voice. In the future, it will be possible to easily synthesize the voices of oneself and others.

[0006]

As described above, the advancement of the speech rule synthesis technology has made it possible to generate a synthesized voice that is close to the actual uttered voice of the recorded speaker, and to synthesize the same voice as that generated by a celebrity, It offers various advantages, such as allowing e-mail to be read aloud by the sender.

[0007] On the other hand, if the voice rule synthesis technology is abused, it is possible to easily "spoof" to another person, so that, for example, a speaker identification or speaker verification technology for identifying or verifying a speaker by voice is meaningless. There is a concern that voiceprint analysis could affect criminal investigations.

SUMMARY OF THE INVENTION The present invention has been made in order to solve such a problem, and it is an object of the present invention to provide a synthesized speech discriminating method and apparatus capable of discriminating whether an input speech signal is a synthesized speech signal, and a speech synthesizing apparatus. Aim.

[0009]

In order to solve the above-mentioned problems, a synthesized speech discriminating method according to the present invention is characterized in that a synthesized speech signal generated by performing speech synthesis in accordance with an input character string is added to the synthesized speech signal. A basic feature is to output an audio signal to which discrimination information for discriminating the presence is added. It is desirable that the discrimination information added to the synthesized speech signal is information that is not lost even by transmission by an analog signal, band limitation, or sound quality deterioration. For example, the discrimination information can be obtained by changing the frequency characteristics of the synthesized speech signal. Is added. More specifically, as the change of the frequency characteristic applied to the synthesized speech signal, a change in signal power in a frequency band outside the main frequency band of the synthesized speech signal is used. Then, the presence or absence of discrimination information is detected from the input speech signal, and when the discrimination information is detected, it is determined that the input speech signal is a synthesized speech signal.

A speech synthesizer according to the present invention performs a speech synthesis according to an input character string to generate a synthesized speech signal, and determines whether the synthesized speech signal is a synthesized speech signal. A discriminating information adding unit that outputs an audio signal to which discriminating information is added.

A synthesized speech discriminating apparatus according to the present invention combined with the speech synthesizing apparatus detects the presence or absence of discrimination information added to the synthesized speech signal in the speech synthesizing apparatus from the input speech signal. It has a discriminating section for discriminating that the input speech signal is a synthesized speech signal, and a display section for displaying a discrimination result of this discriminating section.

Further, according to the present invention, a process of generating a synthesized voice signal by performing voice synthesis according to an input character string, and adding discrimination information for discriminating that the synthesized voice signal is a synthesized voice signal. And a program for causing a computer to execute the process of outputting the converted audio signal, or a recording medium on which the program is recorded.

Further, according to the present invention, the presence or absence of discrimination information added to the synthesized speech signal is detected from the input speech signal, and when the discrimination information is detected, the input speech signal is determined to be the synthesized speech signal. A program for causing a computer to execute a process and a process of displaying a result of the determination, or a recording medium on which the program is recorded can be provided.

As described above, according to the present invention, by outputting a speech signal in which discrimination information indicating that the speech signal is a synthesized signal is added to the synthesized speech signal, the input speech signal is input to the synthesized speech signal on the input side. Or a voice signal generated by a human. Therefore, it is possible to prevent improper use of speech synthesis in crime, that is, spoofing in a speaker identification or speaker verification system due to impersonation of another person, or disturbance of a forensic investigation based on speech. It is another object of the present invention to add discrimination information to a synthesized speech signal so that the side receiving the speech signal can easily determine whether or not the input speech signal is a synthesized speech signal.

[0015]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 2 is a block diagram showing a schematic configuration of a speech synthesis device and a synthesized speech discrimination device according to an embodiment of the present invention.

The voice synthesizing device 12 is provided on an information processing device such as a personal computer, for example, on a CD-ROM,
It is realized by executing dedicated software (sentence / speech synthesis software) supplied by a recording medium such as a flexible disk or a memory card or by a communication medium such as a network, and generates speech from input text. It has a sentence speech synthesis (TTS) processing function.

In the speech synthesizer 12, a text, which is a character string to be subjected to sentence speech synthesis (speech), is stored as a text file 11. According to the sentence / speech synthesis software, a text mixed with kanji or kana is read from the text file 11 and sentence / speech synthesis processing unit 13
And a speech corresponding to the input text is synthesized. The sentence-to-speech synthesis processing unit 13 is roughly divided into a text analysis unit 14 that analyzes input text and generates phonetic symbols.
And a voice generation unit 15 that generates voice from phonetic symbols.

The speech generation unit 15 synthesizes a speech having the characteristics of the specific speaker with reference to the voice dictionary 16 storing the voice quality and the prosodic features of the specific speaker. For details on obtaining voice characteristics and prosodic features of specific speakers and using them in speech rule synthesis, see, for example, the above-mentioned reference: X. Huang et al. “Recent Improve
ments on Microsoft's Trainable Text-To-Speech Syst
em-Whistler ”. IEEE International Conference on Ac
oustics, Speech, andSignal Processing .pp.959-96
2, 1997, etc., and the description is omitted here.

The conventional sentence speech synthesizer outputs a synthesized speech signal as it is. As described above, in a recent speech synthesizer capable of synthesizing a speech that retains the characteristics of a speaker, "spoofing" is easily permitted to a speaker who abuses the high-quality speech.

In order to avoid this problem, in the speech synthesizer 12 of the present embodiment, the synthesized speech signal output from the sentence speech synthesis processing unit 13 is passed through the synthesized speech discrimination information adding unit 17 and is sent to a human (in this case, Discriminating information indicating that the voice signal is different from the voice signal uttered by the (speaker) is added to the synthesized voice signal. In this manner, the speech signal 18 to which the discrimination information has been added by the synthesized speech discrimination information adding unit 17 is output from the speech synthesizer 12, and transmitted via the transmission path 19.

The synthesized speech discriminating device 20 discriminates whether the input speech signal input via the transmission line 19 is a synthesized speech signal or a human-generated speech signal 18.
The synthesized speech discrimination device 20 includes a discrimination unit 21 and a discrimination result display unit 22. The discrimination unit 21 detects the presence or absence of discrimination information from the input speech signal, and when the discrimination information is detected, discriminates that the input speech signal is the synthesized speech signal 18,
The discrimination result is displayed on the discrimination result display unit 22 by visual display or audio display.

Next, the synthesized speech discrimination information adding section 17 will be described in detail. The discrimination information added to the synthesized speech signal in the synthesized speech discrimination information adding unit 17 is information for discriminating that the synthesized speech signal is different from a voice signal generated by a human, and various forms can be considered. For example,
In the transmission of digital video and digital audio, there is a technique called “digital watermark” for preventing unauthorized copying. Basically, digital watermarking directly adds other information to image data or audio data directly to the extent that the image or audio is not perceived. If the voice is handled only as a digital signal, it is possible to add the discrimination information of the synthesized voice using this digital watermarking technique. However, since voice is often transmitted as an analog signal or transmitted in the air and then converted to an electrical signal before being transmitted, the discrimination information is based on analog signal transmission, band limitation in the transmission path, or sound quality degradation. Must not be lost by

Therefore, in the present embodiment, the synthesized speech discrimination information adding section 17 is configured to change the frequency characteristics of the synthesized speech signal. More specifically, the change in the frequency characteristic added to the synthesized voice signal by the synthesized voice determination information adding unit 17 is a change (attenuation or increase) in the signal power in a frequency band outside the main frequency band of the synthesized voice signal. Can be realized, for example, by using a notch filter in the synthesized speech discrimination information adding unit 17.

FIG. 3 shows an example of the frequency characteristic of the notch filter, which has a characteristic of greatly attenuating the signal power in a frequency band having a very narrow bandwidth on a relatively high frequency side. The center frequency of the notch filter, that is, the center frequency fd of the frequency band in which the notch filter attenuates the signal power of the synthesized audio signal is selected to be about 4 to 5 kHz, which is higher than the main frequency band of the audio signal, that is, 2 to 3 kHz.

When such a notch filter is used, the synthesized speech signal output through the synthesized speech discrimination information adding process 27 has a small deterioration in sound quality that can be perceived by human ears. When the analysis is performed, the signal power at the center frequency fd of the notch filter is extremely small, so that it can be easily determined that the voice is a mechanically synthesized voice.

The synthesized speech discriminating device 20 discriminates whether the input speech signal is a synthesized speech signal or a human-generated speech signal by detecting a decrease in the signal power of a specific frequency component by such a notch filter. FIG. 4 is a specific configuration example of the discrimination unit 21 in the synthesized speech discrimination device 20, and shows an example in which the synthesized speech discrimination information adding unit 17 is configured by the above notch filter.

The synthesized speech signal 18 which is an input speech signal is split into two, and on the other hand, passes through a band-pass filter (BPF) 31 having the same center frequency as the center frequency fd of the notch filter constituting the synthesized speech discrimination information adding section 17. And input directly to another power calculator 33.

In the power calculation unit 32, the synthesized speech signal 18
Of the frequency components passing through the band-pass filter 31 are obtained. In the power calculation unit 33, the signal power of the entire frequency band of the synthesized voice signal 18 is obtained. The division unit 34 divides the signal power of the frequency component obtained by the power operation unit 32 by the signal power of the entire frequency band obtained by the power operation unit 33.

The division result of the division unit 34 is output to the threshold processing unit 3
5 is compared with a predetermined threshold. If the result of the division is smaller than the threshold value in the threshold processing section 35, the input voice signal is determined to be a synthesized voice signal, and otherwise, it is determined to be a human uttered voice signal. This judgment result is
It is displayed by the determination result display section 22 of FIG.

Next, another embodiment of the present invention will be described. The basic configurations of the speech synthesis device 12 and the synthesized speech discrimination device 20 in the present embodiment are the same as those in the previous embodiment, and are as shown in FIG.

In the present embodiment, a notch filter whose center frequency changes according to a function of a predetermined time is used as the synthesized voice discrimination information adding unit 17. The manner of this change is made different depending on the maker name and model name of the sentence speech synthesizer 12. For example, as shown in FIGS. 5 and 6, in the X-type speech synthesizer of Company A, the time function of the frequency during the stop of the notch filter is F (t), and in the Y-type speech synthesizer, the frequency between the stop of the notch filter is F (t). Is the time function of G (t)
Unlike the Z-type speech synthesizer of Company B, the time function of the center frequency of the notch filter changes further differently from H (t).

FIG. 7 shows the configuration of the discriminating section 21 in the synthesized speech discriminating apparatus 20 according to the present embodiment. Discriminator 2 in FIG.
In 1, the notch filter of the synthesized speech discrimination information adding unit 17 has bandpass filters (BPF) 41, 42, and 43 having the same center frequency as the center frequency that changes with time according to the three functions described above, respectively. The audio signal input through is input. The output signals of the band-pass filters 41, 42 and 43 are output to the average power calculator 4
4, 45 and 46, and the average power in the utterance section is calculated. On the other hand, the average power in the utterance section of the entire frequency band of the input audio signal that does not pass through the band-pass filter is calculated by the average power calculator 47.

The dividing units 48, 49, and 50 calculate the average power in the vocal section of the output signals of the band-pass filters 41, 42, and 43 calculated by the average power calculating units 44, 45, and 46, respectively. The input voice signal is divided by the average power in the utterance section of the entire frequency band calculated by the unit 47. These divisions 48, 49,
The minimum value of the division result of 50 is obtained by the minimum value detection unit 50, and the minimum value is compared with a predetermined threshold value by the threshold processing unit 52, so that the division result is smaller than the threshold value,
In addition, a manufacturer name and a model name corresponding to the function having the minimum value are determined. In this case, the discriminator 21 assigns the functions F (T), G (t), H to the manufacturer name and model name as shown in FIG.
It is assumed that there is a table corresponding to (t).

The manufacturer name and the model name, which are the results of the determination by the determination section 21, are displayed by the determination result display section 22 of FIG. If the result of the division does not become smaller than the threshold value in any of the functions F (T), G (t), and H (t) shown in FIG. 6, the input audio signal is determined to be a human-generated audio signal. Displayed on the display unit 22.

According to this embodiment, not only can the input voice signal be determined to be a synthesized voice signal, but also the name of the manufacturer and model of the voice synthesizer that generated the synthesized voice signal can be specified. And more effective in crime prevention.

Although the embodiments of the present invention have been described above, the present invention is not limited to the above embodiments. For example, in the above embodiment, the change in the signal power in the specific frequency band of the audio signal is used as the synthesized voice determination information, but the phase information that is hardly perceived by humans is operated so as to be different from human uttered voices. Information, or a method of coarsely quantizing time information (pause length, phoneme duration time, utterance time, etc.), for example, 50 utterances
It is also possible to use a method of synthesizing so that it fits correctly in the time length of [ms] (it is impossible for a human to utter within a precise time). Further, a method of changing the center frequency of the notch filter for each phoneme is also possible, and according to this, it is possible to make deterioration in auditory perception inconspicuous.

In short, the present invention relates to a speech synthesizer originally intended to bring a synthesized speech signal closer to a human utterance, and to intentionally convert a part of the synthesized speech signal to a human utterance speech signal within a range that is difficult for a human to perceive. The main focus is on making it easy to distinguish the synthesized speech signal by making it different from
Various modifications can be made without departing from the scope of the invention.

[0038]

As described above, according to the present invention, it is possible to determine whether or not an input speech signal is a synthesized speech signal, and the speech synthesis can be abused to a crime, that is, spoofed to another person. In addition, it is possible to prevent spoofing in a speaker identification or speaker verification system and disturbance of a forensic investigation based on speech.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a speech rule synthesis system.

FIG. 2 is a block diagram showing a configuration of a speech synthesis device and a synthesized speech discrimination device according to an embodiment of the present invention.

FIG. 3 is a view showing frequency characteristics of a notch filter used in a synthesized speech discrimination information adding unit according to the embodiment;

FIG. 4 is a block diagram showing a configuration of a discriminating unit in the synthesized speech discriminating apparatus according to the embodiment;

FIG. 5 is a diagram illustrating a time function of a center frequency corresponding to a maker name and a model name of a speech synthesizer of a notch filter used in a synthesized speech discrimination information adding unit according to another embodiment of the present invention.

FIG. 6 is a diagram showing frequency characteristics corresponding to various time functions of a notch filter used in a synthesized speech discrimination information adding unit according to the embodiment;

FIG. 7 is a block diagram showing a configuration of a discriminating unit in the synthesized speech discriminating apparatus according to the embodiment;

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 11 ... Text file 12 ... Speech synthesizer 13 ... Sentence speech synthesis processing part 14 ... Text analysis part 15 ... Speech generation part 16 ... Speech dictionary 17 ... Synthesis speech discrimination information addition part 18 ... Synthesis speech signal 19 ... Transmission path 20 ... Synthesis Voice discriminating device 21: discriminating unit 22: discrimination result display unit

Claims (8)

[Claims] An audio signal in which discrimination information for discriminating a synthesized speech signal is added to a synthesized speech signal generated by performing speech synthesis in accordance with an input character string, is output. To determine the synthesized speech to be performed. 2. The method according to claim 1, wherein the determination information is added to the synthesized voice signal by changing a frequency characteristic of the synthesized voice signal. 3. The synthesized voice discrimination method according to claim 2, wherein the change of the frequency characteristic applied to the synthesized voice signal is a change of a signal power in a frequency band outside a main frequency band of the synthesized voice signal. . 4. A synthetic speech discriminating method comprising: detecting presence or absence of said discrimination information from an input speech signal; and discriminating that said input speech signal is said synthesized speech signal when detecting said discrimination information. 5. A voice synthesizing unit for generating a synthesized voice signal by performing voice synthesis according to an input character string, and a voice obtained by adding discrimination information for discriminating that the synthesized voice signal is a synthesized voice signal. And a discrimination information adding unit for outputting a signal. 6. A speech synthesizer according to claim 5, wherein the presence or absence of said discrimination information added to said synthesized speech signal is detected from said input speech signal. A synthesized speech discriminating apparatus comprising: a discriminating unit that discriminates a synthesized speech signal; and a display unit that displays a discrimination result of the discriminating unit. 7. A process for generating a synthesized voice signal by performing voice synthesis in accordance with an input character string, and generating a voice signal obtained by adding discrimination information for discriminating that the synthesized voice signal is a synthesized voice signal. A program for causing a computer to execute output processing. 8. A process for detecting presence / absence of discrimination information added to a synthesized speech signal from an input speech signal, and when detecting the discrimination information, discriminating that the input speech signal is the synthesized speech signal. And a program for causing a computer to execute a process of displaying a result of the process.