JP3068250B2 - Speech synthesizer - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech synthesizer for synthesizing speech by synthesizing a plurality of speech information parameters including spectrum information previously analyzed for each frame by frame using a rule synthesis method.

[0002]

2. Description of the Related Art In a conventional speech synthesizer of this type, when speech is synthesized using speech information parameters obtained by analyzing a sentence for each frame of a fixed time length, for example, spectrum information and speech information are obtained at fixed frame times. Synthesis is performed using parameters such as a residual (pulse). When a high-speed voice utterance is performed by such a voice synthesizer, a voiced or unvoiced voice, a vowel or a consonant is determined with respect to the spectrum information, and the frames determined to be voiced or vowels are thinned out (for example, Japanese Patent Application No. Hei. No. 58609).

[0003]

In such a conventional speech synthesizer, voiced or vowel frames are thinned out based on a fixed reference, and only speech synthesis is performed at a fixed utterance speed.

[0004]

The speech synthesizer of the present invention comprises:
In a speech synthesizer for editing and synthesizing a plurality of speech information parameters including spectrum information analyzed for each frame of a fixed time length in units of the frame, a predictive gain calculation means for the spectrum information and a plurality of predetermined gains are calculated. Selecting means for selecting one threshold value from the threshold values according to a speed instruction given from the outside, and comparing the predicted gain with the threshold value so that the predicted gain is larger than the threshold value. Is provided with control means for performing frame thinning.

[0005]

Next, the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram of one embodiment of the present invention. In the present embodiment, the spectrum information is converted to partial autocorrelation (PA
(RCOR) method. The average prediction residual signal power (Pe) in the frame is expressed by Expression (1) using the partial autocorrelation coefficient (Ki), which is one expression method of speech spectrum information.

[0007]

The preliminary gain (Pg) is expressed by the following equation (2).

P _g = P ₀ −P _e (2) where P ₀ indicates the average power of the input voice. In addition, the order P of the partial autocorrelation coefficient usually selects a value of about 10.

When the input speech is a periodic wave such as a vowel stationary part, the predictive gain (Pg) is a partial autocorrelation coefficient k
_{Since i} approaches 1, the value takes a value close to P _{0 as} can be seen from Expressions (1) and (2). When the input speech is an aperiodic wave such as a consonant part, the partial autocorrelation coefficient k _i
Approaches 0, the prediction gain (Pg) takes a very small value.

Therefore, by comparing the value of the preliminary gain (Pg) with the threshold value, the vowel frame and the consonant frame can be distinguished. By changing the threshold value to P ₀ , a more stable vowel frame can be continuously detected. Using this method, a plurality of threshold values corresponding to the utterance speed are learned while changing the vowel frame determination threshold value in advance, and a threshold value table is created based on the learned threshold values.

In FIG. 1, first, voice data necessary for synthesis is transmitted from a voice file 1 to a voice memory 2 and stored. The audio memory 2 receives the timing control of the frame control circuit 10 and transfers the spectrum information to the prediction gain calculator 3 and the buffer memory 6 for each frame, and transfers the residual to the buffer memory 7.

The prediction gain is calculated by the prediction gain calculator 3 and compared with the threshold value of the threshold value table 14 in which the threshold value previously learned and stored is stored in the judgment unit 4.
Which threshold value to use is determined by a threshold control circuit 13 according to a speed parameter given from the host CPU 12.
Determined by

When the predicted gain is equal to or larger than the threshold value, that is, when the frame is determined not to be thinned, the buffer memory 6 and the buffer memory 7 are controlled by the buffer control circuit 11 connected to the decision unit 4. Buffer memory 6,
Each of the data stored in 7 is sent to a synthesizing filter 8, and the synthesizing filter 8 synthesizes voice and outputs the voice.

If the predicted gain is equal to or smaller than the threshold value, the spectral information and residual for one frame stored in the buffer memories 6 and 7 are discarded, and each data for the next one frame is stored in the buffer memory 6. 7 is accumulated. Discarding of each data is performed by temporarily stopping the synthesis filter 8. This means that one frame has been thinned out.

[0016]

As described above, according to the present invention, a plurality of threshold values corresponding to the utterance speed are prepared in advance by learning, and the amount of frames to be thinned out is changed by changing the threshold value according to the utterance speed. By performing the control, the utterance speed can be variably controlled, which is impossible with a conventional speech synthesizer.

[Brief description of the drawings]

FIG. 1 is a block diagram of an embodiment of the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Audio file 2 Audio memory 3 Predictive gain calculator 4 Judgment device 6, 7 Buffer memory 8 Synthesis filter 10 Frame control circuit 11 Buffer control circuit 12 Host CPU 13 Threshold control circuit 14 Threshold table

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-64-9500 (JP, A) JP-A-61-122700 (JP, A) JP-A-1-93795 (JP, A) JP-A 63-122 234299 (JP, A) JP-A-63-199399 (JP, A) JP-A-59-082608 (JP, A) JP-A-62-102300 (JP, A) JP-A-3-206496 (JP, A) Patent 2758688 (JP, B2) (58) Fields investigated (Int. Cl. ⁷ , DB name) G10L 11/00-13/08 G10L 19/00-21/06 JICST file (JOIS)

Claims (1)

(57) [Claims] 1. A speech synthesizer for editing and synthesizing a plurality of speech information parameters including spectrum information analyzed for each frame of a fixed time length in frame units, and comprising: a predictive gain calculation means for the spectrum information; Selecting means for selecting one of the plurality of threshold values according to a speed instruction given from the outside, and comparing the predicted gain with the threshold value so that the predicted gain is And a control means for performing frame thinning when the threshold value is larger than the threshold value.