JPH0887295A - Sound source data generating method for voice synthesis - Google Patents

Abstract

PURPOSE: To provide a sound source data generating method for voice synthesis capable of eliminating tone quality degradation due to amplitude abnormality in synthesized wave forms associated with an energy control in a voice sythesizer. CONSTITUTION: When the normalization of the sound source of sound source data is carried out in order to control the energy of the sound source based on a rhythm control, a maximum amplitude value of the voice waveform of an original sound used for the normalization is corrected every sound source data using a beforehand generated table (S7). The normalization is performed by dividing the sound source of segmented voice waveforms by the corrected maximum amplitude value (S8). Thus, the energy transition from a consonant to a vowel and a vowel to a consonant is made smooth and the tone quality degradation of a synthesized voice is eliminated. The table for the amplitude correction is produced by visually checking the waveform of synthesized voice produced by the sound source data which are normalized by the maximum amplitude value, confirming the abnormal frame of waveform amplitude by a pitch period unit and tabulating amplitude correction values, which are suitable to eliminate abnormality, every voice data.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for producing sound source data for voice synthesis used in a Japanese rule voice synthesizer or the like for performing voice synthesis by amplitude control using voice data as a sound source.

[0002]

2. Description of the Related Art FIG. 2 shows a flow of generating sound source data for speech synthesis and speech synthesis in a conventional regular speech synthesizer. Generally, in speech synthesis, input text is converted into a phoneme symbol string by Japanese processing or the like, and the time length (speech duration), pitch (sound pitch), An energy (sound volume) pattern is generated, and a speech waveform is synthesized from speech data based on these prosody control patterns.

Here, the sound source data is composed of a vocal tract cross-sectional area pattern representing a vocal tract characteristic from the human throat to the mouth and a sound source, and has a CV-VC data format (where C is a consonant and V is a vowel). I am collecting. FIG. 3 and FIG. 4 show examples of the voice waveform of "ma". 3 shows the transition from / M (consonant C) to / A / (vowel V), and FIG. 4 shows the overall shape. Frames F1 to F6 in pitch period units in FIG. 3 are consonants / M
/, Frame F7-is the vowel / A /.

The sound source data is obtained by analyzing a target waveform of a voice to be synthesized (a waveform of a natural voice (original sound)), and for each frame (F1 to F14) of each pitch period, the vocal tract cross-sectional area pattern and the primary order. A sound source is extracted, and the primary sound source is divided by the maximum amplitude value H of the analysis target waveform to be a normalized sound source, and these vocal tract cross-sectional area patterns and the normalized sound source are sound source data for speech synthesis. It is stored in a database and used.

In speech synthesis, amplitude control is performed based on the control of the energy control unit. Therefore, as described above, the sound source prepared as sound source data corresponds to the primary sound source obtained from the analysis result of the original sound waveform. A sound source is used which is normalized by dividing by the maximum amplitude of the sound waveform of the original sound. By the above amplitude control, the normalized sound source is multiplied by the energy E ′ at the time of synthesis, and this becomes a synthesized voice through a filter having the vocal tract characteristic of the vocal tract cross-sectional area pattern of the corresponding voice data.

[0006]

However, in the sound source normalization method according to the above-mentioned conventional technique, since the maximum amplitude of the original sound waveform is used for normalization, a gap may appear at the waveform level during synthesis depending on the shape of the waveform. There are cases. Taking FIG. 3 as an example, a sinusoidal waveform such as / M /
If a normal waveform with a complicated waveform such as A / is normalized with the same reference, the waveform transition of the original sound is often impaired, which leads to deterioration in sound quality. Specifically, when the amplitude of the consonant part is larger than that of the vowel part, the intelligibility is lost, and when a more local abnormal waveform amplitude appears, a large abnormal sound is produced. FIG. 5 is a diagram showing an example of a waveform of a synthetic voice "Amai", which is a consonant at the boundary between / A / and / M /.
The amplitude of M / is large and abnormal. Consonant / M /
As a whole, the amplitude is larger than that of the vowel / A /.
In addition to the normalization problem, such anomalies are also considered to be factors such as decimation for controlling the time length.

The present invention has been made to solve the above problems, and an object thereof is to eliminate the sound quality deterioration due to the amplitude abnormality of the synthesized waveform accompanying the energy control in the regular speech synthesizer or the like. It is to provide a sound source data creation method for voice synthesis.

[0008]

In order to achieve the above-mentioned object, a method for creating sound source data for speech synthesis according to the present invention analyzes a speech waveform of natural speech to be synthesized and extracts a sound source of the speech waveform. In the method of creating a sound source data for voice synthesis, which is normalized with the maximum amplitude value and uses the normalized sound source as a part of the sound source data, when the synthesized voice created using the sound source data is observed and an abnormal amplitude is recognized, An amplitude correction value for eliminating the amplitude abnormality is made into a table in advance, and when the sound source is normalized, the corresponding amplitude correction value in the table is read to correct the maximum amplitude value, and the corrected maximum amplitude value is used. The feature is that the sound source is normalized.

[0009]

In the method of creating the sound source data for speech synthesis of the present invention,
When normalizing the sound source of the sound source data to control the energy of the sound source based on the prosody control, the maximum amplitude value of the natural speech waveform used for normalization is synthesized by the sound source that is normalized by the maximum amplitude value. By correcting each sound source data by using a table that is analyzed and created in advance, the energy transition of consonant-vowel and vowel-consonant is smoothed, and the deterioration of the sound quality of synthesized speech is eliminated.

[0010]

Embodiments of the present invention will now be described in detail with reference to the drawings.

FIG. 1 is a flow chart of sound source data creation showing an embodiment of the present invention. In the figure, S1 to S9 indicate processing steps. First, the announcer or the like records the original sound (natural sound) of the target sound to be synthesized (S1). Recording is done on tape etc. with analog signal as it is. Next, the recorded original sound is reproduced, A / D-converted, sampled at an appropriate sampling frequency and made into a file (S
2). Note that it is preferable to cut the extra frequency component of the original sound by passing it through a low-pass filter or a high-pass filter, or by filtering after A / D conversion. Next, various control points are manually determined by visually observing the filed voice waveform (S
3). Various control points include the waveform cutout range, C /
There are parameters such as V boundary and prosody control. Next, a waveform mixing process is performed as a pre-process (S4). Here, the connectivity between the waveform data and the waveform data is improved. Next, waveform analysis is performed for each target waveform (S5), and the vocal tract cross-sectional area pattern and the primary sound source are extracted for each frame.

Next, the waveform data is cut out (S6),
Although the normalization is performed, in the past, the normalization was automatically performed using the maximum amplitude value of the waveform, whereas in the present embodiment, the amplitude correction that is made in the table in advance is performed in the normalization. The value (magnification) is read at the time of data extraction and is corrected by multiplying it by the maximum amplitude value used for normalization of the sound source data (S
7) Then, the primary sound source is divided by the corrected value to perform normalization processing (S8). As a result, the sound source data composed of the vocal tract cross-sectional area pattern and the normalized sound source is finally obtained (S9).

In the above method, the amplitude correction value table is created by visually checking the waveform of the synthesized voice created from the sound source data normalized by the maximum amplitude value as in the conventional method, and determining the waveform amplitude in pitch cycle units. This is performed by confirming the abnormal frame of No. 2 and making a table of amplitude correction values suitable for eliminating the abnormalities for each audio data. Note that the correction process can be simplified by setting the magnification to 1.0 for frames that do not require correction.

In the conventional method in which the maximum amplitude of the waveform is automatically normalized when the sound source data is created, and the pattern obtained by the prosody control is applied during synthesis, the consonant and vowel consistency may be poor depending on the waveform shape. , The balance of the overall energy transition of synthetic speech may be lost. In this embodiment, the maximum amplitude value used for sound source data normalization is corrected by a table created by analyzing a voice waveform synthesized in advance with the sound source data according to the above-described conventional method, so that the amplitude is very smooth at the synthesized waveform level. It is possible to realize the transition, and it is possible to eliminate the sound quality deterioration due to the amplitude abnormality of the synthesized waveform accompanying the energy control in the regular speech synthesizer.

[0015]

As is apparent from the above description, according to the sound source data generation method for voice synthesis of the present invention, the normalization is performed at the time of normalization accompanying the energy control at the time of generating the sound source data used for the synthesized voice. It is possible to eliminate the abnormal synthesized waveform amplitude caused by the defect, and it is possible to eliminate the sound quality deterioration.

[Brief description of drawings]

FIG. 1 is a flow chart of sound source data creation showing an embodiment of the present invention.

FIG. 2 is a flow chart of sound source data creation and voice synthesis explaining a conventional technique.

FIG. 3 is a diagram showing a CV transition section of a voice waveform example.

FIG. 4 is a diagram showing an entire example of the above speech waveform.

FIG. 5 is a diagram showing an example of a waveform of synthesized speech according to a conventional technique.

[Explanation of symbols]

 S1 to S9 ... Processing steps

Claims (1)

[Claims] 1. A sound source for voice synthesis in which a sound source extracted by analyzing a voice waveform of a natural voice to be synthesized is normalized by a maximum amplitude value of the voice waveform, and the normalized sound source is a part of sound source data. In the data creation method, the synthesized voice created using the sound source data is observed, and amplitude correction values for eliminating the amplitude abnormalities are preliminarily tabulated when the amplitude abnormality is recognized, and when normalizing the sound source, A method for generating sound source data for speech synthesis, which comprises reading out a corresponding amplitude correction value in the table, correcting the maximum amplitude value, and normalizing the sound source with the corrected maximum amplitude value.