JP3059751B2 - Residual driven speech synthesizer - Google Patents

Description

The present invention relates to a speech synthesizer capable of uttering arbitrary words, and more particularly to a residual drive type speech synthesizer that performs residual drive.

(B) Conventional technology In recent years, research on rule synthesis methods for synthesizing speech from arbitrary sentences has been actively conducted, and currently, prototypes have been produced and practically used in newspaper review devices and reading machines for blind people. There is.

A rule synthesizer for synthesizing speech from an arbitrary sentence, for example, performs sentence analysis on text input to determine pronunciation and accent, and, based on phoneme rules, a speech unit (synthesis unit) which is a necessary synthesis unit. For example, CVC units are determined and combined, the pitch of the voice is determined from the prosodic rules, a time series of voice parameters and a pitch pattern are generated, and a sound source and a digital filter are generated from these parameters. Is generated to generate a synthesized speech.

As speech parameters used for such a speech synthesis method, LSP and formants are generally used.
Impulses and white noise were used as sound sources to reduce memory and simplify processing.

Thus, it has been proven that in speech synthesis of a linear prediction system such as LSP, a synthesized speech close to the original speech can be obtained by using the prediction residual as a driving sound source. As shown in “Transaction Papers 1-2-16”, a method using a residual as a driving sound source has also been proposed for rule synthesis. In this method, residual waveforms are stored for all speech units together with speech units, which are synthesis units used for rule synthesis, and used as a driving sound source during speech synthesis.

However, when the residual is used as a driving sound source in rule synthesis, the following problem occurs. That is, in the rule synthesis, it is necessary to be able to arbitrarily change the pitch period of the sound source in order to generate synthesized sounds at various pitch periods. When the impulse and white noise are used as the sound source, the pitch period can be changed only by changing the time interval of the impulse, but when the residual is used as the driving sound source, the pitch period of the residual is changed by some method. Must be changed.

Therefore, in general, as shown in the above-mentioned collection of lecture papers, when the pitch period is lengthened, 0 is added to the extended portion.
Are shortened, the pitch cycle is changed by truncating the residual waveform on the way. At this time, when the residuals are connected for each changed pitch cycle,
Distortion occurs in the spectrum of the residual, which causes sound quality degradation.

On the other hand, as shown in the latest “Annual Papers of the Acoustical Society of Japan Spring Research Conference, 2-7-18,” the extraction of the residual so that the spectral distortion caused by the change in the pitch period is minimized. A method for determining the position has been proposed.
Although it has been reported that a high-quality synthesized voice could be obtained, the degradation of the synthesized voice is large for a female voice that is greatly affected by a pitch cycle change due to truncation to zero.

(C) Problems to be Solved by the Invention In order to solve the above-mentioned problems, the present invention focuses on the fact that there is a correlation between the amount of change in pitch and the amount of deterioration in sound quality. On the other hand, by storing a plurality of residuals having different pitch periods, a residual having a pitch period closest to the pitch period of the voice to be synthesized is selected from among the residuals, and this is used as a driving sound source. It is an object of the present invention to realize a residual drive type speech synthesizer capable of avoiding degradation of speech synthesis.

(D) Means for Solving the Problems A residual driving type speech synthesizer according to the present invention has a first memory for storing speech segments, which are a sequence of speech parameters required for speech synthesis, and a speech unit. A second memory for storing the corresponding residuals,
A phoneme symbol string generation unit that generates a symbol string indicating a required speech unit from the content to be uttered, a pitch pattern generation unit that determines a change in pitch period from the utterance content, and a symbol generated by the phoneme symbol string generation unit A speech unit connection unit for sequentially connecting necessary speech units based on the sequence, a speech synthesis filter for synthesizing speech using speech parameters included in the connected speech units as coefficients, and a residue corresponding to the speech unit. Using the difference as the driving sound source,
According to the pitch cycle at each point in time determined by the pitch pattern generator, the pitch cycle of the residual is changed and input to the synthesis filter, and the driving sound source generator is stored in the second memory. It comprises a residual selection circuit for selecting a specific residual from a plurality of residuals.

(E) Operation In the residual drive type speech synthesizer, changing the pitch cycle
Conventionally, zero data has been inserted into a part of the residual or truncated, but the sound quality deteriorates. However, according to experiments, when the pitch cycle of the residual is changed, the relationship between the pitch change amount and the sound quality is as follows.
Even when the pitch period is shortened (increased pitch), the larger the amount of change in the pitch period, the worse the subjective evaluation value and the sound quality is degraded.

Since the residual driving type speech synthesizer of the present invention is provided with a residual selection circuit for selecting a specific residual from a plurality of residuals stored in the second memory, A plurality of residuals having different pitch periods are stored in the second memory corresponding to the speech units stored in the memory of (a), and according to the pitch period at each time point determined by the pitch pattern generation unit, The above-described residual selection circuit selects an appropriate residual of the pitch period from the second memory, and the driving sound source generation unit can change the required pitch period with respect to the selected residual.

(F) Embodiment In order to compare the present embodiment with a residual drive type speech synthesizer of the present invention, a conventional apparatus will be described first.

FIG. 1 shows an example of the configuration of a general residual drive type rule synthesizing apparatus. However, this figure does not include the language processing part, and the input is performed using kana characters and accent position information.

According to the apparatus shown in the figure, first, input information is input to the character string buffer (1). For example, "*
* Is input, the phoneme symbol string generation unit (2) converts the input information stored in the character string buffer (1) into phoneme symbols indicating necessary speech units. In this column, to describe the case where the synthesis unit is a CV element,
A phoneme symbol string as shown in FIG. 2 is stored in the phoneme symbol string buffer (3).

The speech unit memory (4) stores speech parameters corresponding to each CV segment, for example, LSP coefficients and the like, and necessary speech is stored in accordance with the phoneme symbols stored in the phoneme symbol string buffer (3). Units are sequentially read from the speech unit memory (4) to the speech unit connection unit (5). Then, the read speech units are connected by a speech unit connection unit (5),
After adjustment of connection length and interpolation processing, audio parameters
Buffer (6).

On the other hand, from the accent information (*) stored in the character string buffer (1) and the information (space) indicating a break in a phrase, a pitch change pattern is generated in a pitch pattern generation unit (7). You. FIG. 4 illustrates the process of generating a pitch pattern by way of an example of "***". As shown in FIG. 3 (a), a frame descending over the entire text. The pitch component shown in FIG. 3 (b), which descends immediately after the accent position (*), is added to the pitch component, thereby generating a pitch change pattern shown in FIG. 3 (c).

The residual waveform memory (8) stores a residual waveform and its pitch period corresponding to each speech unit. The residual waveform corresponding to the sequentially read speech unit and its pitch are stored. The period is read out by the driving sound source generation section (9), and the pitch change pattern generated by the pitch pattern generation section (7).
After the pitch is changed according to the pitch, it is stored in the connected driving sound source buffer (10).

The driving sound source stored in the driving sound source buffer (10) is input to the synthesis filter (11) as a sound source, and the sound parameters are output.
The synthesized speech is generated using the speech parameters stored in the data buffer as coefficients of the synthesis filter (1). The synthesized voice is converted into an analog signal by a DA converter (12) and is emitted by a speaker (13).

As shown in FIG. 4, a driving sound source generating section (9) of the conventional residual driving type speech synthesizer of the present invention has a residual waveform memory (8).
The residual waveform read from the residual waveform buffer (91)
Then, the pitch cycle of the residual waveform is stored in the residual pitch cycle register (92). On the other hand, the pitch change pattern generated by the pitch pattern generation section (7) is stored in the generated pitch cycle buffer (93) in the form of the pitch cycle at each time. Then, of the pitch periods stored in the generated pitch period buffer (93), the pitch period of the voice to be synthesized at that time is set in the target pitch period register (94). Differentiator (95) is a residual pitch period register (92)
Calculates the difference between the pitch cycle of the read residual waveform stored in the target pitch cycle and the pitch cycle of the voice to be synthesized set in the target pitch cycle register (94). To store. When the pitch cycle change value is positive based on the contents of the pitch cycle change value register (96), the pitch control circuit (97) applies the change value to the residual stored in the residual waveform buffer (91). The pitch period is lengthened by inserting zero data by the minute, and when the pitch period change value is negative, the pitch period is shortened by truncating the residual waveform, and the residual period is stored in the driving sound source buffer (10). Store the waveform.

With the above-described configuration, a voice having a desired pitch change pattern can be synthesized. In such a conventional method, for example, the pitch cycle is 33, that is, the pitch frequency is 300 Hz (sampling) in the residual waveform memory (8). If the residual waveform corresponding to the speech unit “ta” (when the period is 10 KHZ) is stored, the first “ta” of “ta * nii **” is about 40 on average.
Since 0HZ and the last "ta" must be synthesized at an average of about 220HZ, it is necessary to round down to zero with a pitch cycle of about 10 (400HZ = pitch cycle 25, 220HZ = pitch cycle 4
5) The residual spectrum is distorted and the synthesized speech is degraded. In the case of longer sentences, the change amount of the pitch period may increase.

On the other hand, FIG. 5 shows a configuration example of a driving sound source generation unit (9) for realizing the present invention.

In the device of the present invention shown in FIG.
A plurality of residual waveforms having different pitch periods, for example, six types,
The head addresses which are stored in the residual waveform memory (8) and store six types of residual waveforms corresponding to each speech unit are the residual address register 1 (981) to the residual address register 6 ( 986). At the same time, the pitch cycle of each residual waveform is read out and set in the residual pitch cycle register 1 (921) to the residual pitch cycle register 6 (926).

FIG. 6 shows an example of the pitch cycle set in the residual pitch cycle register 1 (921) to the residual pitch cycle register 6 (926). As shown in FIG. 6, residual waveforms of various pitch periods are stored in the residual waveform memory (8) corresponding to each speech unit.

On the other hand, the pitch change pattern generated by the pitch pattern generation section (7) is stored in the generated pitch cycle buffer (93) in the form of the pitch cycle at each time. Then, of the pitch periods stored in the generated pitch period buffer (93),
The pitch period of the sound to be synthesized at that time is set in the target pitch period register (94).

In the example of FIG. 3, first, from the pitch frequency 400 Hz corresponding to the first "ta", the pitch period 25 (the sampling frequency 1
0KHZ) is set in the target pitch period register (94). The differentiator (95) first reads out the 20 stored in the residual pitch period register (921), takes the difference from 25 which is the value of the target pitch period register (94), and outputs the difference value 5. , Its value is taken into one input of a comparator (99). The pitch period change value register (96) connected to the output of the comparator (99) stores the smallest difference value at the present time, and 100 is input as a large initial value. The comparator compares 100 set in the pitch cycle change value register (96) with 5 which is the output of the differentiator (95), and outputs the smaller value 5 of the absolute value to change the pitch cycle. The value is set in the value register (96) and the residual address register 1 (98) corresponding to the residual pitch period register (921) input to the difference circuit (95) at that time.
Set the contents of 1) in the residual address register (98).

Next, the differentiator (95) sets the residual pitch period register (92
2) The 26 stored in 2) is read out, the difference from 25 which is the value of the target pitch period register (94) is obtained, and the difference value 1 is output, and the value is taken into one input of the comparator (99). It is. The pitch period change value register (96) connected to the output of the comparator (99) stores the smallest difference value 5 at the time. The comparator uses the pitch period change value register (9
6) is compared with 1 which is the output of the differentiator (95), and the smaller absolute value is output as 1.
The content of the residual address register 1 (982) corresponding to the residual pitch period register (922) input to the difference circuit (95) at that time is set in the pitch cycle change value register (96). Set in the address register (98). Conversely, when the absolute value of the value set in the pitch cycle change value register (96) is smaller, the value of the residual address register (98) is stored as it is.

By repeating the above operation, the pitch cycle 26 closest to the pitch cycle 25 to be synthesized is selected from the residual pitch cycle registers 1 to 6, and the difference value thereof, that is, 1 which is the amount to change the pitch cycle Is set in the pitch cycle change value register (96). The residual address register (98) includes a selected residual pitch period register 2
The value of the residual address register 2 (982) corresponding to (922) is set. Then, when synthesizing the first "ta", the residual is read from the address stored in the residual address register (98) into the residual waveform buffer (91), and the pitch is controlled by the pitch control circuit (97). Zero data is inserted by one.

Similarly, for the last “ta”, the leading address of the residual address register 5 (985) of the pitch cycle 44 is stored in the residual address register (98), and the residual waveform is stored according to the address. The difference waveform is read into the buffer (91). Since the pitch cycle to be synthesized is 45 and −1 is finally set in the pitch cycle change value register (96), the pitch control circuit (97) cuts off the residual waveform by one.

Since the present invention is configured as described above, when the pitch cycle of the residual waveform is changed, in the case of the embodiment, sufficient pitch control can be performed only by truncating the pitch cycle to zero at most. .

(G) Effect of the Invention Since the residual driving type speech synthesizer of the present invention stores a plurality of residuals having different pitch periods for the same speech unit, the pitch according to the pitch pattern generated from the rule is used. When changing the pitch cycle of the residual waveform to the cycle, for example, the driving sound source is generated using the residual waveform of the pitch cycle closest to the pitch cycle to be synthesized, so that the amount of change in the pitch is greatly reduced. Thus, it is possible to avoid deterioration in sound quality due to a change in pitch of the residual waveform.

[Brief description of the drawings]

FIG. 1 is a block diagram of a general residual drive type rule synthesizing apparatus, FIG. 2 is an arrangement diagram of phoneme symbols, FIG. 3 is a pitch pattern pattern diagram, and FIG. 4 is a conventional residual drive type speech. FIG. 5 is a configuration diagram of a driving sound source generation unit in the synthesizing device, FIG. 5 is a configuration diagram of a driving sound source generation unit realizing the present invention, and FIG. 6 is an arrangement diagram of residual pitch period registers 1 to 6. (1) ... character string buffer, (2) ... phoneme symbol string generation unit, (3) ... phoneme symbol string buffer, (4) ... speech unit memory, 5 ... speech unit connection unit, ( 6) Voice parameter buffer, (7) Pitch pattern generation unit,
(8) ... residual waveform memory, (9) ... residual sound source generator, (10) ... driving sound source buffer, (11) ... synthesis filter, (12) ... DA converter, (13) …… Speaker, (9
1) Residual waveform buffer, (92) Residual pitch cycle register, (921) to (926) Residual pitch cycle registers 1 to 6, (93) Generated pitch cycle buffer, (94) )
…… Target pitch period register, (95) …… Differentiator, (9
6) Pitch cycle change value register (97) Pitch control circuit (98) Residual address register (981)
~ (986) ... Residual address registers 1 to 6, (99) ...
... Comparator.

Continuation of front page (56) References JP-A-2-66599 (JP, A) IEICE Technical Report [Sound] Vol. 89, No. 388, SP89-112, "Generation of Driving Sound Source and Pitch Control Method Based on Minimum Criterion for Spectral Distortion", p. 19-26 (issued on January 26, 1990) Proceedings of ASJ Autumn Meeting, 1981 I-1-2-16 "Speech synthesis using phoneme chain and residual waveform" p. 67-68 (October 6, 1981) (58) Fields investigated (Int. Cl. ⁷ , DB Name) G10L 11/00-13/08 G10L 19/00-21/06 JICST file (JOIS)

Claims (2)

(57) [Claims] 1. A first memory for storing a speech unit which is a sequence of speech parameters required for speech synthesis, a second memory for storing a residual corresponding to each speech unit, and a content to be uttered. Phoneme symbol string generation unit that generates a symbol string indicating a natural speech unit,
A pitch pattern that determines the change in pitch period from the utterance content
A speech unit connection unit for sequentially connecting necessary speech units based on the symbol string generated by the phoneme symbol string generation unit, and a speech parameter included in the connected speech unit as a coefficient. A voice synthesis filter for synthesizing voice, a residual corresponding to a voice segment is used as a driving sound source, and a pitch cycle of the residual is changed according to a pitch cycle at each time point determined by the pitch pattern generation unit. In a residual driving type speech synthesizer comprising a driving sound source generating unit for inputting to the synthesis filter, a residual for selecting a specific residual from a plurality of residuals stored in the second memory. A selection circuit for storing a plurality of residuals having different pitch periods in a second memory in correspondence with the speech units stored in the first memory;
According to the pitch period at each point determined by the
The residual selection circuit selects an appropriate residual of the pitch period from the second memory, and the driving sound source generation unit changes a required pitch period for the selected residual. Driven speech synthesizer. 2. The residual selecting circuit according to claim 1, wherein the residual of the pitch cycle closest to the pitch cycle at each time point determined by the pitch pattern generating section is selected from the residuals in the second memory. 2. The apparatus of claim 1, wherein the apparatus is selected.