JPS58117598A - Voice synthesizer - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a speech synthesis device, and particularly to a speech synthesis device using an analysis and synthesis method.

The analysis and synthesis method is a method in which information in the form of speech data sufficient to express the content of speech, such as speech synthesis parameters, is extracted in advance from speech analysis/K, and the speech is resynthesized using this information. This method includes APCM (adaptive PCM) and 5BC (subband coding) methods that aim to reproduce the input waveform, and methods that preserve the spectral properties of the input but do not care about the reproducibility of the waveform. as L
There are a PG (linear predictive coding) method, a PARCOR (partial autocorrelation coefficient) method, and the like. Conventional speech synthesis devices using these methods synthesize speech using speech data obtained with the same analysis density. Therefore, if the length of the audio sentence is long compared to the storage capacity of this audio data, and the audio data cannot be accommodated in the storage unit, the analysis density of the audio data is uniformly lowered and a large amount of audio data is stored. was designed to accommodate.

For example, if the total duration of the audio sentence as audio output is 40 seconds, the memory capacity of the speech synthesizer is 128 bits, and the analysis density of the audio data is 4.8 bits/second. In terms of memory capacity, Maku is 28 bits/
4.8 can only store audio data equivalent to approximately 26 seconds of bits. Therefore, in such cases, the memory analysis density is set to 2.4 bits/second to cope with the long duration of the audio sentence. In this case, about 52 seconds worth of ondo booters can be stored.

However, in this way, the analysis density of voice data is −<,1
If it is set too low, the sound quality will deteriorate and the audio text will become unclear (d+).

In other words, the analysis and synthesis method performs some processing on the natural speech waveform to obtain data with higher information density.For example, 1) AR, COR, and the The displayed audio waveform is sampled every predetermined analysis interval T1, and the audio feature parameters as audio data are stored in a storage memory.

Here, the analysis interval TI (10 mS) is the analysis density of 4.
If it corresponds to 8 bits/second, the audio data in the memory is synthesized into 48 bits/second audio data, and is output as an audio signal as shown in FIG. However, when the analysis density of audio data stored in memory is increased from 4.8 bits/second to 2.4 bits/second, the analysis interval T2 becomes 2 [a (2
0m S), and the sound 16 as shown in Fig.
output as a number.

If the analysis density of the audio data stored in the memory becomes uniformly low in this way, a faithful audio waveform cannot be obtained, and the audio sentence becomes unclear.

The present invention has been made in view of the above problems, and an object thereof is to provide a speech synthesis device that can prevent speech sentences from becoming uniformly unclear.

In order to achieve the above object, the present invention provides a memory for storing a single MI constituting a spoken sentence as speech data obtained by speech analysis for each word block, into word block groups divided into at least two groups depending on the analysis density. For a given audio sentence, speech synthesis is performed using audio data from word blocks with high analysis density for phrases that require high intelligibility, and for phrases that require low intelligibility. The present invention is characterized in that speech synthesis is performed using speech data of a group of word blocks with low analysis density.

Hereinafter, preferred embodiments of the present invention will be described based on the drawings.

FIG. 4 shows a preferred embodiment of the present invention, in which PkRC
A configuration diagram of a speech synthesis device to which the OR method is applied is shown. In the figure, the speech synthesizer includes a control section 2, a parallel interface circuit 4, a buffer relay 6, a speech synthesis section 8,
It has a memory 10 and an amplifier 12, and the amplifier 12 is connected to a speaker 14.

Memo 1710u The words constituting the audio sentence are stored as audio data obtained by audio analysis for each word block, and the audio data of word block groups divided into at least two groups depending on the analysis density is stored. There is.

That is, the analysis density is made to differ depending on whether words in the audio sentence require high intelligibility or not.

For example, [A train is coming to platform 1. It is dangerous, so please stand back to the white line at your feet and wait. ] In the case of audio sentences used for train approaching announcements, where the former sentence requires high intelligibility and the latter sentence has low intelligibility, , in the former sentence, [on platform 1], [the train], [is coming]. Analysis of each word block in ``Density of 48 bits/second audio data,'' each single wave of the latter sentence ``It's dangerous,'' ``To the white line at your feet,'' ``Step back,'' and ``Please wait.'' The analysis density of the block is stored as audio data in 14 bits/second.

The control unit 2 is composed of a microcomputer having a CPU, ROM, RAM, etc.
is supplied through the buffer relay 6 and the parallel interface circuit 4, the voice sentence corresponding to the command is edited by a combination of word blocks corresponding to the word blocks contained in the memory IO, and the word block is selected according to the editing. command. This command is supplied to the speech synthesis section 8 via the parallel interface circuit 4.

The speech synthesis section 8 is composed of a sound source section including a pulse train generator, a noise generator, a digital filter, a D/A converter, a parameter interpolation circuit, a parameter decoding circuit, an interface circuit, etc. Word blocks according to the selection command are selected from memory 0, and the audio data of each word block is synthesized into an audio signal according to the analysis density. In the voice synthesis section 8, the synthesized voice signal is amplified by the amplifier 12 and output as voice from the speaker 4.

Next, the operation when broadcasting the audio text of the train approach announcement described above will be explained based on the flowchart of FIG. 5.

First, in step 200, when the voice output command 100 is supplied to the control unit 2 via the buffer relay 6 and the parallel interface circuit 4i, the process moves to step 202, where the train is placed on platform 1. 〕and〔
This is a dangerous animal, so please stand back to the white line at our feet and wait.

] is edited, and since Tsumugi's broadcast text is a text that requires clarity, the analysis density of the audio data is set to 4.8 bits/second in step 203, and the control unit 2 Based on the selection command, [the train] will come [to platform 1]. ] The audio data for each word is 4.8
A voice is synthesized using an analysis density of bits/second, and the voice is broadcast from the speaker 14 as voice. Next step 204
Step 20, an initial setting process is performed to set the analysis density of the latter broadcast text to bits/second at t-2, 4.
Move on to 5. In step 205, based on the selection command from the control unit 2, [it's dangerous], [go to the white line at your feet], [go back], and [wait for me, please!]. ] Each word block is voice synthesized at an analysis density of 14 bits/second and broadcast from a speaker 14.

In this embodiment, it has been described that the analysis density is different for each sentence, but the same can be done in the case where the analysis density is made different for each word block in the sentence.

In this way, according to this embodiment, the memory capacity t can be saved by mixing audio data with different analysis densities in the memory, so more audio data can be stored than when audio data with the same analysis density is stored. can do. Therefore, costs can be reduced by reducing memory and memory mounting space can be used effectively.

As explained above, according to the present invention, even when a predetermined storage capacity cannot accommodate all the audio sentences, the audio data of the words that make up the audio sentences are divided into at least two groups depending on the analysis density and stored in the memory. This has the advantageous effect of reducing memory capacity and preventing speech sentences from becoming uniformly unclear.

[Brief explanation of the drawing]

Fig. 1 is a natural speech waveform diagram, Fig. 2 is a speech waveform diagram synthesized from the speech waveform shown in Fig. 1 at an analysis density of 4.8 bits/sec, and Fig. 3 is a diagram of the speech waveform shown in Fig. 1. A speech waveform diagram in which a speech waveform is synthesized into Z4 at an analysis density of bits/second, FIG. 4 is a block diagram showing an embodiment of the present invention, and FIG. 5 explains the operation of the device shown in FIG. 4. There is a flowchart for this. 2... Control unit, 4... Parallel interface circuit,
6...Buffer relay, 8...Speech synthesis section, 1o.
...Memo lJ%12...Amplifier, 14...Speaker. Figure 1 Figure 4

Claims (1)

[Claims] 1. A memory that stores the single @ that makes up a spoken sentence in word blocks using audio data obtained through audio analysis;
A control unit that edits an audio sentence according to a combination of word blocks corresponding to the word blocks included in the memory according to an audio output command, and instructs selection of word blocks according to the editing, and a control unit that instructs selection of word blocks according to the editing, and a speech synthesis unit that selects word blocks and synthesizes the speech data of each word block into a speech signal according to analysis density; A speech synthesis device characterized by having a group.