JP2000330591A - Voice synthesizer - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce a circuit scale and current consumption in a voice synthesizer using vector quantization as a method of compressing voice data for storing them in a storage means. SOLUTION: This voice synthesizer is provided with a 1st memory 4 which stores the data compressed by vector-quantizing the voice data on time axis as they are and outputs the compressed data stored in the addresses designated by a read sequencer 3, a control part 2 for controlling the addresses in the 1st memory designated by the read sequencer 3, and a means (a 2nd memory 5, a rate converting part 6, a D/A converter part 7, and a voice output part 8) for expanding the compressed data outputted from the 1st memory 4 and converting them into voice to output.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice synthesizing apparatus for outputting a voice corresponding to an input, which is mounted on, for example, a toy, an electronic book, an alarm, and the like.

[0002]

2. Description of the Related Art A voice synthesizing apparatus which stores voice data in a storage means, reads out the voice data from the storage means in response to an input, and outputs the data is usually used to reduce the storage capacity of the storage means. Is compressed and stored in the storage means, the voice data (compressed voice data) read from the storage means is expanded, and the voice data obtained by expansion is converted into voice and output. I have.

Here, vector quantization exists as a method for compressing data. In vector quantization, the hardware becomes large-scale at the time of compression, and the time required for processing is long. On the other hand, at the time of decompression, the code and bit pattern that the compressed data can take are used as hardware. Only a storage means for storing data (called "codebook") indicating the correspondence is required, and the time required for processing is short. Therefore, vector quantization can be said to be an optimal technique for a speech synthesizer such as an electronic book that performs only speech reproduction.

Japanese Patent Laid-Open No. 5-73100 discloses a speech synthesizer using vector quantization as a method for compressing speech data stored in a storage means. In this speech synthesizer, compressed data obtained by compressing speech data by separating parameters into power information as strength information and spectrum information as acoustic information and subjecting only the spectrum information to vector quantization. Is stored in the storage means.

As described above, by compressing the audio data by vector-quantizing only the spectral information of the audio data, it is necessary to expand the power information and the spectral information of the audio together rather than to perform vector quantization. Can be reduced in size.

[0006]

However, in the above speech synthesizer, compressed data obtained by compressing the speech data in the form of data on the frequency axis is stored in the storage means. Therefore, a circuit for converting the audio data obtained by decompressing the compressed data output from the storage means into data on the time axis is required, and the overall circuit size of the audio synthesizer is large and the current consumption is large. There was a problem.

Accordingly, the present invention provides a speech synthesizer that uses vector quantization as a method for compressing speech data stored in a storage means, and that realizes a reduced circuit scale and reduced current consumption. The purpose is to do.

[0008]

In order to achieve the above object, according to the present invention, in a speech synthesizing apparatus for outputting a speech corresponding to an input, a speech data which is a digital electric signal obtained by converting the speech is converted. Storage means for storing compressed data obtained by compression by vector quantization as it is on the time axis; reading means for reading compressed data from the recording means in response to input to the speech synthesizer; A decompressing means for decompressing the compressed data read by the means, and an output means for converting data obtained by decompressing by the decompressing means into sound and outputting the sound.

With this configuration, the audio data obtained by decompression is data on the time axis, and there is no need to convert data on the frequency axis to data on the time axis.

[0010]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of a speech synthesizer according to an embodiment of the present invention. In the figure,
1 is an input unit, 2 is a control unit, 3 is a read sequencer, 4
Is the first memory, 5 is the second memory, 6 is the rate converter, 7
Is a D / A converter, and 8 is a sound output unit.

The input unit 1 receives an external input.
The control unit 2 controls the read sequencer 3 so that predetermined addresses in the first memory 4 according to the input received by the input unit 1 are specified by the read sequencer 3 in a predetermined order. The read sequencer 3 specifies an address in the first memory 4 under the control of the control unit 2.

The data stored in the first memory 4 will be described. Voice data (8 bits), which is a digital electric signal obtained by converting voice to be output from the voice synthesizer, is converted into data on a time axis,
Compressed data obtained by replacing with a code assigned to a bit pattern having the highest similarity among the N 32-bit bit patterns prepared in advance is stored every four bytes. That is, the audio data is compressed by vector quantization as it is on the time axis, and the compressed data obtained by compression is stored in the first memory 4. Then, the first memory 4 outputs the compressed data stored at the address specified by the read sequencer 3.

As shown in an image diagram of FIG. 2, the second memory 5 stores the N 32-bit bit patterns b (0), b prepared in advance for performing vector quantization.
(1), b (2),..., B (N−1) and codes (ie, codes that can be taken by compressed data) in which audio data is replaced when compression is performed C (0), C (1) , C (2),
.., C (N−1) are stored. The second memory 5 stores the code C (k) (k = 0, 1, 2,...) Of the compressed data output from the first memory 4.
.., N−1) are output in parallel in a bit sequence of the bit pattern b (k). That is, the compressed data output from the first memory 4 is expanded into audio data by the second memory 5.

Here, for example, assuming that the number of types of bit patterns prepared in advance for performing vector quantization is 256, the size of the code book, that is,
The storage capacity of the second memory 5 is 256 × 4 = 1024
Bytes are required.

The rate converter 6 receives the 32-bit data output from the second memory 5, divides the input 32-bit data into four 8-bit data, and divides the data into four data. Are sequentially output in a predetermined order. That is, the audio data obtained by expanding the compressed data by the second memory 5 is 4-byte parallel data, but is converted into serial data through the rate converter 6.

The D / A converter 7 converts the audio data output from the rate converter 6 into an analog electric signal. The audio output unit 8 converts the analog electric signal obtained by the D / A conversion unit 7 into audio and outputs it to the outside.

With the above configuration, audio data obtained by expanding compressed data is data on the time axis, and there is no need to convert data on the frequency axis to data on the time axis. Therefore, a circuit for performing this conversion becomes unnecessary, thereby reducing the circuit scale and the current consumption.

If the number of sound sources is limited by, for example, a single sounding person, or the type of voice to be compressed is limited, the output voice can be set assuming the same codebook size. Sound quality can be improved.

[0019]

As described above, according to the speech synthesizing apparatus of the present invention, speech data is compressed by vector quantization as it is on the time axis, and the compressed data obtained by compression is stored in the storage means. Since the compressed data read from the storage means is decompressed and then converted into speech, the data is stored on the frequency axis, which is necessary in the conventional speech synthesizer using vector quantization. It is not necessary to convert the data into the data described above, whereby the circuit scale and the current consumption can be reduced.

[Brief description of the drawings]

FIG. 1 is a block diagram of a speech synthesizer according to an embodiment of the present invention.

FIG. 2 is an image diagram showing that a second memory in the speech synthesizer shown in FIG. 1 stores a code book.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Input part 2 Control part 3 Read-out sequencer 4 1st memory 5 2nd memory 6 Rate conversion part 7 D / A conversion part 8 Audio output part

Claims (1)

[Claims] 1. A speech synthesizer for outputting speech in accordance with an input, wherein compressed data obtained by vector-quantizing speech data obtained by converting speech into a digital electric signal without performing parameter separation is used. A first storage unit that stores the data, a reading unit that reads compressed data from the first storage unit in response to an input to the speech synthesizer, and a codebook that reads the compressed data read by the reading unit. A speech synthesizer comprising: a second storage unit for decompressing; and an output unit for converting data obtained by decompression by the second storage unit into speech and outputting the speech.