Compressing & using a concatenative speech database in text-to-speech systems

• US 20020143543A1
• Filed: 03/30/2001
• Published: 10/03/2002
• Est. Priority Date: 03/30/2001
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Sirivara, Sudheer
• Granted Patent: US 7,035,794 B2
• Time in Patent Office: Days
• US Class Current: 704/260
• CPC Class Codes: G10L 13/06 Elementary speech units used in speech synthesisers; Concatenation rules
• CPC Class Codes: G10L 19/06 Determination or coding of the spectral characteristics, e.g. of the short-term prediction coefficients

Abstract

A method and apparatus are provided for compressing and using a concatenative speech database in TTS systems to improve the quality of speech output generated by handheld TTS systems by allowing synthesis to occur on the client. According to one embodiment of the present invention, a G.723 encoder receives diphone waveforms, and compresses them into diphone residuals. While compressing the diphone waveforms, the encoder generates Linear Predictive Coding (LPC) coefficients. The diphone residuals, and the encoder-generated LPC coefficients are then stored in encoder-generated compressed packet.

Claims

1. A method comprising:

receiving diphone waveforms;

compressing the diphone waveforms into diphone residuals, wherein the compressing is performed using an encoder;

generating linear predictive coding (LPC) coefficients, wherein the LPC coefficients are generated by the encoder; and

storing the diphone residuals and the encoder-generated LPC coefficients in a compressed packet, wherein the compressed packet is generated by the encoder.

2. The method of claim 1 further comprising:

a waveform synthesizer requesting diphone residuals;

locating the requested diphone residuals in the compressed packet;

extracting the located diphone residuals from the compressed packet;

decompressing the extracted diphone residuals, wherein the decompressing is performed using a decoder; and

supplying the diphone residuals to the waveform synthesizer.

3. The method of claim 2 further comprising supplying the encoder-generated LPC coefficients to the waveform synthesizer.

4. The method of claim 2 further comprising supplying pitch marks to the waveform synthesizer.

5. The method of claim 2 further comprising the waveform synthesizer producing speech output.

6. The method of claim 1, wherein the encoder is a G.723 encoder.

7. The method of claim 1, wherein the decoder is a modified G.723 decoder.

8. A method comprising:

receiving diphone waveforms;

compressing the diphone waveforms into diphone residuals, wherein the compressing is performed using an encoder;

generating linear predictive coding (LPC) coefficients, wherein the LPC coefficients are generated by the encoder;

storing the diphone residuals and the coder-generated LPC coefficients in a compressed packet, wherein the compressed packet is generated by the encoder;

a waveform synthesizer requesting the diphone residuals;

locating the requested diphone residuals in the compressed packet;

extracting the located diphone residuals from the compressed packet; and

decompressing the extracted diphone residuals, wherein the decompressing is performed using a decoder; and

supplying the diphone residuals and the encoder-generated LPC coefficients to the waveform synthesizer.

9. The method of claim 8 further comprising supplying pitch marks to the waveform synthesizer.

10. The method of claim 8, wherein the encoder is a G.723 encoder.

11. The method of claim 8, wherein the decoder is a G.723 decoder.

13. The system of claim 12, wherein the text-to-speech system comprising:

a text analysis module for processing a text into forms of linguistic representations;

a linguistic and prosodic analysis module for analyzing the forms of linguistic representations corresponding to their assigned language system; and

a waveform synthesizer for producing a speech output.

14. The system of claim 12, wherein the concatenative speech database comprising:

diphone waveforms;

LPC coefficients; and

pitch marks.

15. The system of claim 14, wherein the diphone waveforms are compressed to diphone residuals.

16. The system of claim 12, wherein the coder is a G.723 coder.

17. The system of claim 16, wherein the G.723 coder comprises:

a G.723 encoder for compressing the concatenative speech database; and

a G.723 decoder for decompressing the concatenative speech database.

19. The method of claim 18, wherein the compressed packets comprising diphone residuals and audio encoder-generated LPC coefficients.

12. A system for compressing and using concatenative speech databases in text-to-speech systems comprising:

a text-to-speech system;

a concatenative speech database; and

a coder.

18. A method of producing a compressed concatenative diphone database comprising:

compressing diphone waveforms and generating linear predictive coding (LPC) coefficients by applying an audio encoder to the diphone waveforms; and

storing compressed packets produced by the audio encoder and uncompressed pitch mark values as a compressed concatenative diphone database.

20. The method for a handheld device with a text-to-speech system using a compressed concatenative diphone database comprising:

compressing diphone waveforms into diphone residuals and generating linear predictive coding (LPC) coefficients by applying an audio encoder to the diphone waveforms;

storing compressed packets produced by the audio encoder and uncompressed pitch mark values as a compressed concatenative diphone database;

decompressing the compressed concatenative diphone database by applying an audio decoder to the diphone residuals and the LPC coefficients; and

synthesizing the decompressed concatenative diphone database including the uncompressed pitch mark values to produce an output by applying a waveform synthesizer.

21. The method of claim 20 further comprising the handheld device downloading a customizable speech database.

22. The method of claim 20, wherein the synthesizing is client-based.

24. The concatenative speech database structure of claim 23, wherein the diphone waveforms are reduced to diphone residuals after compression.

25. The concatenative speech database structure of claim 23, wherein the difference equation is a linear predictor expressing each new sample of a signal as a linear combination of previous samples.

26. The concatenative speech database structure of claim 23, wherein the formants are the resonance characterizing vocal tract.

27. The concatenative speech database structure of claim 23, wherein the pitch mark values correspond to changes in fundamental frequency.

23. A concatenative speech database structure comprising:

diphone waveforms indicating smallest units of speech for efficient text-to-speech conversion that are derived from phonemes;

linear predictive coefficients of a difference equation for characterizing formants; and

pitch mark values marking positions in an utterance indicating varying pitch.