Celp-Based speech coding for fine grain scalability by altering sub-frame pitch-pulse

• US 6,996,522 B2
• Filed: 09/13/2001
• Issued: 02/07/2006
• Est. Priority Date: 03/13/2001
• Status: Active Grant

• Current Assignee: Industrial Technology Research Institute
• Original Assignee: Industrial Technology Research Institute
• Inventors: Chen, Fang-Chu
• Primary Examiner(s): Chawan, Vijay B.
• Application Number:
  US09/950,633
• Publication Number:
  US 20020133335A1
• Time in Patent Office: 1,608 Days
• Field of Search: 704/219, 704/264, 704/238, 704/270, 704/267, 704/500, 704/220, 704/222, 704/200.1, 704/203, 704/207, 704/212, 704/223, 704/236, 704/229, 704/208, 704/214, 382/238
• US Class Current: 704/219
• CPC Class Codes: G10L 19/10 the excitation function being a multipulse excitation

Abstract

Methods and systems for providing a CELP-based speech coding with fine grain scalability include a parameter encoder that generates a basic bit-stream from LPC coefficients for a frame, pitch-related information for all the sub-frames obtained by searching an adaptive codebook, and first pulse-related information for even sub-frames obtained by searching a fixed codebook. The parameter encoder also generates enhancement bits, which are preceded by the basic bit-stream, from second pulse-related information for odd sub-frames. The quality of synthesized speech is improved on a basis of one additional odd sub-frame pulse, as more of the second pulse-related information in the enhancement bits is received by a decoder.

Claims

1. A method of encoding a speech signal in a code excited linear prediction (CELP)-based speech processing system that includes an adaptive codebook and a fixed codebook, wherein the speech signal is divided into frames and each frame is further divided into sequential sub-frames, the method comprising:

generating linear prediction coding (LPC) coefficients for a frame;

generating pitch-related information by using the adaptive codebook, for the sequential sub-frames of the frame;

generating fixed-code pulse information by using the fixed codebook, for a plurality of selected sub-frames of the frame;

generating a first bit-stream corresponding to the frame for the LPC coefficients, the pitch-related information, and the fixed-code pulse information for the plurality of selected sub-frames;

generating fixed-code pulse information by using the fixed codebook, for unselected sub-frames; and

separately generating a second bit-stream corresponding to speech enhancement of the frame from the fixed-code pulse information for the unselected sub-frames.

2. The method of claim 1, wherein the first bit-stream provides a minimum quality when synthesized into speech, and the second bit-stream provides improved quality of the synthesized speech.

3. The method of claim 2,

wherein the selected sub-frames are even sub-frames of the frame, and the unselected sub-frames are odd sub-frames of the frame.

4. The method of claim 1, further comprising placing the second bit-stream after the first bit-stream.

5. The method of claim 4, wherein the generating of fixed-code pulse information for the unselected sub-frames includes generating information for a plurality of pulses, and in the second bit-stream, placing all information for one pulse before information of another pulse.

6. The method of claim 1, further comprising:

using the pulse-related information in addition to the pitch-related information for a selected sub-frame to generate pitch-related information and fixed-code pulse information for a succeeding sub-frame; and

using the pitch-related information without the pulse-related information for an unselected sub-frame to generate pitch-related information and fixed-code pulse information for a succeeding sub-frame.

7. The method of claim 1, further comprising:

searching the adaptive codebook and the fixed codebook to minimize a difference between a synthesized speech and a target signal to generate the pitch-related information and the fixed-code pulse information; and

linearly attenuating a magnitude of samples in the target signal for an unselected sub-frame, the number of samples corresponding to the order of an LP-synthesis filter.

8. A method of synthesizing speech in a code excited linear prediction (CELP)-based speech processing system that includes an adaptive codebook and a fixed codebook, wherein a speech signal is divided into frames and each frame is further divided into sub-frames, the method comprising:

receiving a basic bit-stream which includes

linear prediction coding (LPC) coefficients for a frame,

pitch-related information for all sub-frames of the frame, and

first pulse-related information for a plurality of selected sub-frames of the frame;

receiving enhancement bits which include second pulse-related information for unselected sub-frames of the frame;

generating an excitation

by referring to the adaptive codebook

based on the pitch-related information included in the basic bit-stream; and

by referring to the fixed codebook

based on the first pulse-related information included in the basic bit-stream;

generating an excitation

by referring to the adaptive codebook

based on the pitch-related information included in the basic bit-stream and

by referring to the fixed codebook

based on the part or the whole of the second pulse-related information included in the enhancement bits; and

outputting synthesized speech according to the excitations and the LPC coefficients.

9. The method of claim 8, wherein the plurality of selected sub-frames are even sub-frames of the frame, and the unselected sub-frames of the frame.

10. The method of claim 8, wherein the second pulse-related information includes information for a plurality of pulses, and quality of the synthesized speech is improved each time information for one pulse is added to the enhancement bits received.

11. The method of claim 8, further comprising:

feeding back the excitation generated from the first pulse-related information in addition to the pitch-related information, for generating an excitation for a succeeding sub-frame; and

feeding back another excitation generated from the pitch-related information without the second pulse-related information, for generating an excitation for a succeeding sub-frame.

12. A speech processing system based on code excited linear prediction (CELP) for encoding a speech signal, wherein the speech signal is divided into frames and each frame is further divided into sub-frames, the system comprising:

a generator of linear prediction coding (LPC) coefficients for a frame;

a first portion including an adaptive codebook for generating pitch-related information for each sub-frame of the frame;

a second portion including a fixed codebook for generating fixed-code pulse information for each sub-frame of the frame, the pulse-related information including first fixed-code pulse information for a first kind of sub-frame and second fixed-code pulse information for a second kind of sub-frame; and

a parameter encoder for generating a basic bit-stream from the LPC coefficients, the pitch-related information, and the first fixed-code pulse information, and for generating enhancement bits from the second pulse-related information.

13. The system according to claim 12, further comprising

a transmitter for transmitting the basic bit-stream and a part of the enhancement bits onto a channel, the part being determined based on traffic of the channel.

14. The system according to claim 12, wherein the pitch-related information is reused in the first portion for a succeeding sub-frame, the first fixed-code pulse information being reused in addition to the pitch-related information, the second fixed-code pulse information not being reused.

15. The system according to claim 12, further comprising:

an analysis-by-synthesis loop including a synthesizer for searching the adaptive codebook and the fixed codebook to minimize a difference between a synthesized speech and a target signal; and

a target signal processor for linearly attenuating a magnitude of samples in the target signal provided to the analysis-by-synthesis loop for the second kind of sub-frame, the number of samples corresponding to the order of an LP-synthesis filter.

16. A speech processing system based on code excited linear prediction (CELP) for synthesizing speech, wherein a speech signal is divided into frames and each frame is further divided into sub-frames, the system comprising:

a parameter decoder for extracting

linear prediction coding (LPC) coefficients for a frame,

pitch-related information for all the sub-frames of the frame, and

first pulse-related information for a plurality of selected sub-frames of the frame,

from a basic bit-stream received, and

for extracting a second pulse-related information for unselected sub-frames of the frame from enhancement bits received;

a first portion including an adaptive codebook for generating an excitation based on the pitch-related information;

a second portion including a fixed codebook for generating an excitation

based on the first pulse-related information or

based on the second pulse-related information; and

a synthesizer for outputting synthesized speech according to the excitations and the LPC coefficients.

17. The system according to claim 16, wherein the second pulse-related information includes information for a plurality of pulses, and the parameter decoder extracts, from the enhancement bits received, information for each pulse and provides the second portion with the information for each pulse.

18. The system according to claim 16, wherein:

the excitation generated from the pitch-related information is fed back to the first portion for a succeeding sub-frame,

the excitation generated from the first pulse-related information being fed back in addition to the excitation from the pitch-related information, and

the excitation generated from the second pulse-related information not being fed back.