Excitation vector generator, speech coder and speech decoder

• US 6,345,247 B1
• Filed: 11/15/1999
• Issued: 02/05/2002
• Est. Priority Date: 11/07/1996
• Status: Expired due to Term

• Current Assignee: Godo Kaisha IP Bridge 1 (IP Bridge, Inc.)
• Original Assignee: Matsushita Electric Industrial Company Limited (Panasonic Holdings Corporation)
• Inventors: Morii, Toshiyuki, Ehara, Hiroyuki, Yasunaga, Kazutoshi
• Primary Examiner(s): Tsang, Fan
• Assistant Examiner(s): Opsasnick, Michael N.
• Application Number:
  US09/440,199
• Time in Patent Office: 813 Days
• Field of Search: 704/219, 704/220
• US Class Current: 704/219
• CPC Class Codes: G10L 19/12 the excitation function being a code excitation, e.g. in code excited linear prediction [CELP] vocoders
• CPC Class Codes: G10L 19/135 Vector sum excited linear prediction [VSELP]
• CPC Class Codes: G10L 2019/0007 Codebook element generation
• CPC Class Codes: G10L 2019/0013 Codebook search algorithms

Abstract

A random code vector reading section and a random codebook of a conventional CELP type speech coder/decoder are respectively replaced with an oscillator for outputting different vector streams in accordance with values of input seeds, and a seed storage section for storing a plurality of seeds. This makes it unnecessary to store fixed vectors as they are in a fixed codebook (ROM), thereby considerably reducing the memory capacity.

Claims

1. A code excited linear prediction speech decoder, comprising:

an adaptive codebook capable of generating an adaptive code vector;

a random codebook capable of generating a random code vector;

a synthesis filter that receives a signal based on said adaptive code vector and said random code vector, and is capable of performing linear prediction coefficient synthesis on said signal; and

said random codebook comprising;

an input vector providing system capable of providing an input vector comprising at least one pulse, each pulse having a predetermined position with a respective polarity;

a fixed waveform storage system capable of storing one or more fixed waveforms; and

an arranging system capable of arranging said fixed waveforms in accordance with the position and the polarity of said at least one pulse of said input vector.

2. The code excited linear prediction speech decoder of claim 1, further comprising:

a determiner that determines whether input speech is at least one of voiced and unvoiced; and

wherein said random codebook generates a different random code vector based on a result of said determiner.

3. The code excited linear prediction speech decoder of claim 2, wherein said determiner determines whether input speech is at least one of voiced and unvoiced based on a value of a quantized pitch gain.

4. The code excited linear prediction speech decoder of claim 2, wherein said determiner determines whether input speech is at least one of voiced and unvoiced based on a value of an adaptive codebook gain.

5. The code excited linear prediction speech decoder of claim 2, wherein said determiner determines whether input speech is at least one of voiced and unvoiced based on a value of a quantized long term prediction gain.

6. A code excited linear prediction speech decoder, comprising:

an adaptive codebook capable of generating an adaptive code vector;

a random codebook capable of generating a random code vector;

a synthesis filter that receives a signal based on said adaptive code vector and said random code vector, and is capable of performing linear prediction coefficient synthesis on said signal; and

said random codebook comprising;

an input vector providing system capable of providing an input vector comprising at least one pulse, each pulse having a predetermined position with a respective polarity;

a fixed waveform storage system capable of storing one or more fixed waveforms; and

a shifting system capable of shifting said fixed waveforms in accordance with the position and the polarity of said at least one pulse of said input vector.

7. The code excited linear prediction speech decoder of claim 6, further comprising:

a determiner that determines whether input speech is at least one of voiced and unvoiced; and

wherein said random codebook generates a different random code vector based on a result of said determiner.

8. The code excited linear prediction speech decoder of claim 7, wherein said determiner determines whether input speech is at least one of voiced and unvoiced based on a value of a quantized pitch gain.

9. The code excited linear prediction speech decoder of claim 7, wherein said determiner determines whether input speech is at least one of voiced and unvoiced based on a value of an adaptive codebook gain.

10. The code excited linear prediction speech decoder of claim 7, wherein said determiner determines whether input speech is at least one of voiced and unvoiced based on a value of a quantized long term prediction gain.

11. A code excited linear prediction speech decoder, comprising:

an adaptive codebook capable of generating an adaptive code vector;

a random codebook capable of generating a random code vector;

a synthesis filter that receives a signal based on said adaptive code vector and said random code vector, and is capable of performing linear prediction coefficient synthesis on said signal; and

said random codebook comprising;

an input vector providing system capable of providing an input vector comprising at least one pulse, each pulse having a predetermined position and a respective polarity;

a fixed waveform storage system capable of storing one or more fixed waveforms; and

a convolution system capable of convoluting said fixed waveforms with said input vector.

12. The code excited linear prediction speech decoder of claim 11, further comprising:

a determiner that determines whether input speech is at least one of voiced and unvoiced; and

wherein said random codebook generates a different random code vector based on a result of said determiner.

13. The code excited linear prediction speech decoder of claim 12, wherein said determiner determines whether input speech is at least one of voiced and unvoiced based on a value of a quantized pitch gain.

14. The code excited linear prediction speech decoder of claim 12, wherein said determiner determines whether input speech is at least one of voiced and unvoiced based on a value of an adaptive codebook gain.

15. The code excited linear prediction speech decoder of claim 12, wherein said determiner determines whether input speech is at least one of voiced and unvoiced based on a value of a quantized long term prediction gain.

19. The code excited linear prediction speech decoder of claim 12, wherein said determiner determines whether input speech is at least one of voiced and unvoiced based on a value of a quantized long term prediction gain.

16. A code excited linear prediction speech decoder, comprising:

a random codebook capable of generating different random code vectors;

said random codebook including a determiner that determines whether input speech is at least one of voiced and unvoiced; and

wherein said random codebook generates said different random code vectors based on a result of said determiner, a fixed waveform associated with said random codebook being arranged in accordance with a position and a polarity of a pulse of an input vector associated with said random codebook.

17. The code excited linear prediction speech decoder of claim 16, wherein said determiner determines whether input speech is at least one of voiced and unvoiced based on a value of a quantized pitch gain.

18. The code excited linear prediction speech decoder of claim 16, wherein said determiner determines whether input speech is at least one of voiced and unvoiced based on a value of an adaptive codebook gain.

20. A method of producing synthesized speech, said method comprising:

providing an adaptive code vector;

determining whether input speech is at least one of voiced and unvoiced based on a gain;

providing an unvoiced random code vector if said input speech is unvoiced, comprising;

providing at least one input vector comprising at least one pulse, each pulse having a position and a polarity;

storing at least one fixed waveform; and

modifying said at least one fixed waveform in accordance with the polarity and the position of said at least one pulse of said at least one input vector;

providing a voiced random code vector if said speech is voiced, said voiced random code vector including a plurality of pulses; and

performing linear prediction coefficient synthesis on a signal based on said adaptive code vector and one of said voiced random code vector and said unvoiced random code vector.

21. The method of claim 20, wherein said gain is one of a quantized pitch gain, an adaptive codebook gain, and a quantized long term prediction gain.

22. A method for producing synthesized speech, comprising:

providing an adaptive code vector;

determining whether input speech is at least one of voiced and unvoiced based on a gain;

providing an unvoiced random code vector if said input speech is unvoiced, comprising;

providing a plurality of input vectors, each input vector comprising a plurality of pulses, each of said pulses having a position and a polarity;

providing a plurality of fixed waveforms;

modifying said plurality of fixed waveforms in accordance with a position and a polarity of the plurality of the pulses of said plurality of input vectors, respectively; and

adding said modified fixed waveforms;

providing a voiced random code vector if said speech is voiced, said voiced random code vector including a plurality of pulses; and

performing linear prediction coefficient synthesis on a signal based on said adaptive code vector and one of said voiced random code vector and said unvoiced random code vector.

23. The method of claim 22, wherein said gain is one of a quantized pitch gain, an adaptive codebook gain, and a quantized long term prediction gain.

24. A system used in the production of synthesized speech, comprising:

an adaptive code vector;

a random code vector generator capable of generating a random code vector, comprising;

an input vector providing system capable of providing a plurality of input vectors, each input vector comprising a plurality of pulses, each of said pulses having a position and a polarity;

a fixed waveform storage system capable of storing a plurality of fixed waveforms; and

a modification system capable of modifying each of said plurality of fixed waveforms in accordance with a polarity and a position of said plurality of pulses of a different one of said plurality of input vectors, respectively;

an adder that adds each of said plurality of fixed waveforms as modified;

a determiner that determines whether input speech is at least one of voiced and unvoiced based on a gain; and

a synthesis filter capable of performing linear prediction coefficient synthesis on a signal based on at least said adaptive code vector and said random code vector,

wherein, when said input speech is voiced, said random code vector comprises a plurality of pulses, and when said input speech is unvoiced, said random code vector comprises said plurality of fixed waveforms as modified by said modification system and added by said adder.

25. The method of claim 24, wherein said gain is one of a quantized pitch gain, an adaptive codebook gain, and a quantized long term prediction gain.

26. A system used in the production of synthesized speech, comprising:

a plurality of input vectors, each comprising a plurality of pulses, each of said pulses having a position and a polarity;

at least first and second sets of fixed waveforms;

a switch movable between a plurality of positions, each position being responsive to one of plurality of speech conditions based on a gain, said conditions including at least voiced and unvoiced input speech; and

a modification system;

wherein, when said switch is in a first position, an output of said system is a result from on a modification of each fixed waveform of said first set of fixed waveforms in accordance with the polarity and the position of said plurality of pulses of a different one of said plurality of input vectors, as modified by said modification system; and

wherein, when said switch is in a second position, an output of said system is based at least partially upon said second set of fixed waveforms.

27. The method of claim 26, wherein said gain is one of a quantized pitch gain, an adaptive codebook gain, and a quantized long term prediction gain.

28. A method of providing an excitation vector used in the production of synthesized speech, said method comprising:

providing a plurality of input vectors, each input vector comprising a plurality of pulses, each of said pulses having a position and a polarity;

providing first and second sets of fixed waveforms;

determining whether input speech is at least one of voiced and unvoiced based on a gain;

outputting, if said input speech is unvoiced, a signal resulting from modifying said each fixed waveform of said first set of fixed waveforms in accordance with the polarity and the position of the pulses of a different one of said plurality of input vectors, respectively; and

outputting, if said input speech is voiced, a signal based at least partially on said second set of fixed waveforms.

29. The method of claim 28, wherein said gain is one of a quantized pitch gain, an adaptive codebook gain, and a quantized long term prediction gain.

30. A method of producing synthesized speech, said method comprising:

providing an adaptive code vector;

determining whether input speech is at least one of voiced and unvoiced based on a gain;

providing an unvoiced random code vector if said input speech is unvoiced, comprising;

providing at least one input vector comprising at least one pulse, each pulse having a position and a polarity;

storing at least one fixed waveform; and

convoluting said at least one fixed waveform with said at least one input vector;

providing a voiced random code vector if said speech is voiced, said voiced random code vector including a plurality of pulses; and

performing linear prediction coefficient synthesis on a signal based on said adaptive code vector and one of said voiced random code vector and said unvoiced random code vector.

31. The method of claim 30, wherein said gain is one of a quantized pitch gain, an adaptive codebook gain, and a quantized long term prediction gain.