Voice recognition with parallel gender and age normalization

• US 20070198261A1
• Filed: 02/21/2006
• Published: 08/23/2007
• Est. Priority Date: 02/21/2006
• Status: Active Grant

• Current Assignee: Sony Interactive Entertainment Inc. (Sony Group Corp.)
• Original Assignee: Sony Computer Entertainment Incorporated (Sony Group Corp.)
• Inventors: Chen, Ruxin
• Granted Patent: US 8,010,358 B2
• Time in Patent Office: Days
• US Class Current: 704/240
• CPC Class Codes: G10L 15/065 Adaptation
• CPC Class Codes: G10L 15/32 Multiple recognisers used in sequence or in parallel; Score combination systems therefor, e.g. voting systems

Abstract

Methods and apparatus for voice recognition are disclosed. A voice signal is obtained and two or more voice recognition analyses are performed on the voice signal. Each voice recognition analysis uses a filter bank defined by a different maximum frequency and a different minimum frequency and wherein each voice recognition analysis produces a recognition probability r_i of recognition of one or more speech units, whereby there are two or more recognition probabilities r_i. The maximum frequency and the minimum frequency may be adjusted every time speech is windowed and analyzed. A final recognition probability P_ƒ is determined based on the two or more recognition probabilities r_i.

Claims

1. A method for voice recognition, the method comprising:

obtaining a voice signal;

performing two or more voice recognition analyses on the voice signal, wherein each voice recognition analysis uses a filter bank defined by a different maximum frequency and a different minimum frequency and wherein each voice recognition analysis produces a recognition probability r_i of recognition of one or more speech units, whereby there are two or more recognition probabilities r_i; and

determining a final recognition probability P_ƒ

based on the two or more recognition probabilities r_i.

2. The method of claim 1 wherein determining the final recognition probability includes selecting a maximum probability of the two or more recognition probabilities as the final recognition probability P_ƒ

.

3. The method of claim 1 wherein obtaining a voice signal includes over-sampling the voice signal at a sampling frequency that is greater than a working feature analysis frequency.

4. The method of claim 3 wherein the sampling frequency is greater than a training time speech sampling rate.

5. The method of claim 1 wherein performing the two or more voice recognition analyses on the voice signal includes performing the two or more voice recognition analyses in parallel.

6. The method of claim 1 performing the two or more voice recognition analyses includes;

performing a first voice recognition analysis based on a first maximum frequency ƒ

_max and a first minimum frequency ƒ

_min;

wherein ƒ

_min and ƒ

_max are adjusted dynamically during the first voice recognition; and

performing one or more additional voice recognition analyses is based on a different maximum frequency given by ƒ

_max±

Δ

ƒ

_max and a different minimum frequency given by ƒ

_min±

Δ

ƒ

_min, where Δ

ƒ

_max<

ƒ

_max and Δ

ƒ

_min<

ƒ

_min.

7. The method of claim 6 wherein ƒ

_min and ƒ

_max are adjusted for each time window of the first voice recognition.

8. The method of claim 6 wherein different voice recognition analyses use the same values of ƒ

_min and ƒ

_max.

9. The method of claim 6 wherein Δ

ƒ

_min is between about 1% and about 5% of ƒ

_min.

10. The method of claim 6 wherein Δ

ƒ

_max is between about 1% and about 5% of ƒ

_max.

11. The method of claim 6 wherein performing one or more additional voice recognition analyses includes:

performing a second voice recognition analysis based on a maximum frequency given by ƒ

_max−

Δ

ƒ

_max and a minimum frequency given by ƒ

_min−

Δ

ƒ

_min;

performing a third voice recognition analyses is based on a maximum frequency given by ƒ

_max+Δ

ƒ

_max and a minimum frequency given by ƒ

_min−

Δ

ƒ

_min;

performing a fourth voice recognition analysis based on a maximum frequency given by ƒ

_max−

Δ

ƒ

_max and a minimum frequency given by ƒ

_min+Δ

ƒ

_min; and

performing a fifth voice recognition analysis based on a maximum frequency given by ƒ

_max+Δ

ƒ

_max and a minimum frequency given by ƒ

_min+Δ

ƒ

_min.

12. The method of claim 11 wherein the first, second, third, fourth and fifth voice recognition analyses are performed in parallel.

13. The method of claim 1 wherein performing two or more voice recognition analyses on the voice signal includes performing one or more gender-specific and/or age-specific voice recognition analyses.

14. The method of claim 1 wherein performing two or more voice recognition analyses on the voice signal includes performing two or more voice recognition analyses having different speaker category assumptions.

15. The method of claim 14 wherein determining the fmal recognition probability includes weighting the final recognition probability according to one or more speaker category probabilities associated with the different speaker category assumptions.

16. The method of claim 13 wherein the final recognition probability P_ƒ

is determined by;

$P_f=\underset{i}{\sum <br><br>}<br><br>\left(C_i\times <br><br>r_i\right),$

where C_i is a probability that a speaker is a member of the i^th category and r_i is a recognition probability resulting from a voice analysis that assumes the speaker is a member of the i^th category.

17. The method of claim 14 wherein performing two or more voice recognition analyses having different speaker category assumptions includes:

performing a first voice recognition based on an assumption that the voice signal is obtained from a male speaker;

performing a second voice recognition based on an assumption that the voice signal is obtained from a female speaker;

performing a third voice recognition based on an assumption that the voice signal is obtained from a child speaker.

18. The method of claim 1, further comprising determining a confidence measure based on the final recognition probability and the two or more recognition probabilities r_i.

19. The method of claim 18 wherein the confidence measure is a maximum of the two or more recognition probabilities r_i divided by a sum of the two or more recognition probabilities r_i.

20. The method of claim 1 wherein performing two or more voice recognition analyses on the voice signal includes adjusting the maximum frequency and the minimum every time speech is windowed and analyzed.

21. A voice recognition system, comprising:

an interface adapted to obtain a voice signal;

one or more processors coupled to the interface; and

a memory coupled to the interface and the processor, the memory having embodied therein a set of processor readable instructions for configured to implement a method for voice recognition, the processor readable instructions including;

one or more instructions that direct the one or more processors to perform two or more voice recognition analyses on the voice signal, wherein each voice recognition analysis uses a filter bank defined by a different maximum frequency and a different minimum frequency and wherein each voice recognition analysis produces a recognition probability r_i of recognition of one or more speech units, whereby there are two or more recognition probabilities r_i; and

an instruction to determine a final recognition probability P_ƒ

based on the two or more recognition probabilities.

22. The system of claim 21 wherein the one or more processors include two or more processor elements configured to perform parallel processing.

23. The system of claim 22 wherein the two or more processor units include a power processor element and one or more synergistic processor elements.

24. The system of claim 22 wherein the one or more instructions that direct the one or more processors to perform two or more voice recognition analyses on the voice signal are configured to perform the two or more voice recognition analyses in parallel.

25. The system of claim 21 wherein one or more instructions that direct the one or more processors to perform two or more voice recognition analyses adjust the maximum frequency and the minimum every time speech is windowed and analyzed.