Weighted spectral distance calculator

• US 6,983,245 B1
• Filed: 06/07/2000
• Issued: 01/03/2006
• Est. Priority Date: 06/07/1999
• Status: Expired due to Fees

• Current Assignee: Telefonaktiebolaget LM Ericsson
• Original Assignee: Telefonaktiebolaget LM Ericsson
• Inventors: Rasmusson, Jim, Jimenez Felstrom, Alberto
• Primary Examiner(s): CHAWAN, VIJAY B
• Application Number:
  US09/588,629
• Time in Patent Office: 2,036 Days
• Field of Search: 704/226, 704/233, 704/238, 704/246
• US Class Current: 704/238
• CPC Class Codes: G10L 15/10 using distance or distortion measures between unknown speech and reference templates

Abstract

A spectral distance calculator includes a calculator for performing spectral distance calculations to compare a spectrum of an an input signal in the presence of a noise signal and a reference spectrum. A memory pre-stores a noise spectrum from the noise signal. A masking unit masks the spectral distance between the input spectrum and the reference spectrum with respect to the pre-stored noise spectrum.

Claims

1. A speech recognition system, comprising:

a spectral distance calculator including;

a calculator for performing a spectral distance calculation comparing an input spectrum of an input signal in the presence of a first known noise signal used to perform a function unrelated to speech recognition and a reference spectrum;

a memory for pre-storing one or more noise spectrums of one or more known noise signals including the first known noise signal; and

masking circuitry for masking the spectral distance between the input spectrum and the reference spectrum using the pre-stored noise spectrum of the first known noise signal, and

a selector for selecting a reference spectrum minimizing a spectral distance between the input spectrum and the reference spectrum.

2. A speech recognition system according to claim 1, wherein the calculator is configured to assign the spectral distance between the input spectrum and the reference spectrum a zero value for each frequency of the input spectra which is due to noise.

3. A speech recognition system according to claim 1, wherein the noise has a lower level than a level of the input spectrum.

4. A speech recognition system according to claim 1, wherein the spectral distance calculation includes calculating the following expression for spectral distance D_n:

$D_n=\underset{i}{\sum <br><br>}<br><br>A_i<br><br><br><br>R_n<br><br>\left(f_i\right)-S_n<br><br>\left(f_i\right)<br><br>.$

where R_n(f_i) is the reference spectrum, S_n(f_i) is the input signal spectrum, and A_i is equal to zero if a frequency f_i of the input signal is due to a known noise and A_i is unity if no noise is present at the frequency f_i.

5. A speech recognition system according to claim 1, wherein the spectral distance is the sum of the spectral distance calculations for a number of samples discerning the reference spectra from each other.

6. A mobile telephone including the speech recognition system according to claim 1, comprising:

call answering circuitry operatively connected to the speech recognition system and responsive to one or more speech answering commands each forming an input spectrum.

7. A mobile telephone according to claim 6, wherein the call answering circuitry is responsive to an accept call command for accepting a call.

8. A mobile telephone according to claim 6, wherein the call answering circuitry is responsive to a reject call command for rejecting a call.

9. A mobile telephone according to claim 6, wherein the call answering circuitry is responsive to a forward call command for forwarding a call.

10. The speech recognition system in claim 1, wherein the first known noise signal is a periodic signal with a repeating pattern used to indicate a message.

11. The speech recognition system in claim 1, wherein the first known noise signal is a ring signal used to indicate a message.

12. The speech recognition system in claim 1, wherein the first known noise signal is a melody or a buzzer signal used to indicate a message.

13. The speech recognition system in claim 1, wherein the first known noise signal is a signal output from a speaker.

14. The speech recognition system in claim 1, wherein the function unrelated to speech recognition is to drive a speaker.

15. A speech recognition method, comprising:

comparing an input spectrum of an input signal in the presence of a first known noise signal used to perform a function unrelated to speech recognition and a reference spectrum to obtain a spectral distance;

pre-storing one or more noise spectrums of one or more known noise signals including the first known noise signal;

masking the spectral distance between the input spectrum and the reference spectrum using the pre-stored noise spectrum of the first known noise signal; and

selecting a reference spectrum minimizing a spectral distance between the input spectrum and the reference spectrum.

16. A speech recognition method according to claim 15, further comprising:

assigning the spectral distance between the input spectrum and the reference spectrum a zero value for each frequency of the input spectra which is due to noise.

17. A speech recognition method according to claim 15, wherein the noise has a lower level than a level of the input spectrum.

18. A speech recognition method according to claim 15, further comprising:

calculating the following expression for spectral distance D_n;

$D_n=\underset{i}{\sum <br><br>}<br><br>A_i<br><br><br><br>R_n<br><br>\left(f_i\right)-S_n<br><br>\left(f_i\right)<br><br>,$

where R_n(f_i) is the reference spectrum, S_n(f_i) is the input signal spectrum, and A_i is equal to zero if a frequency f_i of the input signal is due to a known noise and A_i is unity if no noise is present at the frequency f_i.

19. A speech method according to claim 15, wherein the spectral distance is the sum of the spectral distance calculations for a number of samples discerning the reference spectra from each other.

20. The speech recognition method according to claim 15 associated with a telephone, further comprising:

detecting and recognizing one or more speech answering commands, each forming an input spectrum.

21. The speech recognition method according to claim 20, further comprising:

responding to an accept call command to accept a call to the telephone.

22. The speech recognition method according to claim 20, further comprising:

responding to a reject call command to reject a call to the telephone.

23. The speech recognition method according to claim 20, further comprising:

responding to a forward call command is forward a call to the telephone.

24. The speech recognition method according to claim 15, wherein the first known noise signal is a periodic signal with a repeating pattern.

25. The speech recognition method according to claim 15, wherein the first known noise signal is a ring signal used to indicate a message.

26. The speech recognition method according to claim 15, wherein the first known noise signal is a melody or a buzzer signal.

27. The speech recognition method according to claim 15, wherein the first known noise signal is a signal output from a speaker.

28. The speech recognition method according to claim 15, wherein the function unrelated to speech recognition is to drive a speaker.