DOUBLE TALK DETECTOR

• US 20100063809A1
• Filed: 02/21/2007
• Published: 03/11/2010
• Est. Priority Date: 02/21/2007
• Status: Active Grant

• Current Assignee: Telefonaktiebolaget LM Ericsson
• Original Assignee: Telefonaktiebolaget LM Ericsson
• Inventors: Trump, Tonu
• Granted Patent: US 8,260,613 B2
• Time in Patent Office: Days
• US Class Current: 704/228
• CPC Class Codes: G10L 2021/02082 the noise being echo, reverberation of the speech
• CPC Class Codes: G10L 21/0208 Noise filtering
• CPC Class Codes: G10L 25/48 specially adapted for particular use
• CPC Class Codes: H04M 9/082 using echo cancellers echo cancellers per se H04B3/23

Abstract

A double talk detector for controlling the echo path estimation in a telecommunication system by indicating when a received coded speech signal is dominated by a non-echo signal; i.e., that so-called double talk exists. This is determined by extracting LSPs from a coded speech frame of the received coded speech signal when the signal power exceeds a first threshold value, converting each of said extracted LSPs into LSFs, and calculating the distance between each two adjacent LSFs. For each distance that is smaller than a second threshold, a spectral peak is located between the two LSFs, and it is determined whether said spectral peak is an echo or not. When a predetermined number of non-echo spectral peaks are located in the received speech signal, double talk will be indicated, and the echo path estimation may be disabled.

Claims

1. A method of detecting that a received coded speech signal is dominated by a non-echo signal, comprising the steps of:

if the signal power of a received speech signal exceeds a first threshold value, then extracting the Line Spectral Pairs (LSPs) from a coded speech frame of said received speech signal;

converting each of said extracted LSPs into Line Spectral Frequencies (LSFs), ω

i, and calculating the distance between each two adjacent LSFs; and

,

for each of said distances that is smaller than a second threshold value, calculating the frequency of the spectral peak, ω

c, surrounded by said LSFs, and determining whether said spectral peak is an echo.

2. A method according to claim 1, comprising the further steps of:

incrementing a counter for each located spectral peak that is not an echo;

indicating double-talk when the counter reaches a predetermined threshold value.

3. A method according to claim 1, wherein the determination of whether a spectral peak is an echo comprises the steps of:

extracting the LSPs from a coded speech frame of a first speech signal and converting said LSPs into the corresponding LSFs; and

,

determining that said spectral peak is an echo if the distance between the adjacent LSFs surrounding said spectral peak in the first speech signal is smaller than a third threshold value.

4. A method according to claim 1, further comprising determining whether a spectral peak is an echo only if the frequency of said spectral peak is lower than a fourth threshold value.

5. A method according to claim 3, further comprising searching for a closely located second spectral peak in the first speech signal if said spectral peak is not an echo, comprising the steps of:

calculating a second spectral peak in the first speech signal from two adjacent LSFs; and

,

determining that said second spectral peak is an echo if the distance between the calculated second spectral peak and the centre frequency is smaller than a fifth threshold value.

6. A method according to claim 1, wherein the LSFs are obtained by converting each of the extracted LSPs into the corresponding LSF using the relationship qi=cos(ω

i).

7. A method according to claim 1, wherein the centre frequency of a spectral peak is determined by locating two adjacent LSPs, ω

i+1−

ω

i, and calculating ω

c=(ω

i+1−

ω

i)/2.

8. A method according to claim 2, wherein an indication of double talk deactivates an echo path estimation in a telecommunication system.

9. A method of estimating an echo path in a telecommunication system, wherein the echo path estimation is deactivated by a double talk indication obtained by a method according to claim 2.

10. A method of estimating the echo path in a telecommunication system, wherein the echo path estimation is deactivated when a received coded speech signal is dominated by a non-echo-signal, the method comprising the steps of:

if the signal power of a received speech signal exceeds a first threshold value, then extracting the Line Spectral Pairs (LSPs) from a coded speech frame of said received speech signal;

converting each of said extracted LSPs into a corresponding Line Spectral Frequency (LSF), ω

i, and calculating the distance between each two adjacent LSFs;

for each distance that is smaller than a second threshold value, calculating the frequency of the spectral peak, ω

c, surrounded by said LSF, and determining whether said spectral peak is an echo; and

,

deactivating the echo path estimation when a predetermined number of spectral non-echo peaks are located.

11. A method according to claim 10, wherein the determination of whether a spectral peak is an echo comprises the of:

extracting the LSPs from a coded speech frame of a first speech signal and converting each extracted LSP into the corresponding LSF; and

,

determining that said spectral peak is an echo if the distance between the LSFs surrounding said spectral peak in the first speech signal is smaller than a third threshold value.

12. A method according to claim 10, further comprising determining whether a spectral peak is an echo only if the frequency of said spectral peak is lower than a fourth threshold value.

13. A method according to claim 10, wherein the speech coding is based on AMR-CELP-coding.

14. A double talk detector arranged to be connected to a coded received speech signal and to a coded transmitted first speech signal in a telecommunication system for determining when said received speech signal is dominated by a non-echo signal, the double talk detector comprising:

a signal power monitor arranged to determine when the signal power of a received speech signal exceeds a first threshold value;

a spectral peak locator arranged to;

extract the LSPs from a coded speech frame of the received speech signal and to convert each of said extracted LSPs into a corresponding LSF; and

,

calculate the distance between each two adjacent LSFs, and calculate a spectral peak, ω

c, surrounded by two LSFs separated by a distance that is smaller than a second threshold value; and

,

an echo locator arranged to determine whether a located spectral peak is an echo.

15. A double talk detector according to claim 14, further arranged to indicate that said received speech signal is dominated by a non-echo signal if a predetermined number of spectral non-echo peaks are located in the received signal.

16. A double talk detector according to claim 14, further comprising a counter for counting the number of spectral non-echo peaks in the received speech signal.

17. A double talk detector according to claim 14, wherein the echo locator is arranged to:

extract the LSPs from a coded speech frame of the first speech signal and convert said extracted LSPs into the corresponding LSFs; and

,

calculate the distance between the adjacent LSFs surrounding said spectral peak and determine that the spectral peak is an echo if said calculated distance is smaller than a third threshold value.

18. A double talk detector according to claim 14, wherein the echo locator is arranged to determine whether a spectral peak is an echo only if the frequency of said spectral peak is lower than a fourth threshold value.

19. A double talk detector according to claim 17, wherein said echo locator is further arranged to search for a closely located second spectral peak in the first speech signal, if said spectral peak is not an echo, by:

calculating a second spectral peak from said LSPs and determining that said second spectral peak is an echo if the distance to the centre frequency is smaller than a fifth threshold value.

20. A double talk detector according to claim 14, wherein the LSFs are obtained by converting each of the extracted LSPs into the corresponding LSF using the relationship qi=cos(ω

i).

21. A double talk detector according to claim 14, wherein the centre frequency of a spectral peak is determined by locating two adjacent LSFs, ω

i+1−

ω

i, and calculating ω

c=(ω

i+1−

ω

i)/2.

22. An echo path estimating device for a telecommunication system, arranged to deactivate the estimation of an echo path dominated by a non-echo-signal, characterized by comprising a double talk detector for a coded speech signal according to claim 14.

23. An echo path estimating device for a telecommunication system, arranged to deactivate the estimation of an echo path dominated by a non-echo-signal, the echo path estimating device comprising:

a double talk detector arranged to be connected to a coded received speech signal and to a coded transmitted first speech signal in said telecommunication system, said double talk detector comprising;

a signal power monitor arranged to determine when the signal power of a received speech signal exceeds a first threshold value;

a spectral peak locator arranged to;

extract Line Spectral Pairs (LSPs) from a coded speech frame of the received speech signal and to convert each of said extracted LSPs into a corresponding Line Spectral Frequency (LSF), ω

i;

calculate the distance between each two adjacent LSFs, and calculate a spectral peak, ω

c, surrounded by two LSFs separated by a distance that is smaller than a second threshold value; and

,

an echo locator arranged to determine whether a located spectral peak is an echo.

24. An echo path estimating device according to claim 23, wherein the double talk detector is further arranged to indicate that said received speech signal is dominated by a non-echo signal if a predetermined number of spectral non-echo peaks are located in the received signal.

25. An echo path estimating device according to claim 24, wherein the double talk detector comprises a counter for counting the number of spectral non-echo peak in the received signal.

26. An echo path estimating device according to claim 23, wherein the echo locator is further arranged to:

extract the LSPs from a coded speech frame of the first speech signal, and convert said extracted LSPs into the corresponding LSFs; and

,

calculate the distance between the adjacent LSFs surrounding said spectral peak and determine that the spectral peak is an echo if said calculated distance is smaller than a third threshold value.

27. An echo path estimating device according to claim 23, wherein the echo locator is arranged to determine whether a spectral peak is an echo only if the frequency of said spectral peak is lower than a fourth threshold value.

28. An echo path estimating device according to claim 26, wherein said echo locator is further arranged to search for a closely located second spectral peak in the first speech signal if said spectral peak is not an echo, by calculating a second spectral peak from two adjacent LSFs, and determining that said second spectral peak is an echo if the distance to the centre frequency is smaller than a fifth threshold value.