Speech communication system and method for improving speech intelligibility

1. A speech communication system for improving speech intelligibility, comprising:

one or more processors; and

a memory storing instructions that, when executed by the one or more processors, cause the system to perform:

determining a cutoff frequency based on a spectrum of noise in a sound signal and Signal-Noise-Ratios (SNRs) of a plurality of sub-bands of frequency of the sound signal, wherein the cutoff frequency defines a noise dominant region of frequency, and the determining comprises, from a lowest sub-band of frequency to an upper frequency limit of the spectrum of the noise, in each of the plurality of sub-bands of frequency:

determining whether an SNR of the sub-band is higher than a threshold, and in response to determining that the SNR is higher than the threshold, determining a previous sub-band of frequency as the cutoff frequency;

lifting a spectrum of a speech above the noise dominant region of frequency, wherein a frequency range of the spectrum of the speech is lifted by the cutoff frequency; and

applying an adaptive filter to the speech to achieve echo cancelation, wherein the adaptive filter is controlled by a volume of the noise.

2. The system according to claim 1 , wherein

the sound signal is received through a microphone of the system.

3. The system of claim 1 , wherein the SNRs are instantaneous SNRs, and wherein the instantaneous SNRs are smoothed over frames of the sound signal and adjacent sub-bands of frequency.

4. The system of claim 3 , wherein determining the cutoff frequency based on the spectrum of the noise and the SNRs further comprises:

in response to determining that the SNR is not higher than the threshold, determining whether a power of the noise below the sub-band of frequency is greater than a threshold percentage of a total power of the noise; and

in response to determining that the power of the noise below the sub-band of frequency is greater than the threshold percentage of the total power of the noise, setting the current sub-band of frequency as the cutoff frequency.

5. The system of claim 1 , wherein lifting a spectrum of a speech above the noise dominant region of frequency further comprises:

classifying a frame of the speech into one of the categories of vowel and consonant; and

if the frame of the speech is classified as a vowel, lifting the spectrum of the frame of the speech to a sub-band of frequency based on the cutoff frequency.

6. The system of claim 1 , wherein the instructions, when executed by the one or more processors, further cause the system to perform:

applying equalization to the lifted spectrum of the speech.

7. The system of claim 6 , wherein applying the equalization on the lifted speech further comprises:

transforming the spectrum of the speech from a linear frequency domain to critical bands of frequency domain, wherein a critical band of frequency is a band of frequency within which a first tone interferes with perception of a second tone; and

performing equalization on the speech in the critical bands of frequency.

8. The system of claim 6 , wherein applying the equalization on the lifted speech further comprises:

adjusting the lifted spectrum of the speech based on loudness of the speech.

9. The system of claim 6 , wherein the instructions, when executed by the one or more processors, further cause the system to perform:

applying spectra smoothing to the speech.

10. The system of claim 1 , wherein a sparsity of the adaptive filter increases if the volume of noise increases.

11. A computer-implemented method for speech communication, comprising:

determining a cutoff frequency based on a spectrum of noise in a sound signal and Signal-Noise-Ratios (SNRs) of a plurality of sub-bands of frequency of the sound signal, wherein the cutoff frequency defines a noise dominant region of frequency, and the determining comprises, from a lowest sub-band of frequency to an upper frequency limit of the spectrum of the noise, in each of the plurality of sub-bands of frequency:

determining whether an SNR of the sub-band is higher than a threshold, and

in response to determining that the SNR is higher than the threshold, determining a previous sub-band of frequency as the cutoff frequency;

lifting a spectrum of a speech above the noise dominant region of frequency, wherein a frequency range of the spectrum of the speech is lifted by the cutoff frequency; and

applying an adaptive filter to the speech to achieve echo cancelation, wherein the adaptive filter is controlled by a volume of the noise.

12. The method of claim 11 , wherein

the sound signal is received through a microphone of the system.

13. The method of claim 11 , wherein the SNRs are instantaneous SNRs, and wherein the instantaneous SNRs are smoothed over frames of the sound signal and adjacent sub-bands of frequency.

14. The method of claim 13 , wherein determining the cutoff frequency based on the spectrum of the noise and the SNRs further comprises:

in response to determining that the SNR is not higher than the threshold, determining whether a power of the noise below the sub-band of frequency is greater than a threshold percentage of a total power of the noise; and

in response to determining that the power of the noise below the sub-band of frequency is greater than the threshold percentage of the total power of the noise, setting the current sub-band of frequency as the cutoff frequency.

15. The method of claim 11 , wherein lifting the spectrum of the speech that is within the noise dominant region of frequency to a sub-band of frequency higher than the cutoff frequency further comprises:

classifying a frame of the speech into one of the categories of vowel and consonant; and

if the frame of the speech is classified as a vowel, lifting the spectrum of the frame of the speech that is within the noise dominant region of frequency to the sub-band of frequency higher than the cutoff frequency.

16. The method of claim 11 , further comprising:

applying equalization on the lifted spectrum of the speech.

17. The method of claim 16 , wherein applying the equalization on the lifted speech further comprises:

transforming the spectrum of the speech from a linear frequency domain to critical bands of frequency domain, wherein a critical band of frequency is a band of frequency within which a first tone interferes with perception of a first tone; and

performing equalization on the speech in the critical bands of frequency.

18. The method of claim 16 , wherein applying the equalization on the lifted speech further comprises:

adjusting the lifted spectrum of the speech based on loudness of the speech.

19. The method of claim 16 , further comprising: applying spectra smoothing to the speech.

20. A non-transitory computer-readable storage medium coupled to one or more processors and comprising instructions that, when executed by the one or more processors, cause the one or more processors to perform a method for speech communication, the method comprising:

determining a cutoff frequency based on a spectrum of noise in a sound signal and Signal-Noise-Ratios (SNRs) of a plurality of sub-bands of frequency of the sound signal, wherein the cutoff frequency defines a noise dominant region of frequency, and the determining comprises, from a lowest sub-band of frequency to an upper frequency limit of the spectrum of the noise, in each of the plurality of sub-bands of frequency:

determining whether an SNR of the sub-band is higher than a threshold, and

in response to determining that the SNR is higher than the threshold, determining a previous sub-band of frequency as the cutoff frequency;

lifting a spectrum of a speech above the noise dominant region of frequency, wherein a frequency range of the spectrum of the speech is lifted by the cutoff frequency; and

applying an adaptive filter to the speech to achieve echo cancelation, wherein the adaptive filter is controlled by a volume of the noise.