Signal processor with side-tone noise reduction for a headset

1. A signal processor for a headset configured with a microphone terminal for receiving a microphone signal, a loudspeaker terminal for outputting a loudspeaker signal, and a far-end terminal for communicating an inbound signal and an outbound signal with a far-end; comprising:

an outbound path from the microphone terminal to the far-end terminal;

a side-tone path configured to generate a side-tone signal from the microphone signal via a side-tone filter,

an inbound path from the far-end terminal to the loudspeaker terminal and comprising a combiner configured to combine the side-tone signal and the inbound signal to generate the loudspeaker signal; and

a side-tone filter controller that receives the microphone signal and computes a first noise estimate with a signal-to-noise level of the microphone signal at respective frequency bands and based thereon controls the side-tone filter to improve a signal-to-noise ratio of the side-tone signal;

wherein the side-tone filter in the side-tone path comprises multiple filter stages that are individually controlled based on the first noise estimate; and

wherein the multiple filter stages are configured from a group consisting of: multiple peak filters and a high-shelf filter;

and wherein the side-tone filter controller executes an iterative process of:

determining parameters for a filter stage based on the first noise estimate;

computing a frequency domain gain estimate of the frequency domain gain transfer function for a filter stage;

computing a second noise estimate by subtracting the frequency domain gain estimate from the first noise estimate;

updating the first noise estimate to be in accordance with the second noise estimate;

wherein the filter stages are controlled via coefficients computed from the parameters for a respective filter stage computed over iterations of the iterative process; and

wherein determining parameters for a filter stage comprises fitting a filter that has a largest possible gain-bandwidth product while optimizing the signal-to-noise ratio in an audio band of interest.

2. A signal processor according to claim 1 , wherein fitting a filter comprises:

over a range of frequencies, iteratively selecting a frequency; and

determining a gain-bandwidth product of a band centred at the frequency and determining a gain factor thereof based on the signal-to-noise ratio and determining a bandwidth factor thereof as the bandwidth wherein the signal-to-noise ratio is not better than approximately half the signal-to-noise ratio at the frequency;

when the range of frequencies has been iterated, selecting a favourable gain-bandwidth product, such as the largest gain-bandwidth product, and estimating parameters for a band-attenuate filter, such as a peak-filter, with a centre frequency and a gain to approximately fit the determined band.

3. A signal processor according to claim 1 , wherein fitting a filter comprises:

fitting a high-shelf filter by selecting a predetermined shelf mid-point frequency and determining a shelf gain, such as an attenuation, from an average of the signal-to-noise ratio at frequencies above the shelf mid-point frequency.

4. A signal processor according to claim 1 , wherein the signal processor:

at a first time, computes a reference gain based on a minimum attenuation in the noise estimate and scales the microphone signal based on the reference gain; and

at a second time, following the first time, and iteratively selecting a frequency.

5. A signal processor according to claim 4 , wherein the signal processor:

at a second time, computes a frequency domain gain estimate of a high-shelf filter, and then computes a second noise estimate by subtracting the frequency domain gain estimate from the first noise estimate;

at a third time, following the second time, and iteratively selecting a frequency, wherein the filter is a stop-band filter or a peak-filter.

6. A signal processor according to claim 1 , wherein a cost function is defined for an optimization process; and

wherein the cost function is a frequency domain estimate of the signal-to-noise ratio which in a first iteration is computed from the microphone signal and in subsequent iterations is computed from the cost function and estimates in the frequency domain of selected filters; and

wherein the optimization process iteratively selects a filter with an optimization strategy to optimize the cost function and then updates the cost function.

7. A signal processor according to claim 1 , wherein determining parameters for a filter stage by fitting a filter comprises dampening the rate and/or distance the centre frequency and/or bandwidth is allowed to change over time.

8. A signal processor according to claim 1 , comprising a voice activity detector which indicates presence of speech;

wherein a first amplitude or power spectrum for the microphone signal is computed at a time when the voice activity detector is detecting speech;

wherein a second amplitude or power spectrum for the microphone signal is computed at a time when the voice activity detector is not detecting speech; and

wherein the signal-to-noise ratio is computed from the first amplitude or power spectrum and the second amplitude or power spectrum.

9. A headset comprising a voice microphone, an earpiece with a loudspeaker and a signal processor according to claim 1 .

10. A signal processing method for generating a side-tone in a headset, comprising:

generating a side-tone signal via a side-tone filter and from a microphone signal picked up via a microphone of the headset,

combining the side-tone signal and an inbound signal to the headset to generate a loudspeaker signal for a loudspeaker of the headset;

computing a first noise estimate of the microphone signal at respective frequency bands and based thereon controlling the side-tone filter to improve a signal-to-noise ratio of the side-tone signal;

wherein the side-tone filter comprises multiple filter stages that are individually controlled based on the first noise estimate; and

wherein the multiple filter stages are configured from a group consisting of: multiple peak filters and a high-shelf filter.

11. A computer-readable medium carrying a program configured to perform the method as claimed in claim 10 , wherein the method is a computer-implemented method run on a computer.

12. A signal processor for a headset configured with a microphone terminal for receiving a microphone signal, a loudspeaker terminal for outputting a loudspeaker signal, and a far-end terminal for communicating an inbound signal and an outbound signal with a far-end; comprising:

an outbound path from the microphone terminal to the far-end terminal;

a side-tone path configured to generate a side-tone signal from the microphone signal via a side-tone filter,

an inbound path from the far-end terminal to the loudspeaker terminal and comprising a combiner configured to combine the side-tone signal and the inbound signal to generate the loudspeaker signal; and

a side-tone filter controller that receives the microphone signal and computes a first noise estimate with a signal-to-noise level of the microphone signal at respective frequency bands and based thereon controls the side-tone filter to improve a signal-to-noise ratio of the side-tone signal;

wherein the side-tone filter in the side-tone path comprises multiple filter stages that are individually controlled based on the first noise estimate; and

wherein the multiple filter stages are configured from a group consisting of: multiple peak filters and a high-shelf filter.

13. A signal processor according to claim 12 wherein fitting a filter comprises:

over a range of frequencies, iteratively selecting a frequency; and

determining a gain-bandwidth product of a band centered at the frequency and determining a gain factor thereof based on the signal-to-noise ratio and determining a bandwidth factor thereof as the bandwidth wherein the signal-to-noise ratio is not better than approximately half the signal-to-noise ratio at the frequency;

when the range of frequencies has been iterated, selecting a favorable gain-bandwidth product, such as the largest gain-bandwidth product, and estimating parameters for a band-attenuate filter, such as a peak-filter, with a center frequency and a gain to approximately fit the determined band.

14. A signal processor according to claim 13 wherein fitting a filter comprises:

fitting a high-shelf filter by selecting a predetermined shelf mid-point frequency and determining a shelf gain, such as an attenuation, from an average of the signal-to-noise ratio at frequencies above the shelf mid-point frequency.

15. A signal processor according to claim 13 , wherein the signal processor:

at a first time, computes a reference gain based on a minimum attenuation in the noise estimate and scales the microphone signal based on the reference gain; and

at a second time, following the first time, and iteratively selecting a frequency.

16. A signal processor according to claim 15 , wherein the signal processor:

at a second time, computes a frequency domain gain estimate of a high-shelf filter, and then computes a second noise estimate by subtracting the frequency domain gain estimate from the first noise estimate;

at a third time, following the second time, and iteratively selecting a frequency, wherein the filter is a stop-band filter or a peak-filter.

17. A signal processor according to claim 13 , wherein determining parameters for a filter stage by fitting a filter comprises dampening the rate and/or distance the center frequency and/or bandwidth is allowed to change over time.

18. A signal processor according to claim 13 , comprising a voice activity detector which indicates presence of speech;

wherein a first amplitude or power spectrum for the microphone signal is computed at a time when the voice activity detector is detecting speech;

wherein a second amplitude or power spectrum for the microphone signal is computed at a time when the voice activity detector is not detecting speech; and

wherein the signal-to-noise ratio is computed from the first amplitude or power spectrum and the second amplitude or power spectrum.

19. A signal processor according to claim 12 , wherein a cost function is defined for an optimization process; and

wherein the cost function is a frequency domain estimate of the signal-to-noise ratio which in a first iteration is computed from the microphone signal and in subsequent iterations is computed from the cost function and estimates in the frequency domain of selected filters; and

wherein the optimization process iteratively selects a filter with an optimization strategy to optimize the cost function and then updates the cost function.