Granted Patent B2

US 12,272,375 · App. 17/959,487 · Granted Apr 8, 2025

Inventors: Muhammed Zahid Ozturk (Beltsville, MD); Chenshu Wu (Hong Kong, CN); Beibei Wang (Clarksville, MD); K. J. Ray Liu (Potomac, MD); Oscar Chi-Lim Au (Greenbelt, MD); Min Wu (Clarksville, MD)

Assignee: ORIGIN RESEARCH WIRELESS, INC.

G10L25/78G10L25/18G10L25/30G10L25/90H01Q3/22H04W4/021G10L2025/783

Abstract

Claims (74)

1. A system for radio-based voice activity detection, comprising:

a transmitter configured to transmit a standard-compliant radio signal through a wireless channel of a venue, wherein the standard-compliant radio signal is compliant to a standard being one of: a mobile cellular communication standard, a wireless local area network (WLAN) standard, a WiFi standard, an IEEE 802.11 standard, or an IEEE 802.11bf standard;

a receiver configured to receive the standard-compliant radio signal through the wireless channel, wherein the wireless channel is impacted by a voice activity of a target voice source and a contemporaneous voice activity of a non-target voice source in the venue; and

a processor configured for:

extracting a set of time series of channel information (TSCI) of the wireless channel from the received standard-compliant radio signal, wherein each channel information (CI) comprises one of: channel state information (CSI), channel frequency response (CFR) or channel impulse response (CIR),

performing beamforming based on the set of TSCI,

computing a directional TSCI (DTSCI) associated with a direction of the target voice source based on the beamforming,

computing a time series of radio feature (TSRF) associated with the target voice source based on the DTSCI, each radio feature computed based on a component of at least one respective CI of the DTSCI,

computing a radio spectrogram based on the TSRF,

detecting a pitch profile comprising a time profile of a pitch of the voice activity and harmonics of the pitch associated with the target voice source based on the radio spectrogram,

detecting a sequence of intermittent voiced speech and unvoiced speech based on a locally continuous time trend of the detected time profile of the pitch and the detected harmonics, and

detecting the voice activity of the target voice source, separately from the contemporaneous voice activity of the non-target voice source which is associated with another component of the at least one respective CI of the DTSCI or another DTSCI computed based on the beamforming, based on the detected sequence of intermittent voiced speech and unvoiced speech without using any media signal.

2. The system of claim 1 , wherein:

the media signal comprises at least one of: a microphone signal, a speech signal, a vocal signal, an audio signal, a signal less than 40 kHz, an acoustic signal, an audible signal, a telephone signal, a tele-conferencing signal, an audio-telephony signal, a conference call signal, a visual signal, a video signal, a video-telephony signal, a video-conferencing signal, a media streaming signal, or a multimedia signal.

3. The system of claim 2 , wherein:

the standard-compliant radio signal is a data communication signal; and

the voice activity of the target voice source is detected without using any data payload communicated in the data communication signal.

4. The system of claim 3 , wherein:

the voice activity of the target voice source is detected without using any media signal data communicated in the data communication signal.

5. The system of claim 4 , wherein:

the voice activity of the target voice source is associated with a voice producing motion of the target voice source.

6. The system of claim 1 , wherein:

each radio feature (RF) of the TSRF is computed based on a respective sliding window of the DTSCI; and

the TSRF is a baseband signal bandlimited to 1 MHz.

7. The system of claim 1 , wherein detecting the voice activity further comprises:

detecting an instantaneous pitch associated with a current time instance in the TSRF, wherein an instantaneous fundamental frequency associated with the instantaneous pitch is greater than a lower threshold and less than an upper threshold; and

detecting the voice activity based on the detected instantaneous pitch.

8. The system of claim 7 , wherein detecting the voice activity further comprises:

detecting at least one instantaneous harmonics of the instantaneous pitch associated with the current time instance in the TSRF, wherein a frequency associated with each respective harmonics is an integer multiple of the instantaneous fundamental frequency associated with the instantaneous pitch; and

detecting the voice activity based on the at least one instantaneous harmonics of the instantaneous pitch.

9. The system of claim 8 , wherein detecting the voice activity further comprises:

detecting the pitch profile comprising a plurality of instantaneous pitches associated with a plurality of respective time instances in the TSRF, wherein each instantaneous pitch associated with a respective time instance is associated with a respective instantaneous fundamental frequency greater than the lower threshold and less than the upper threshold;

detecting a voice-related time trend of the plurality of instantaneous pitches in the pitch profile; and

detecting the voice activity based on the pitch profile and the voice-related time trend of the plurality of instantaneous pitches.

10. The system of claim 9 , wherein the voice-related time trend comprises at least one of:

a local continuity of instantaneous pitches,

a local continuity of the instantaneous fundamental frequencies,

a local continuity of frequency of the at least one instantaneous harmonics,

a habitual pitch, a long term pitch, a variation of pitch around the habitual pitch,

a timing or pacing of pitch change,

a fast pitch change within a tone, or

a slow pitch change reflecting prosody.

11. The system of claim 10 , wherein the processor is further configured for:

processing the TSRF with a neural network; and

detecting the pitch profile based on the processing of the TSRF.

12. The system of claim 10 , wherein the processor is further configured for:

computing a frequency decomposition of the TSRF by computing at least one of: a spectrogram, a short-time Fourier transform (STFT), a wavelet transform, a filter-bank representation, a harmonic analysis, a Fourier analysis, a multi-resolution analysis, a time-frequency decomposition, a time-frequency representation, a sonograph, a voiceprint, a voicegram, or a waterfall display; and

detecting the pitch profile based on the frequency decomposition.

13. The system of claim 12 , wherein detecting the pitch profile comprises:

processing the frequency decomposition of the TSRF with a neural network; and

detecting the pitch profile based on the processing of the frequency decomposition of the TSRF.

14. The system of claim 13 , wherein:

the processor is further configured for detecting an instantaneous pitch based on the frequency decomposition of the TSRF in a time window associated with the current time instance.

15. The system of claim 14 , wherein:

the processor is further configured for detecting an instantaneous harmonics based on the frequency decomposition of the TSRF in a time window associated with the current time instance.

16. The system of claim 15 , wherein:

the transmitter or the receiver has an array of antennas;

each TSCI in the set of TSCI is associated with a respective antenna in the array of antennas; and

the DTSCI is associated with a direction relative to the array of antennas.

17. The system of claim 16 , wherein the processor is further configured for:

associating the target voice source with a component of the DTSCI, wherein there is at least one non-target voice source undergoing respective asynchronous voice producing motion, wherein the wireless channel is impacted asynchronously by the respective asynchronous voice producing motion of the at least one non-target voice source;

associating each non-target voice source with a different component of the DTSCI or a different directional TSCI;

selecting the component of the DTSCI; and

rejecting the at least one non-target voice source by computing the TSRF based on the selected component of the DTSCI.

18. A method for radio-based voice activity detection, comprising:

obtaining a standard-compliant radio signal transmitted from a transmitter to a receiver through a wireless channel of a venue, wherein the wireless channel is impacted by a voice activity of a target voice source and a contemporaneous voice activity of a non-target voice source in the venue, wherein the standard-compliant radio signal is compliant to a standard being one of: a mobile cellular communication standard, a wireless local area network (WLAN) standard, a WiFi standard, an IEEE 802.11 standard, or an IEEE 802.11bf standard;

extracting a set of time series of channel information (TSCI) of the wireless channel from the obtained standard-compliant radio signal, wherein each channel information (CI) comprises one of: channel state information (CSI), channel frequency response (CFR) or channel impulse response (CIR);

performing beamforming based on the set of TSCI;

computing a directional TSCI (DTSCI) associated with a direction of the target voice source based on the beamforming;

computing a time series of radio feature (TSRF) associated with the target voice source based on the DTSCI, each radio feature computed based on a component of at least one respective CI of the DTSCI;

computing a radio spectrogram based on the TSRF;

detecting a pitch profile comprising a time profile of a pitch of the voice activity and harmonics of the pitch associated with the target voice source based on the radio spectrogram;

detecting a sequence of intermittent voiced speech and unvoiced speech based on a locally continuous time trend of the detected time profile of the pitch and the detected harmonics; and

detecting the voice activity of the target voice source, separately from the contemporaneous voice activity of the non-target voice source which is associated with another component of the at least one respective CI of the DTSCI or another DTSCI computed based on the beamforming, based on the detected sequence of intermittent voiced speech and unvoiced speech without using any media signal.

Assignments (1)

Continuity (40)

References Cited (8)

Cited By (2)

US 12,417,566 US 12,571,898