Back to results
US 2018/0293998 A1Published application
Methods and apparatus for low cost voice activity detector
AssigneeTexas Instruments Incorporated
InventorsGanesan Thiagarajan, Tarkesh Pande, David Patrick Magee
Life of this patent
Estimating the term…
Abstract
In described examples, a method for detecting voice activity includes: receiving a first input signal containing noise; sampling the first input signal to form noise samples; determining a first value corresponding to the noise samples; subsequently receiving a second input signal; sampling the second input signal to form second signal samples; determining a second value corresponding to the second signal samples; forming a ratio of the second value to the first value; comparing the ratio to a predetermined threshold value; and responsive to the comparing, indicating whether voice activity is detected in the second input signal.
Claims29
(canceled)
A method of detecting voice activity, the method comprising:
- receiving a first input signal containing noise;
- sampling the first input signal to form noise samples;
- determining a first value corresponding to the noise samples;
- subsequently receiving a second input signal;
- sampling the second input signal to form signal samples;
- determining a second value corresponding to the signal samples;
- forming a ratio of the second value to the first value, in which forming the ratio includes computing a log likelihood ratio by using an inverse covariance of the noise samples and a covariance of the signal samples;
- comparing the ratio to a predetermined threshold value;
- and
- responsive to the comparing, indicating whether voice activity is detected in the second input signal.
7 more elements
A method of detecting voice activity, the method comprising:
- receiving a first input signal containing noise;
- sampling the first input signal to form noise samples;
- determining a first value corresponding to the noise samples;
- subsequently receiving a second input signal;
- sampling the second input signal to form signal samples;
- determining a second value corresponding to the signal samples;
- forming a ratio of the second value to the first value, in which forming the ratio includes computing a mean ratio by computing the second value as a mean energy value of the signal samples and by computing the first value as a mean energy value of the noise samples;
- comparing the ratio to a predetermined threshold value;
- and
- responsive to the comparing, indicating whether voice activity is detected in the second input signal.
7 more elements
(canceled)
A method of detecting voice in an input signal, the method comprising:
- in a noise state, sampling an input signal containing noise and no voice to form noise samples, and computing a noise value from the noise samples;
- in the noise state, sampling an input signal to form first signal samples, and computing a first voice value from the first signal samples;
- in the noise state, detecting potential voice activity in the first signal samples of the input signal;
- responsive to the detecting, transitioning to a voice state;
- in the voice state, sampling the input signal to form second signal samples, and computing a second voice value from the second signal samples;
- in the voice state, computing a first ratio of the second voice value to the noise value, including computing a log likelihood ratio of the second voice value to the noise value, in which the second voice value is a covariance matrix of the second signal samples, and the noise value is a covariance matrix of the noise samples;
- in the voice state, if the first ratio exceeds a predetermined threshold, indicating a voice activity detection, and remaining in the voice state;
- in the voice state, if the first ratio does not exceed the predetermined threshold, transitioning to an interim state;
- in the interim state, sampling the input signal to form third signal samples, and computing a third voice value from the third signal samples;
- in the interim state, computing a second ratio of the third voice value to the noise value;
- in the interim state, if the second ratio exceeds the predetermined threshold, transitioning to the voice state;
- and
- in the interim state, if the second ratio does not exceed the predetermined threshold, transitioning to the noise state.
10 more elements
A method of detecting voice in an input signal, the method comprising:
- in a noise state, sampling an input signal containing noise and no voice to form noise samples, and computing a noise value from the noise samples;
- in the noise state, sampling an input signal to form first signal samples, and computing a first voice value from the first signal samples;
- in the noise state, detecting potential voice activity in the first signal samples of the input signal;
- responsive to the detecting, transitioning to a voice state;
- in the voice state, sampling the input signal to form second signal samples, and computing a second voice value from the second signal samples;
- in the voice state, computing a first ratio of the second voice value to the noise value, including computing a mean ratio in which the second voice value is a mean of energy of the second signal samples, and the noise value is a mean of energy of the noise samples;
- in the voice state, if the first ratio exceeds a predetermined threshold, indicating a voice activity detection, and remaining in the voice state;
- in the voice state, if the first ratio does not exceed the predetermined threshold, transitioning to an interim state;
- in the interim state, sampling the input signal to form third signal samples, and computing a third voice value from the third signal samples;
- in the interim state, computing a second ratio of the third voice value to the noise value;
- in the interim state, if the second ratio exceeds the predetermined threshold, transitioning to the voice state;
- and
- in the interim state, if the second ratio does not exceed the predetermined threshold, transitioning to the noise state.
10 more elements
(canceled)
The apparatus of claim 18, in which the MCU further includes a programmable processor to execute instructions to implement the state machine.
Apparatus, comprising:
- an input terminal to receive voice input signals;
- an analog front end, coupled to the input terminal, to output digital signals corresponding to signals received at the input terminal, and having an automatic gain control input;
- a filter bank coupled to receive and output samples of the digital signals;
- a microcontroller unit (MCU) to implement a state machine, in which the MCU is configured to:
- in a noise state, compute:
- a noise value from first samples corresponding to noise and no voice at the input terminal;
- and a second value from second samples at the input terminal;
- in the noise state, compute a ratio of the second value to the noise value, including computing a log likelihood ratio using:
- the noise value, which is a covariance matrix of the first samples;
- and the second value, which is a covariance matrix of the second samples;
- in the noise state, transition to a voice state if the ratio exceeds a threshold;
- in the voice state, update the second value with additional samples from the input terminal, update the ratio, and transition to an interim state if the updated ratio is less than the threshold;
- and
- in the interim state, again update the second value with additional samples from the input terminal, again update the ratio, transition to the noise state if the updated ratio is less than the threshold, and transition to the noise state if the automatic gain control input to the analog front end is increasing.
11 more elements
Dependencies are inferred from the claim text.