# US 2019/0172481 A1 · Pitch detection algorithm based on pwvt of teager energy operator

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2019/0172481 A1
- Document ID: US20190172481A1
- Status: Published application
- Kind: A1
- Application number: 16/272,034
- Filed: 2019-02-11
- Published: 2019-06-06
- Earliest claimed: 2016-03-31
- Assignees: OmniSpeech LLC
- Applicants: OmniSpeech LLC
- Inventors: Vahid KHANAGHA
- Web page: https://patentgrep.com/patent/US20190172481A1

## CPC

- G10L 25/90: Pitch determination of speech signals
- G10L 15/1815: {Semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning}
- G10L 25/06: the extracted parameters being correlation coefficients
- G10L 25/18: the extracted parameters being spectral information of each sub-band
- G10L 2015/223: {Execution procedure of a spoken command}
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING

## Abstract

A pitch detection method. Such a pitch detection method may use a Teager Energy Operator (TEO) with a Pseudo Weigner Ville Transformation (PWVT) to recover speech from noise and to recover low-frequency information of the speech signal in its detection of a pitch value. Also, the pitch detection method may use the combinatory PWVT and the respective state machine for decision making.

## Claims

### Claim 1

A method for a pitch detection comprising:
applying, by a processor communicatively coupled to one or more memory devices, a Teager Energy Operator (TEO) to a first signal to generate a second signal;
computing, by the processor, a Pseudo Weigner Ville Transformation (PWVT) to generate a third signal from the first signal;
computing, by the processor, the PWVT to generate a fourth signal from the second signal;
generating, by the processor, a combined signal from the third signal and the fourth signal;
computing, by the processor, to generate a pitch candidate, a harmonic summation on at least one of the combined signal and the third signal; and
deciding, by the processor, the pitch candidate as a pitch value, the pitch candidate is larger than a threshold.

### Claim 2 (depends on 1)

The method of claim 1, further comprising:
sampling, by the processor, a signal to generate the first signal so that the first signal is a first discrete time signal and the second signal is a second discrete time signal;
generating, by the processor, a first complex valued signal from the first discrete time signal; and
generating, by the processor, a second complex valued signal from the second discrete time signal,
wherein when sampling the signal, the signal is each frame of a speech signal.

### Claim 3 (depends on 2)

The method of claim 2, further comprising:
applying, by the processor, a Hilbert Transform to the first discrete time signal to generate the first complex valued signal; and
applying, by the processor, a Hilbert Transform to the second discrete time signal to generate the second complex valued signal,
wherein each of the first complex valued signal and the second complex valued signal is low-pass filtered, buffered and down-sampled,
wherein the processor computes the PWVT on each of the first complex valued signal and the second complex valued signal to generate a spectral representation so that the third signal and the fourth signal are a first spectral representation and a second spectral representation, and
wherein the combined signal is a combined spectral representation generated from the first spectral representation and the second spectral representation.

### Claim 4 (depends on 1)

The method of claim 1, wherein the TEO acts as a non-linear filter which outputs a constant value from a tonal periodic waveform and recovers a low-frequency content of a high-pass filtered impulsive waveform, and
wherein the PWVT is computed by at least one of Fourier Transform as a frequency domain representation of the signal and 512 points Discrete Fast Fourier Transform (DFFT) of a down-sampled Hilbert transform of the signal to form a frequency domain representation of the signal over time in the frequency range of 0 to 1 Khz.

### Claim 5 (depends on 3)

The method of claim 3, wherein when generating the combined spectral representation, a low-frequency part of the combined spectral representation is taken from the second spectral representation and a remaining part of the combined spectral representation is taken from the first spectral representation,
wherein the harmonic summation is computed on the combined spectral representation if a relative amount of a low-frequency energy of the signal is larger than a pre-determined threshold, and
wherein the harmonic summation is computed within a human pitch frequency range.

### Claim 6 (depends on 3)

The method of claim 3, wherein when deciding the pitch candidate as the pitch value,
the pitch candidate is a corresponding frequency location of a maximum value of the harmonic summation, and
the threshold is fifty times of a median value of at least one of the combined spectral representation and the first spectral representation.

### Claim 7 (depends on 2)

The method of claim 2, wherein the threshold is at least one of the fifty times of the median value of the PWVT of a current frame and an average value of more than one hundred frames of the speech signal in which the pitch value is not detected.

### Claim 8 (depends on 3)

The method of claim 3, wherein when applying the Hilbert Transform, a half band Finite-Impulse-Response (FIR) filter is applied for a real part, and shift in an amount of π/2 is applied for an imaginary part,
wherein an anti-aliasing low-pass filter with cut-off frequency of 1 KHz is applied to low-pass filter each of the first complex valued signal and the second complex valued signal,
wherein each of the first complex valued signal and the second complex valued signal is buffered to form the analysis window with 60 msec length, and
wherein each of the first complex valued signal and the second complex valued signal is down sampled by a factor 4.

### Claim 9

A system for a pitch detection comprising:
one or more memory devices; and
at least one processor communicatively coupled to the one or more memory devices, the processor operable to:
apply a Teager Energy Operator (TEO) to a first signal to generate a second signal;
compute a Pseudo Weigner Ville Transformation (PWVT) to generate a third signal from the first signal;
compute the PWVT to generate a fourth signal from the second signal;
generate a combined signal from the third signal and the fourth signal;
compute a harmonic summation on at least one of the combined signal and the third signal to generate a pitch candidate; and
decide the pitch candidate as a pitch value, the pitch candidate is larger than a threshold.

### Claim 10 (depends on 9)

The system of claim 9, the processor further operable to:
perform a sampling a signal to generate the first signal so that the first signal is a first discrete time signal and the second signal is a second discrete time signal;
generate a first complex valued signal from the first discrete time signal; and
generate a second complex valued signal from the second discrete time signal,
wherein in the sampling the signal, the signal is each frame of a speech signal.

### Claim 11 (depends on 10)

The system of claim 10, the processor further operable to:
apply a Hilbert Transform to the first discrete time signal to generate the first complex valued signal; and
apply a Hilbert Transform to the second discrete time signal to generate the second complex valued signal,
wherein each of the first complex valued signal and the second complex valued signal is low-pass filtered, buffered and down-sampled,
wherein the processor computes the PWVT on each of the first complex valued signal and the second complex valued signal to generate a spectral representation so that the third signal and the fourth signal are a first spectral representation and a second spectral representation, and
wherein the combined signal is a combined spectral representation generated from the first spectral representation and the second spectral representation.

### Claim 12 (depends on 9)

The system of claim 9, wherein the TEO acts as a non-linear filter which outputs a constant value from a tonal periodic waveform and recovers a low-frequency content of a high-pass filtered impulsive waveform, and
wherein the PWVT is computed by at least one of Fourier Transform as a frequency domain representation of the signal and 512 points Discrete Fast Fourier Transform (DFFT) of a down-sampled Hilbert transform of the signal to form a frequency domain representation of the signal over time in the frequency range of 0 to 1 Khz.

### Claim 13 (depends on 11)

The system of claim 11, wherein when generating the combined spectral representation, a low-frequency part of the combined spectral representation is taken from the second spectral representation and a remaining part of the combined spectral representation is taken from the first spectral representation,
wherein the harmonic summation is computed on the combined spectral representation if a relative amount of a low-frequency energy of the signal is larger than a pre-determined threshold, and
wherein the harmonic summation is computed within a human pitch frequency range.

### Claim 14 (depends on 11)

The system of claim 11, wherein when deciding the pitch candidate as the pitch value,
the pitch candidate is a corresponding frequency location of a maximum value of the harmonic summation, and
the threshold is fifty times of a median value of at least one of the combined spectral representation and the first spectral representation.

### Claim 15 (depends on 10)

The system of claim 10, wherein the threshold is at least one of the fifty times of the median value of the PWVT of a current frame and an average value of more than one hundred frames of the speech signal in which the pitch value is not detected.

### Claim 16 (depends on 11)

The system of claim 11, wherein when applying the Hilbert Transform, a half band Finite-Impulse-Response (FIR) filter is applied for a real part, and shift in an amount of π/2 is applied for an imaginary part,
wherein an anti-aliasing low-pass filter with cut-off frequency of 1 KHz is applied to low-pass filter each of the first complex valued signal and the second complex valued signal,
wherein each of the first complex valued signal and the second complex valued signal is buffered to form the analysis window with 60 msec length, and
wherein each of the first complex valued signal and the second complex valued signal is down sampled by a factor 4.
