# US 10,504,539 B2 · Voice activity detection systems and methods > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 10,504,539 B2 - Document ID: US10504539B2 - Status: Granted - Kind: B2 - Application number: 15/832,709 - Filed: 2017-12-05 - Published: 2019-06-06 - Granted: 2019-12-10 - Earliest claimed: 2017-12-05 - Assignees: Synaptics Incorporated - Applicants: Synaptics Incorporated - Inventors: Saeed Mosayyebpour Kaskari; Francesco Nesta - Web page: https://patentgrep.com/patent/US10504539B2 ## CPC - G10L 25/78: Detection of presence or absence of voice signals (switching of direction of transmission by voice frequency in two-way loud-speaking telephone systems H04M9/10) - G10L 15/02: Feature extraction for speech recognition; Selection of recognition unit - G10L 15/16: using artificial neural networks - G10L 15/22: Procedures used during a speech recognition process, e.g. man-machine dialogue - G10L 25/30: using neural networks - G10L 25/18: the extracted parameters being spectral information of each sub-band - G10L 25/21: the extracted parameters being power information - G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING ## Abstract An audio processing device or method includes an audio transducer operable to receive audio input and generate an audio signal based on the audio input. The audio processing device or method also includes an audio signal processor operable to extract local features from the audio signal, such as Power-Normalized Coefficients (PNCC) of the audio signal. The audio signal processor also is operable to extract global features from the audio signal, such as chroma features and harmonicity features. A neural network is provided to determine a probability that a target audio is present in the audio signal based on the local and global features. In particular, the neural network is trained to output a value indicating whether the target audio is present and locally dominant in the audio signal. ## Claims ### Claim 1 An audio processing device comprising: an audio sensor operable to receive audio input and generate an audio signal based on the audio input; and an audio signal processor comprising: a local feature extraction component operable to extract local features from the audio signal; a global feature extraction component operable to extract global features from the audio signal, wherein the global feature extraction component comprises a spectrum smoothing component operable to generate a smoothed spectrum of the audio signal, a chroma feature component operable to determine chroma features based on the smoothed spectrum, and a harmonicity feature component operable to determine harmonicity features based on the smoothed spectrum; and a neural network operable to determine a probability that a target audio is present in the audio signal based on the local and global features; wherein the chroma feature component comprises a chroma transformation matrix operable to perform a chroma transformation, a derivative component operable to estimate a derivative of chroma vector, and an energy component operable to determine an energy of the chroma vector; and wherein the harmonicity feature component is operable to determine Harmonic Additive Spectrum (HAS) features and Harmonic Difference Spectrum (HDS) features. ### Claim 2 (depends on 1) The audio processing device of claim 1, wherein the target audio is associated with a person's voice. ### Claim 3 (depends on 1) The audio processing device of claim 1, wherein the audio signal processor further comprises: a sub-band analysis component operable to transform the audio signal in sub-bands in frequency domain; and a magnitude determining component operable to determine magnitudes of the sub-bands in frequency domain. ### Claim 4 (depends on 1) The audio processing device of claim 1, wherein the local feature extraction component comprises: a Gammatone filter bank operable to determine Power-Normalized Coefficients (PNCC) of the audio signal, and a delta component operable to determine delta features and double delta features from the PNCC. ### Claim 5 (depends on 4) The audio processing device of claim 4, wherein the local feature extraction component is operable to store the PNCC, the delta features, and the double delta features as the local features of the audio signal. ### Claim 6 (depends on 1) The audio processing device of claim 1, wherein the neural network is operable to output a binary value indicating whether the target audio is present in the audio signal. ### Claim 7 A method for processing audio signals, comprising: receiving audio input and generating an audio signal based on the audio input by an audio transducer; extracting local features from the audio signal; extracting global features from the audio signal by generating a smoothed spectrum of the audio signal, determining chroma features based on the smoothed spectrum, and determining harmonicity features based on the smoothed spectrum; and determining, by a neural network, a probability that a target audio is present in the audio signal based on the local and global features wherein the determining chroma features comprises performing a chroma transformation by chroma transformation matrix, estimating a derivative of chroma vector; and determining an energy of the chroma vector; and wherein the determining harmonicity features comprises determining Harmonic Additive Spectrum (HAS) features and Harmonic Difference Spectrum (HDS) features. ### Claim 8 (depends on 7) The method of claim 7, wherein the target audio is associated with a person's voice. ### Claim 9 (depends on 7) The method of claim 7, further comprising: transforming the audio signal in sub-bands in frequency domain; and determining magnitudes of the sub-bands in frequency domain. ### Claim 10 (depends on 7) The method of claim 7, wherein the extracting local features comprises: determining Power-Normalized Coefficients (PNCC) of the audio signal using a Gammatone filter bank, and determining delta features and double delta features from the PNCC. ### Claim 11 (depends on 10) The method of claim 10, wherein the extracting local features comprises storing the PNCC, the delta features, and the double delta features as the local features of the audio signal. ### Claim 12 (depends on 7) The method of claim 7, further comprising determining by the neural network, a binary value indicating whether the target audio is present in the audio signal. ### Claim 13 (depends on 7) The method of claim 7, further comprising training the neural network by updating weights and biases of the neural network based on an output error of a forward pass. ### Claim 14 (depends on 7) The method of claim 7, further comprising estimating a ground-truth garget value for training the neural network.