# US 12,014,730 B2 · Voice processing method, electronic device, and storage medium > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,014,730 B2 - Document ID: US12014730B2 - Status: Granted - Kind: B2 - Application number: 17/322,238 - Filed: 2021-05-17 - Published: 2022-05-26 - Granted: 2024-06-18 - Earliest claimed: 2020-11-23 - Assignees: Beijing Xiaomi Mobile Software Co., Ltd. - Applicants: Beijing Xiaomi Mobile Software Co., Ltd. - Inventors: Xiangyan Xu - Web page: https://patentgrep.com/patent/US12014730B2 ## CPC - G10L 15/04: Segmentation; Word boundary detection - G10L 15/20: Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech (G10L21/02 takes precedence) - G10L 15/02: Feature extraction for speech recognition; Selection of recognition unit - G10L 15/08: Speech classification or search - 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 2015/025: {Phonemes, fenemes or fenones being the recognition units} - G10L 2015/088: {Word spotting} - G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING ## Abstract A voice processing method includes: collecting a voice signal by a microphone of an electronic device, and signal-processing the collected voice signal to obtain a first voice frame segment; performing voice recognition on the first voice frame segment to obtain a first recognition result; in response to the first recognition result not matching a target content and a plurality of tokens in the first recognition result meeting a preset condition, performing frame compensation on the first voice frame segment to obtain a second voice frame segment; and performing voice recognition on the second voice frame segment to obtain a second recognition result. A matching degree between the second recognition result and the target content is greater than a matching degree between the first recognition result and the target content. ## Claims ### Claim 1 A voice processing method, comprising: collecting a voice signal by a microphone of an electronic device, and performing voice activity detection (VAD) processing of the collected voice signal to obtain a first voice frame segment; detecting whether a phoneme is a filler phoneme starting from the first phoneme of the first voice frame segment, wherein the filler phoneme indicates non-keywords; in response to determining that a probability of the phoneme being the filler phoneme is greater than a probability of the phoneme being not the filler phoneme, skipping the filler phoneme and performing voice recognition on the first voice frame segment to obtain a first recognition result; in response to the first recognition result not matching a target content and a number of tokens in the first recognition result whose matching probability to the target content is greater than a second set threshold exceeding a preset number, determining that there is a VAD processing truncation in the first voice frame segment; in response to the VAD processing truncation in the first voice frame segment, estimating a target frame length compensated for the first voice frame segment according to a length of historical target content counted, determining a next voice frame segment in the collected voice signal adjacent to the first voice frame segment, obtaining, in the collected voice signal, a third voice frame segment having the target frame length and including complete token units, from a start position of the next voice frame segment, and obtaining a second voice frame segment by splicing the third voice frame segment behind the first voice frame segment; and performing voice recognition on the second voice frame segment to obtain a second recognition result, wherein a matching degree between the second recognition result and the target content is greater than a matching degree between the first recognition result and the target content. ### Claim 2 (depends on 1) The method of claim 1, wherein obtaining the second voice frame segment comprises: obtaining a fourth voice frame segment having a set frame length and including complete token units from a start position of the next voice frame segment, and obtaining the second voice frame segment by splicing the fourth voice frame segment behind the first voice frame segment. ### Claim 3 (depends on 1) The method of claim 1, wherein a unit of each token of the plurality of tokens comprises at least one of a word, a phone, a monophone and a triphone. ### Claim 4 An electronic device, comprising: a microphone configured to collect a voice signal; a processor; and a memory storing instructions that when executed by the processor, control the processor to perform voice activity detection (VAD) processing of the collected voice signal to obtain a first voice frame segment; detect whether a phoneme is a filler phoneme starting from the first phoneme of the first voice frame segment, wherein the filler phoneme indicates non-keywords; in response to determining that a probability of the phoneme being the filler phoneme is greater than a probability of the phoneme being not the filler phoneme, skip the filler phoneme and perform voice recognition on the first voice frame segment to obtain a first recognition result; in response to the first recognition result not matching a target content and a number of tokens in the first recognition result whose matching probability to the target content is greater than a second set threshold exceeding a preset number, determine that there is a VAD processing truncation in the first voice frame segment; in response to the VAD processing truncation in the first voice frame segment, estimate a target frame length compensated for the first voice frame segment according to a length of historical target content counted, determine a next voice frame segment in the collected voice signal adjacent to the first voice frame segment, obtain, in the collected voice signal, a third voice frame segment having the target frame length and including complete token units, from a start position of the next voice frame segment, and obtain a second voice frame segment by splicing the third voice frame segment behind the first voice frame segment; and perform voice recognition on the second voice frame segment to obtain a second recognition result, wherein a matching degree between the second recognition result and the target content is greater than a matching degree between the first recognition result and the target content. ### Claim 5 (depends on 4) The device of claim 4, wherein the processor is further configured to: obtain a fourth voice frame segment having a set frame length and including complete token units from a start position of the next voice frame segment, and obtain the second voice frame segment by splicing the fourth voice frame segment behind the first voice frame segment. ### Claim 6 (depends on 4) The device of claim 4, wherein a unit of each token of the plurality of tokens comprises at least one of a word, a phone, a monophone and a triphone. ### Claim 7 A non-transitory computer-readable storage medium having instructions stored thereon, when the instructions are executed by a processor in an electronic device, a voice processing method is implemented, the method comprising: collecting a voice signal by a microphone of an electronic device, and performing voice activity detection (VAD) processing of the collected voice signal to obtain a first voice frame segment; detecting whether a phoneme is a filler phoneme starting from the first phoneme of the first voice frame segment, wherein the filler phoneme indicates non-keywords; in response to determining that a probability of the phoneme being the filler phoneme is greater than a probability of the phoneme being not the filler phoneme, skipping the filler phoneme and performing voice recognition on the first voice frame segment to obtain a first recognition result; in response to the first recognition result not matching a target content and a number of tokens in the first recognition result whose matching probability to the target content is greater than a second set threshold exceeding a preset number, determining that there is a VAD processing truncation in the first voice frame segment; in response to the VAD processing truncation in the first voice frame segment, estimating a target frame length compensated for the first voice frame segment according to a length of historical target content counted, determining a next voice frame segment in the collected voice signal adjacent to the first voice frame segment, obtaining, in the collected voice signal, a third voice frame segment having the target frame length and including complete token units, from a start position of the next voice frame segment, and obtaining a second voice frame segment by splicing the third voice frame segment behind the first voice frame segment; and performing voice recognition on the second voice frame segment to obtain a second recognition result, wherein a matching degree between the second recognition result and the target content is greater than a matching degree between the first recognition result and the target content. ### Claim 8 (depends on 7) The method of claim 7, wherein obtaining the second voice frame segment comprises: obtaining a fourth voice frame segment having a set frame length and including complete token units from a start position of the next voice frame segment, and obtaining the second voice frame segment by splicing the fourth voice frame segment behind the first voice frame segment. ### Claim 9 (depends on 7) The method of claim 7, wherein a unit of each token of the plurality of tokens comprises at least one of a word, a phone, a monophone and a triphone.