# US 12,603,098 B2 · Automatic gain control method and apparatus for voice interaction system, and system > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,603,098 B2 - Document ID: US12603098B2 - Status: Granted - Kind: B2 - Application number: 18/293,643 - Filed: 2022-07-29 - Published: 2025-04-10 - Granted: 2026-04-14 - Earliest claimed: 2021-07-30 - Assignees: Espressif Systems (Shanghai) Co., Ltd. - Applicants: Espressif Systems (Shanghai) Co., Ltd. - Inventors: Xiangyu Sun - Web page: https://patentgrep.com/patent/US12603098B2 ## CPC - G10L 21/0272: Voice signal separating - G10L 21/034: Automatic adjustment - G10L 15/08: Speech classification or search - G10L 15/22: Procedures used during a speech recognition process, e.g. man-machine dialogue - G10L 15/28: Constructional details of speech recognition systems - G10L 21/028: using properties of sound source - G10L 21/0364: for improving intelligibility - G10L 15/063: {Training} - G10L 15/285: {Memory allocation or algorithm optimisation to reduce hardware requirements} - G10L 2015/088: {Word spotting} - 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 The present application discloses an automatic gain control method and apparatus for a voice interaction system, a system, and a computer-readable storage medium. The method comprises: receiving a voice signal; dividing the received voice signal into a plurality of sound signals of different sound sources by blind source separation; invoking a pre-trained wake-up word detection model to perform wake-up word detection on the sound signals of different sound sources respectively, to determine a sound signal in which a wake-up word is detected; calculating, based on the sound signal in which the wake-up word is detected, a gain required for automatic gain control of the voice interaction system, to adjust a volume of picked-up audio data within a preset range. ## Claims ### Claim 1 An automatic gain control method for a voice interaction system, comprising: receiving a voice signal; dividing the received voice signal into a plurality of sound signals of different sound sources by blind source separation; invoking a pre-trained wake-up word detection model to perform wake-up word detection on the sound signals of different sound sources respectively, to determine a sound signal in which a wake-up word is detected; and for each of the plurality of sound signals of different sound sources, using the sound signal in which the wake-up word is detected as an adjustment reference for performing automatic gain control, and calculating a gain required for automatic gain control of the voice interaction system, to adjust a volume of picked-up audio data within a preset range. ### Claim 2 (depends on 1) The automatic gain control method according to claim 1, wherein after the dividing the received voice signal into a plurality of sound signals of different sound sources by blind source separation, the method further comprises: maintaining a corresponding FIFO queue for each of the sound signals of different sound sources; wherein a value of a length N of the FIFO queue is determined based on a length of the wake-up word and a frame time of the wake-up word detection; and reading each of the sound signals of different sound sources on a frame-by-frame basis, calculating a maximum absolute value of an amplitude value of the signal for each frame, and saving the maximum absolute value in the FIFO queue. ### Claim 3 (depends on 2) The automatic gain control method according to claim 2, wherein a product of the length N of the FIFO queue and the frame time of the wake-up word detection is equal to a length of the wake-up word. ### Claim 4 (depends on 2) The automatic gain control method according to claim 2, wherein the calculating a gain required for automatic gain control of the voice interaction system comprises: calculating a maximum value Amax in the FIFO queue corresponding to the sound signal in which the wake-up word is detected, and according to G = { A ref A max , A max < A ref 1 , A max ≥ A ref calculating a gain G required for a picked-up audio device in the voice interaction system; wherein Aref is an amplitude value of a reference audio signal, and the reference audio signal is a fixed value. ### Claim 5 (depends on 4) The automatic gain control method according to claim 4, wherein after the calculating a gain required for automatic gain control of the voice interaction system, the method further comprises: adjusting an analog gain to a required gain G; if an adjustable range Ga of the analog gain is less than the required gain G, then adjusting the analog gain to maximum and adjusting a digital gain to G-Ga. ### Claim 6 (depends on 5) The automatic gain control method according to claim 5, further comprising: restoring the gain to an initial gain Ginit after detecting the completion of the voice interaction. ### Claim 7 (depends on 6) The automatic gain control method according to claim 6, wherein the initial gain Ginit is: a gain corresponding to ensuring that the audio data picked up by a picked-up audio device is not saturated when the reference audio signal value is determined. ### Claim 8 (depends on 1) The automatic gain control method according to claim 1, wherein after the dividing the received voice signal into a plurality of sound signals of different sound sources by blind source separation, the method further comprises: maintaining a corresponding FIFO queue for each of the sound signals of different sound sources; wherein a value of a length N of the FIFO queue is determined based on a length of the wake-up word and a frame time of the wake-up word detection; and reading each of the sound signals of different sound sources on a frame-by-frame basis, calculating an absolute value of an amplitude value of the signal for each frame, and performing a convolution operation on the absolute value by using a Gaussian window, and saving the maximum value in the FIFO queue after the convolution operation. ### Claim 9 (depends on 8) The automatic gain control method according to claim 8, wherein a product of the length N of the FIFO queue and the frame time of the wake-up word detection is equal to a length of the wake-up word. ### Claim 10 (depends on 8) The automatic gain control method according to claim 8, wherein the calculating a gain required for automatic gain control of the voice interaction system comprises: calculating a maximum value Amax in the FIFO queue corresponding to the sound signal in which the wake-up word is detected, and according to G = { A ref A max , A max < A ref 1 , A max ≥ A ref calculating a gain G required for a picked-up audio device in the voice interaction system; wherein Aref is an amplitude value of a reference audio signal, and the reference audio signal is a fixed value. ### Claim 11 (depends on 10) The automatic gain control method according to claim 10, wherein after the calculating a gain required for automatic gain control of the voice interaction system, the method further comprises: adjusting an analog gain to a required gain G; if an adjustable range Ga of the analog gain is less than the required gain G, then adjusting the analog gain to maximum and adjusting a digital gain to G−Ga. ### Claim 12 (depends on 11) The automatic gain control method according to claim 11, further comprising: restoring the gain to an initial gain Ginit after detecting the completion of the voice interaction. ### Claim 13 (depends on 12) The automatic gain control method according to claim 12, wherein the initial gain Ginit is: a gain corresponding to ensuring that the audio data picked up by a picked-up audio device is not saturated when the reference audio signal value is determined. ### Claim 14 (depends on 1) The automatic gain control method according to claim 1, wherein the wake-up word detection model is a model obtained by pre-training using delta-LFBE as a feature. ### Claim 15 A voice interaction system, comprising: a memory and a processor; wherein the memory is configured to store instructions; the processor, when invoking the instructions, executes an automatic gain control method for the voice interaction system, the method comprising: receiving a voice signal; dividing the received voice signal into a plurality of sound signals of different sound sources by blind source separation; invoking a pre-trained wake-up word detection model to perform wake-up word detection on the sound signals of different sound sources respectively, to determine a sound signal in which a wake-up word is detected; and for each of the plurality of sound signals of different sound sources, using the sound signal in which the wake-up word is detected as an adjustment reference for performing automatic gain control, calculating a gain required for automatic gain control of the voice interaction system, to adjust a volume of picked-up audio data within a preset range. ### Claim 16 A non-transitory computer-readable storage medium comprising instructions that, when executed, implement an automatic gain control method for a voice interaction system, the method comprising: receiving a voice signal; dividing the received voice signal into a plurality of sound signals of different sound sources by blind source separation; invoking a pre-trained wake-up word detection model to perform wake-up word detection on the sound signals of different sound sources respectively, to determine a sound signal in which a wake-up word is detected; and for each of the plurality of sound signals of different sound sources, using the sound signal in which the wake-up word is detected as an adjustment reference for performing automatic gain control, calculating a gain required for automatic gain control of the voice interaction system, to adjust a volume of picked-up audio data within a preset range.