Search

CN-121983062-A - Ultralow-power-consumption voice wake-up system with self-adaptive threshold adjustment capability

CN121983062ACN 121983062 ACN121983062 ACN 121983062ACN-121983062-A

Abstract

The invention relates to the technical field of voice recognition and audio signal processing, in particular to an ultra-low power consumption voice awakening system with self-adaptive Threshold adjustment capability, which comprises an input voice processing module and an energy trend detection module, wherein the input voice processing module monitors and receives an audio signal x (t) from a microphone to output an envelope signal E n (t), the self-adaptive control module updates the Gain of a programmable amplifier according to the envelope signal E n (t), the programmable amplifier amplifies the audio signal x (t) by adopting the updated Gain to obtain a signal x ' ' (t), the signal x ' ' (t) is subjected to a band-pass filter to obtain a signal V ' BPF (t), and a Threshold judgment and awakening output module self-adaptively adjusts a awakening Threshold according to the envelope signal E n (t) to output a awakening mark digital signal.

Inventors

  • WANG SHUO
  • LI ZEHONG
  • REN MIN
  • WANG YAHAO
  • CHEN CAI
  • WANG XUAN

Assignees

  • 重庆邮电大学
  • 电子科技大学重庆微电子产业技术研究院

Dates

Publication Date
20260505
Application Date
20260123

Claims (6)

  1. 1. The ultra-low power consumption voice awakening system with the self-adaptive threshold adjustment capability is characterized by comprising an input sound processing module, an energy trend detection module, a self-adaptive control module and a threshold judgment and awakening output module, wherein the input sound processing module and the energy trend detection module comprise a programmable amplifier, a band-pass filter, a full-wave rectifier and a low-pass filter; ultra-low power consumption voice wake-up the system workflow comprises: The input sound processing module and the energy trend detection module monitor and receive the audio signal x (t) from the microphone and output the envelope signal E n (t) of the current frame; The self-adaptive control module updates the Gain of the programmable amplifier according to the envelope signal E n (t); The programmable amplifier amplifies the audio signal x (t) by adopting the updated Gain to obtain a signal x ' ' (t), and the signal x ' ' (t) is subjected to a band-pass filter to obtain a signal V ' BPF (t); The Threshold judging and waking output module adaptively adjusts a waking Threshold according to the envelope signal E n (t), compares the adjusted waking Threshold with the signal V ' BPF (t), and generates a waking flag digital signal with a single period or a preset width effective for triggering the main system to switch from a dormant state to a high-performance voice processing state if the signal V' BPF (t) is larger than the adjusted waking Threshold and the minimum continuous jitter elimination time is met.
  2. 2. The ultra-low power consumption voice wake-up system with adaptive threshold adjustment capability of claim 1, wherein the processing procedure of the input sound processing module and the energy trend detecting module comprises: The programmable amplifier receives an audio signal x (t) from the microphone and amplifies the audio signal according to the Gain to obtain a signal x' (t); The signal x' (t) passes through a band-pass filter to filter high-frequency noise, and meanwhile, a frequency band which is relevant to voice in the range of 200 Hz to 6 kHz is reserved to obtain a signal V BPF (t); The full-wave rectifier converts the signal V BPF (t) from an alternating current signal to a direct current envelope to obtain a signal V rect (t); The signal V rect (t) is smoothed by a low-pass filter to obtain the envelope signal E n (t) of the current frame.
  3. 3. The ultra-low power consumption voice wakeup system with adaptive threshold adjustment capability of claim 2, wherein the envelope signal E n (T) is updated according to a fixed frame length T frame , and each update is implemented by using an exponential sliding average method, expressed as: E n (t) = α · E n (t-1) + (1-α) · V rect (t), Where α represents a smoothing coefficient, which is determined by the fixed frame length T frame and a time constant preset by the system.
  4. 4. The ultra-low power consumption voice wakeup system with adaptive threshold adjustment capability of claim 1, wherein the adaptive control module operates as follows: The window comparator receives the envelope signal E n (t) of the current frame, compares the envelope signal E n (t) with a preset fixed reference sound pressure level E set and outputs a Trend signal Trend n of the current frame; The dynamic counter receives a Trend signal Trend n of the current frame, calculates an integral count value C n =C n-1 +Trend n ,C n-1 of the current frame to represent an integral count value of a previous frame; the Gain lookup table module updates the Gain of the programmable amplifier according to the integral count value C n of the current frame.
  5. 5. The ultra-low power consumption voice wake-up system with adaptive threshold adjustment capability according to claim 4, wherein the window comparator judges a Trend of background noise according to a relation between an envelope signal E n (t) of a current frame and a preset fixed reference sound pressure level E set , and outputs a Trend signal Trend n of the current frame, and the processing procedure comprises: When the E n (t)>E set + [ delta ] E is detected, the current background noise is judged to be in an ascending Trend, and a Trend signal Trend n =1 of the current frame; When E set -△E≤E n (t)≤E set + [ delta ] E, determining that the current background noise is stable, and the Trend signal Trend n =0 of the current frame; when E n (t)<E set -delta E, judging that the current background noise is in a descending Trend, and the Trend signal Trend n = -1 of the current frame; where Δe represents the absolute value of the difference between the preset fixed reference sound pressure level E set and the window comparator limit.
  6. 6. The ultra-low power consumption voice wakeup system with adaptive Threshold adjustment capability according to claim 1, wherein the Threshold determination and wakeup output module adaptively adjusts the wakeup Threshold according to the envelope signal E n (t) includes: Sequentially performing clipping processing, inter-frame smoothing and linear mapping on the envelope signal E n (t) to obtain a background NOISE reference level NOISE_REF, wherein the background NOISE reference level NOISE_REF is used for representing the average energy level of the background NOISE of the current environment; a wakeup Threshold is calculated according to the background NOISE reference level NOISE_REF: Threshold = NOISE_REF × (1 + K) + Threshold_OFFSET; where K represents a configurable noise margin coefficient and threshold_OFFSET represents a fixed bias term.

Description

Ultralow-power-consumption voice wake-up system with self-adaptive threshold adjustment capability Technical Field The invention relates to the technical field of voice recognition and audio signal processing, in particular to an ultra-low power consumption voice awakening system with self-adaptive threshold adjustment capability. Background In recent years, with rapid development in the fields of smart home, wearable devices, etc., voice activation technology has become a key component in portable electronic devices (such as smart watches, bluetooth headsets, remote controllers, etc.). In order to reduce the power consumption of the system and prolong the battery endurance time, a wake-up circuit which always keeps monitoring but has extremely low power consumption is widely researched and practically applied. The current voice wake-up system mostly adopts a fixed sound pressure threshold method, the method presets a fixed sound pressure threshold value based on the environmental noise amplitude, and the system is triggered to enter a wake-up state only when an input signal exceeds the threshold value. Such techniques are typically implemented by analog threshold detection or comparator circuits, which only maintain low power consumption trigger module operation during operation, while the main microphone circuit is in a powered-off state (e.g., a fixed threshold voice wake-up scheme described in patent document voice activity detection method and system (application number CN105810201 a)). However, the fixed threshold based wake-up approach suffers from the following significant drawbacks: First, it lacks the ability to adapt to the environment. The fixed sound pressure threshold cannot be adaptively adjusted according to the ambient noise variation. In a scene with stronger noise, the system is easy to wake up by mistake, and in a very quiet environment, wake-up omission can occur, so that the user experience is unstable. For example, in a noisy traffic environment, a fixed threshold may cause frequent false triggers, while in a quiet office, it is difficult for the system to recognize valid voice commands. Second, existing schemes still suffer from deficiencies in terms of power consumption. Although the system power consumption is lower in the sleep mode, the main system still needs to be kept working in the wake-up and sound monitoring phases, so that the overall energy consumption is higher. Particularly in the continuous listening state, it is difficult to realize long standby time. For example, the wake-up system proposed in the patent "real-time voice wake-up audio device, operation method, apparatus and storage medium" (CN 110312235A) needs to maintain high sensitivity listening, and especially consumes power in complex noise environments. Finally, it is difficult to achieve intelligent decisions while compromising environmental adaptability and power consumption in the prior art. Most of the prior schemes are based on simple sound pressure detection, and cannot introduce intelligent judging mechanisms such as frequency domain analysis, envelope detection or feature extraction, so that the system cannot effectively distinguish background noise from target voice signals, and the recognition accuracy is limited. In complex acoustic environments, problems may arise with inaccuracy in voice recognition. In summary, the existing fixed sound pressure threshold wake-up scheme still has the defects of false wake-up, missed wake-up, higher power consumption, slow response and the like in various application scenes. Therefore, a mechanism capable of dynamically adjusting the threshold according to the environmental noise is needed to realize accurate voice wake-up on the premise of maintaining extremely low power consumption, thereby improving the reliability and user experience of the voice wake-up system. Disclosure of Invention Aiming at the problems of weak environmental adaptability, frequent false wake-up and missed wake-up phenomena, higher power consumption and the like caused by a fixed sound pressure threshold in the existing voice wake-up technology, the invention provides an ultra-low power consumption voice wake-up system with self-adaptive threshold adjustment capability, which comprises an input sound processing module, an energy trend detection module, a self-adaptive control module and a threshold judgment and wake-up output module, wherein the input sound processing module and the energy trend detection module comprise a programmable amplifier, a band-pass filter, a full-wave rectifier and a low-pass filter; ultra-low power consumption voice wake-up the system workflow comprises: The input sound processing module and the energy trend detection module monitor and receive the audio signal x (t) from the microphone and output the envelope signal E n (t) of the current frame; The self-adaptive control module updates the Gain of the programmable amplifier according to the envelope signal E n (