Method and equipment for target sound detection
By adopting a multi-level target sound detector in the audio context detection system, using a low-complexity binary classifier and a high-complexity multi-objective sound classifier, the problem of high power consumption of the audio detection system in the prior art is solved, and the target of low power consumption is achieved when detecting a large number of target sounds.
Patent Information
- Application Number
- CN202180024648.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2020-04-01
- Filing Date
- 2021-03-01
- Publication Date
- 2025-03-25
- Estimated Expiration
- 2041-03-01
AI Technical Summary
Existing audio context detection systems lead to relatively large power consumption when continuously scanning audio inputs to detect sound events, especially in mobile devices, which reduce battery life, and system complexity and power consumption increase with the number of detected sound events.
Multi-stage target sound detectors are used, including the first and second stages. The first level uses a low-complex binary target sound classifier to perform binary classification of audio data, and activates the second level in response to detection of the target sound. The second level uses a more powerful multi-objective sound classifier for further analysis and classification.
By using a low-power first-level binary classifier in normal operation and selectively activating a high-power but high-performance second-level classifier, it is possible to use relatively low average power when detecting relatively large amounts of target sounds of interest, thereby reducing overall power consumption.