Splitting and normalizing method and device, audio feature extractor and chip

Abstract

The invention relates to a splitting and normalizing method and device, an audio feature extractor and a chip. In order to improve the robustness of an audio front end to background noise, splitting and normalizing processing is carried out on each channel. The method comprises the following steps that: a threshold is generated according to the average pulse number of an input pulse sequence in an average window; and in each clock period, whether an integral distribution counting module is enabled to count is judged, if yes, the integral distribution counting module starts to count, and when the count value of the integral distribution counting module reaches the threshold, an output pulse is generated, and counting is started again. Compared with the prior art, according to the technical scheme, an expensive and complex splitting and normalizing method is improved, the structure is simpler, implementation is easier, the precision is higher, and the device and the chip applying the splitting and normalizing processing have better statistical performance and lower cost and power consumption.

Description

technical field [0001] The invention relates to a splitting and normalization method, device, audio feature extractor, and chip, and in particular to a method, device, audio feature extractor, and chip for splitting and normalizing background noise during audio signal processing. Background technique [0002] The audio signal processing module in the chip usually uses the audio front end (Audio front end, AFE) to process the original audio signal, and extracts audio features, and the extracted audio features are encoded and identified by a classifier (for example, a pulse neural network SNN). Such as figure 1 As shown, it can be used for always-on keyword spotting (keyword spotting, KWS), voice activity detection (voice activity detect, VAD), vibration anomaly detection (Vibration anomaly detection), smart agriculture, animal wear, environmental sound detection, etc. [0003] The implementation principle of the audio front-end using analog signal processing (Analog signal p...

AI Technical Summary

Problems solved by technology

However, such split-normalized DN processing by SNN requires observation and storage of instantaneous power information over a long period of time, which is very expensive and difficult to do in practice

Images