A low-power mfcc speech feature extraction circuit based on serial fft

Abstract

The invention discloses a low-power consumption MFCC speech feature extraction circuit based on serial FFT, and belongs to the technical field of calculation, calculation or counting. The circuit is oriented to the field of intelligence. By optimizing the MFCC algorithm to adapt to the hardware circuit design, making full use of the serial FFT algorithm and the approximate operation of the multiplication operation, the circuit area and power consumption are greatly reduced. The circuit as a whole includes: preprocessing module, framing and windowing module, FFT module, Mel filter module, logarithm and DCT module, the improved FFT algorithm adopts serial pipeline to process data, and effectively utilizes the time of audio frame to meet In the case of output requirements, the storage area and the number of operations of the circuit are reduced. At the same time, the pre-emphasis module uses shift operations instead of multiplication and addition operations, and the logarithmic operation uses a look-up table, which further reduces the area and power consumption of the circuit.

Description

technical field [0001] The invention discloses a low power consumption MFCC speech feature extraction circuit based on serial FFT, relates to the field of speech feature signal processing and integrated circuit design, and belongs to the technical field of calculation, calculation or counting. Background technique [0002] With the gradual development of intelligence in the current era, the importance of voice interaction in the field of intelligence has become more and more important, and voice interaction devices have also begun to play an important role in our lives, such as voice assistants for smartphones, access control systems, and in-vehicle systems. . The key to voice interaction is voice recognition, and voice feature extraction is the key technology of voice recognition. The low-loss extraction of audio feature signals with less power consumption can effectively improve the recognition accuracy of the voice recognition system. [0003] At present, traditional spe...

AI Technical Summary

Problems solved by technology

The algorithm finally obtains the result of the audio through preprocessing of the audio, frame-by-frame windowing, Fast Fourier Transform (FFT, Fast FourierTransform), Mel logarithmic operation, and discrete cosine transform (DCT, Discrete Cosine Transform), but The traditional MFCC algorithm has poor adaptability to hardware. Among them, the large number of multiplication and addition operations and storage area required by the FFT operation process and Mel filter operation increase the difficulty and cost of hardware implementation.

Images