Audio bandwidth expansion method and system based on genetic algorithm optimization model parameters

Abstract

The invention discloses an audio bandwidth extension method based on a genetic algorithm to optimize model parameters. The method includes: step 1) preprocessing an input audio signal x(n) to obtain a filtered signal; and step 2) modulating and overlapping the filtered signal Transform to obtain low-frequency modulation overlapping transform coefficients; step 3) divide the low-frequency modulation overlapping transform coefficients into subbands, calculate the root mean square energy of each subband, and obtain the low-frequency spectrum envelope sequence; step 4) according to the low-frequency spectrum envelope sequence, use gray The model GM(1,1) estimates the high-frequency sub-band energy of the audio signal to obtain the high-frequency spectrum envelope; step 5) uses spectrum replication, spectrum folding, nonlinear calculation, comprehensive multi-band excitation or nonlinear prediction methods to The details of the audio frequency spectrum are expanded to obtain the details of the high-frequency spectrum; step 6) restore the high-frequency spectrum information of the audio signal x(n) according to the above-mentioned high-frequency spectrum envelope and the details of the high-frequency spectrum; step 7) use the modulation overlap inverse transformation to realize Bandwidth extension of x(n).

Description

technical field [0001] The present invention relates to the technical field of audio bandwidth extension, and more specifically, the present invention relates to an audio bandwidth extension method and system based on genetic algorithm optimization model parameters. Background technique [0002] According to people's actual demand for audio quality, the bandwidth of the signal reproduced by digital audio technology will be limited to a certain extent. For this reason, audio signals can be classified into four types: narrowband, wideband, ultrawideband, and fullband. The effective bandwidth of the full-band signal covers the entire frequency range (20Hz to 20kHz) that the human ear can perceive. This type of audio is mainly used for high-fidelity reproduction of music signals, and the required network transmission rate and terminal computing power are relatively high. Traditional telephone voice usually uses narrowband signals, its frequency distribution is in the range of 3...

AI Technical Summary

Problems solved by technology

However, the current traditional audio bandwidth expansion methods generally use traditional statistical models to expand the spectrum envelope, rather than studying the change trend of the audio signal's high and low frequency spectrum envelope itself, resulting in a more uniform and natural high frequency portion of the expansion. Poor, with low differentiation between different audio extensions

Images