Multitask Noisy Speech Enhancement System
- Speech band equalizer
- Dynamics processing
- Noise gate
- Signal level limiter
- Clipping restoration
- Noise reduction
- Noise whitening
- Blind deconvolution
- Spectrum analyser
- Time stretching
- Spectral expander
- Fourier corrector
- Neural network corrector
- Joint approximation
- Homomorphic approximation
Speech band equalizer
The speech band equalizer modifies the spectrum of the speech signal using the filter bank. The speech frequency range is divided into a number of frequency bands and the signal may be amplified or attenuated in each of these bands independently. One of two sets of the frequency bands may be chosen:
The concept of speech critical bands is based on the selection of narrow frequency bands in speech signal spectrum which may be treated as the elementary signals equivalent to the tones. The center frequencies Fc and bandwidths ΔF of the critical bands (according to the definition) are as follows.
Three versions of filter responses may be selected:
The other method uses equally-contributing bands according to French and Steinberg. The frequency range of the speech signal is divided into 20 bands. Each of these bands equally contributes to the total speech intelligibility. The filter bank has flat frequency response in this method.
The user selects one of the four methods of equalization in the window of equalizer module. The gain of each filter may be adjusted using the sliders (-90 dB to 24 dB) or entered directly using the keyboard. The current frequency responses of the filters are presented on the plot. The equalizer module works on-line - each change in filter bank settings is immediately applied and the user can hear the result. The settings of the filter bank may be saved as a preset and restored later.
It is recommended to use the filter bank with flat frequency response (equally-contributing bands or critical bands with modified bandwidth). Before the equalization, spectral analysis of the sound file should be performed in order to find the frequency bands containing the distortions (and to reduce filter gain in these bands during the equalization).
|© 2004 Multimedia Systems Department, Gdansk University of Technology and Air Force Academy in Deblin|