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
- Decorrelation
- Joint approximation
- Homomorphic approximation
- Reverberation
- Synchronisation
- Normalisation
Contact info


Processing of the distorted signal using various digital signal processing algorithm may result in unnatural sounding of speech. A simple way of enhancing the signal quality is to add reverb to the signal. The reverberation module should be used in the final stage of the restoration process, after the signal is processed by all other modules.

The reverberation module performs a convolution of the signal with the impulse response of a room providing good listening conditions. The processed signal is a sum of the original signal and all reflected signal replicas (delayed and attenuated):

where x(n) is the original signal, y(n) - processed signal, d(k) delay of k-th reflection, w(k) - amplitude of k-th reflection, K - number of reflections.

Three parameters may be set by the user: first reflection delay (1-100 ms, default 20 ms), gain (0.1-0.9, default 0.5) and reverberation time (1-200 ms, default 200 ms). The signal is processed by the module in off-line mode.