Normalisation of the SNR of a (mel) spectrum. More...

Functions
int	spr_noise_masking_setup (SprSspInfo Info, const char descript, void aux_info)

int	spr_noise_masking_process (SprSspInfo Info, const void frame_in, void *frame_out)

void	spr_noise_masking_reset (SprSspInfo Info, SprSspStatus action)

void	spr_noise_masking_free (SprSspInfo *Info)

Detailed Description

Normalisation of the SNR of a (mel) spectrum.

This presprocessing module adds noise to each spectral parameter (filterbank output) until the target SNR for that parameter is reached. Some noise is added to the SNR as to avoid a (nearly) hard flooring of the parameter.

Parameter [i] is calculated as follows (in energy):
  spec[i] += mean_energy[i] /
             exp(M_LN10 * (SNR[i] + SNRvar[i]*(rand()-0.5)) / 10)

SNR[i] and SNRvar[i] are given by the user. The values of the mean_energy of the parameters are calculated over the first input frames of the input file. They are updated for each new input frame, using a weight.

[noise_masking]
`nfr <number>`
Number of frames used for initial mean_energy calculation. Set it to -1 if all frames should be used. Default 100. Overruled if a file with means is given.
`file <fname>`
Starting mean_enery values are given in a file.
`alpha <number>`
Weight for mean_energy update, default 0.05.
`no_reset`
Do not reset the mean spectrum at the beginning of a new file.
`spdet_in <copy/move> <buf_name>`
Input from a silence speech detector: the spectral mean is only update for speech frames.
`snr <coeff1> ...`
Define the target SNR for each spectral parameter. If only one coefficient is given, it is used for all parameters. Default 30 dB.
`snr_var <coeff1> ...`
Define the variance on the target SNR for each spectral parameter. If only one coefficient is given, it is used for all parameters. Default 3 dB.
`db_input`
Input and output in dB, default the parameters are given as energy. Calculations are based on energy values.

Author: Jacques Duchateau

Date: 26 April 2000

Functions

Detailed Description