begin/end point detection of speech More...

Functions
void	spr_bed_free (SprSspInfo *Info)

int	spr_bed_setup (SprSspInfo Info, const char descript, void aux_info)

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

void	spr_bed_reset (SprSspInfo Info, SprSspStatus action)

Detailed Description

begin/end point detection of speech

This program removes the leading and trailing noise frames from a data stream.

[online_bed]
`nfr_lead <number>(20)`
Number of extra leading noise frames.
`nfr_tail <number>(20)`
Number of extra trailing noise frames.
`spch_win <number>(60)`
Window used to detect a speech fragment.
`min_spch <number>(30)`
Min. number of speech frames in the speech window.
`sil_win <number>(60)`
Window used to detect silence fragments.
`max_spch <number>(10)`
Max. number of speech frames in the silence window.
`clean_sil_spch`
Return clean silence speech codes instead of removing frames.
`trigger_pos`
Return the trigger positions only (nfr with respect to the previous state).
`method <sent/trunc/glob>(glob)`
Determines how bed works. The <sent> mode work in a continuous mode, e.g. marking sentences in a dialog. The <trunc> mode will detect the first speech segment only (the rest will be classified as noise). The <glob> method will remove leading and trailing noise in a sentence, intermediate noise will be marked as speech to.
`sil_range <min_range>(0.0) <max_range>(0.0)`
Build in SpchDet: The <min_range> parameter is needed for signals that are almost noise free, and gives a lower bound on the amplitude of the noise. The <max_range> gives an upper bound on the amplitude of the noise, and gives more robustness if the speech detector is started in a speech segment.
`spch_raise <raise>(0.25)`
Build in SpchDet: minimum raise in the speech level (averaged over 4 consecutive frames) needed to detect the very first word.
`spch_range <range>(20.0)`
Build in SpchDet: signals (amplitude) below the average speech level divided by <range> are also treated as noise.

Author: Kris Demuynck

Date: 12 September 1995

Functions

Detailed Description