feat_s Struct Reference

Data Fields
int	refcount
	Reference count.
char *	name
	Printable name for this feature type.
int32	cepsize
	Size of input speech vector (typically, a cepstrum vector).
int32	n_stream
	Number of feature streams; e.g., 4 in Sphinx-II.
int32 *	stream_len
	Vector length of each feature stream.
int32	window_size
	Number of extra frames around given input frame needed to compute corresponding output feature (so total = window_size*2 + 1).
int32	n_sv
	Number of subvectors.
int32 *	sv_len
	Vector length of each subvector.
int32 **	subvecs
	Subvector specification (or NULL for none).
mfcc_t *	sv_buf
	Temporary copy buffer for subvector projection.
int32	sv_dim
	Total dimensionality of subvector (length of sv_buf).
cmn_type_t	cmn
	Type of CMN to be performed on each utterance.
int32	varnorm
	Whether variance normalization is to be performed on each utt; Irrelevant if no CMN is performed.
agc_type_t	agc
	Type of AGC to be performed on each utterance.
void(*	compute_feat )(struct feat_s *fcb, mfcc_t **input, mfcc_t **feat)
	Feature computation function.
cmn_t *	cmn_struct
	Structure that stores the temporary variables for cepstral means normalization.
agc_t *	agc_struct
	Structure that stores the temporary variables for acoustic gain control.
mfcc_t **	cepbuf
	Circular buffer of MFCC frames for live feature computation.
mfcc_t **	tmpcepbuf
	Array of pointers into cepbuf to handle border cases.
int32	bufpos
	Write index in cepbuf.
int32	curpos
	Read index in cepbuf.
mfcc_t ***	lda
	Array of linear transformations (for LDA, MLLT, or whatever).
uint32	n_lda
	Number of linear transformations in lda.
uint32	out_dim
	Output dimensionality.

---

Detailed Description

Definition at line 155 of file feat.h.

---

Field Documentation

void(* feat_s::compute_feat)(struct feat_s *fcb, mfcc_t **input, mfcc_t **feat)

Feature computation function.

Parameters:

	fcb	the feat_t describing this feature type
	input	pointer into the input cepstra
	feat	a 2-d array of output features (n_stream x stream_len)

Returns:

0 if successful, -ve otherwise.

Function for converting window of input speech vector (input[-window_size..window_size]) to output feature vector (feat[stream][]). If NULL, no conversion available, the speech input must be feature vector itself.

---

The documentation for this struct was generated from the following file:

• include/feat.h

---