This class includes configuration variables relating to the online iVector extraction, but not including configuration for the "base feature", i.e. More...

#include <online-ivector-feature.h>

Collaboration diagram for OnlineIvectorExtractionConfig:

[legend]

Public Member Functions
	OnlineIvectorExtractionConfig ()

void	Register (OptionsItf *opts)

Public Attributes
std::string	lda_mat_rxfilename

std::string	global_cmvn_stats_rxfilename

std::string	splice_config_rxfilename

std::string	cmvn_config_rxfilename

bool	online_cmvn_iextractor

std::string	diag_ubm_rxfilename

std::string	ivector_extractor_rxfilename

int32	ivector_period

int32	num_gselect

BaseFloat	min_post

BaseFloat	posterior_scale

BaseFloat	max_count

int32	num_cg_iters

bool	use_most_recent_ivector

bool	greedy_ivector_extractor

BaseFloat	max_remembered_frames

Detailed Description

This class includes configuration variables relating to the online iVector extraction, but not including configuration for the "base feature", i.e.

MFCC/PLP/filterbank, which is an input to this feature. This configuration class can be used from the command line, but before giving it to the code we create a config class called OnlineIvectorExtractionInfo which contains the actual configuration classes as well as various objects that are needed. The principle is that any code should be callable from other code, so we didn't want to force configuration classes to be read from disk.

Definition at line 55 of file online-ivector-feature.h.

Constructor & Destructor Documentation

◆ OnlineIvectorExtractionConfig()

OnlineIvectorExtractionConfig ( )

inline

Definition at line 105 of file online-ivector-feature.h.

                                  : online_cmvn_iextractor(false),
                                    ivector_period(10), num_gselect(5),
                                    min_post(0.025), posterior_scale(0.1),
                                    max_count(0.0), num_cg_iters(15),
                                    use_most_recent_ivector(true),
                                    greedy_ivector_extractor(false),
                                    max_remembered_frames(1000) { }

Member Function Documentation

◆ Register()

void Register ( OptionsItf * opts )

inline

Definition at line 113 of file online-ivector-feature.h.

References OptionsItf::Register().

Referenced by main(), and OnlineSilenceWeightingConfig::RegisterWithPrefix().

                                   {
     opts->Register("lda-matrix", &lda_mat_rxfilename, "Filename of LDA matrix, "
                    "e.g. final.mat; used for iVector extraction. ");
     opts->Register("global-cmvn-stats", &global_cmvn_stats_rxfilename,
                    "(Extended) filename for global CMVN stats, used in iVector "
                    "extraction, obtained for example from "
                    "'matrix-sum scp:data/train/cmvn.scp -', only used for "
                    "iVector extraction");
     opts->Register("cmvn-config", &cmvn_config_rxfilename, "Configuration "
                    "file for online CMVN features (e.g. conf/online_cmvn.conf),"
                    "only used for iVector extraction.  Contains options "
                    "as for the program 'apply-cmvn-online'");
     opts->Register("online-cmvn-iextractor", &online_cmvn_iextractor,
                    "add online-cmvn to feature pipeline of ivector extractor, "
                    "use the cmvn setup from the UBM.  Note: the default of "
                    "false is what we historically used; we'd use true if "
                    "we were using CMVN'ed features for the neural net.");
     opts->Register("splice-config", &splice_config_rxfilename, "Configuration file "
                    "for frame splicing (--left-context and --right-context "
                    "options); used for iVector extraction.");
     opts->Register("diag-ubm", &diag_ubm_rxfilename, "Filename of diagonal UBM "
                    "used to obtain posteriors for iVector extraction, e.g. "
                    "final.dubm");
     opts->Register("ivector-extractor", &ivector_extractor_rxfilename,
                    "Filename of iVector extractor, e.g. final.ie");
     opts->Register("ivector-period", &ivector_period, "Frequency with which "
                    "we extract iVectors for neural network adaptation");
     opts->Register("num-gselect", &num_gselect, "Number of Gaussians to select "
                    "for iVector extraction");
     opts->Register("min-post", &min_post, "Threshold for posterior pruning in "
                    "iVector extraction");
     opts->Register("posterior-scale", &posterior_scale, "Scale for posteriors in "
                    "iVector extraction (may be viewed as inverse of prior scale)");
     opts->Register("max-count", &max_count, "Maximum data count we allow before "
                    "we start scaling the stats down (if nonzero)... helps to make "
                    "iVectors from long utterances look more typical.  Interpret "
                    "as a frame-count times --posterior-scale, typically 1/10 of "
                    "a number of frames.  Suggest 100.");
     opts->Register("use-most-recent-ivector", &use_most_recent_ivector, "If true, "
                    "always use most recent available iVector, rather than the "
                    "one for the designated frame.");
     opts->Register("greedy-ivector-extractor", &greedy_ivector_extractor, "If "
                    "true, 'read ahead' as many frames as we currently have available "
                    "when extracting the iVector.  May improve iVector quality.");
     opts->Register("max-remembered-frames", &max_remembered_frames, "The maximum "
                    "number of frames of adaptation history that we carry through "
                    "to later utterances of the same speaker (having a finite "
                    "number allows the speaker adaptation state to change over "
                    "time).  Interpret as a real frame count, i.e. not a count "
                    "scaled by --posterior-scale.");
   }