kaldi.ivector¶

Functions

`agglomerative_cluster`	Calls C++ function
`compute_vad_energy`	Calls C++ function
`estimate_ivectors_online`	Obtains periodically an estimate of the iVector including all frames up to that point.

Classes

`AgglomerativeClusterer`	CLIF wrapper for ::kaldi::AgglomerativeClusterer
`AhcCluster`	CLIF wrapper for ::kaldi::AhcCluster
`IvectorEstimationOptions`	Options for estimating iVectors, during both trainning and test.
`IvectorExtractor`	CLIF wrapper for ::kaldi::IvectorExtractor
`IvectorExtractorEstimationOptions`	CLIF wrapper for ::kaldi::IvectorExtractorEstimationOptions
`IvectorExtractorOptions`	CLIF wrapper for ::kaldi::IvectorExtractorOptions
`IvectorExtractorStats`	CLIF wrapper for ::kaldi::IvectorExtractorStats
`IvectorExtractorStatsOptions`	CLIF wrapper for ::kaldi::IvectorExtractorStatsOptions
`IvectorExtractorUtteranceStats`	Stats for a particular utterance, i.e., the sufficient stats for estimating an iVector
`LogisticRegression`	CLIF wrapper for ::kaldi::LogisticRegression
`LogisticRegressionConfig`	CLIF wrapper for ::kaldi::LogisticRegressionConfig
`OnlineIvectorEstimationStats`	CLIF wrapper for ::kaldi::OnlineIvectorEstimationStats
`Plda`	CLIF wrapper for ::kaldi::Plda
`PldaConfig`	CLIF wrapper for ::kaldi::PldaConfig
`PldaEstimationConfig`	CLIF wrapper for ::kaldi::PldaEstimationConfig
`PldaStats`	CLIF wrapper for ::kaldi::PldaStats
`PldaUnsupervisedAdaptor`	CLIF wrapper for ::kaldi::PldaUnsupervisedAdaptor
`PldaUnsupervisedAdaptorConfig`	CLIF wrapper for ::kaldi::PldaUnsupervisedAdaptorConfig
`VadEnergyOptions`	CLIF wrapper for ::kaldi::VadEnergyOptions

class kaldi.ivector.AgglomerativeClusterer¶

CLIF wrapper for ::kaldi::AgglomerativeClusterer

cluster() → list<int>¶: Calls C++ function void ::kaldi::AgglomerativeClusterer::Cluster(::std::vector< ::int32>*)

class kaldi.ivector.AhcCluster¶

CLIF wrapper for ::kaldi::AhcCluster

id¶: C++ ::int32 AhcCluster.id

parent1¶: C++ ::int32 AhcCluster.parent1

parent2¶: C++ ::int32 AhcCluster.parent2

size¶: C++ ::int32 AhcCluster.size

utt_ids¶: C++ ::std::vector< ::int32> AhcCluster.utt_ids

class kaldi.ivector.IvectorEstimationOptions¶

Options for estimating iVectors, during both trainning and test.

acoustic_weight¶: C++ double IvectorEstimationOptions.acoustic_weight

max_count¶: C++ double IvectorEstimationOptions.max_count

register(opts:OptionsItf)¶: Calls C++ function void ::kaldi::IvectorEstimationOptions::Register(::kaldi::OptionsItf *)

class kaldi.ivector.IvectorExtractor¶

CLIF wrapper for ::kaldi::IvectorExtractor

feat_dim() → int¶: Calls C++ function int ::kaldi::IvectorExtractor::FeatDim()

get_acoustic_auxf_gconst(utt_stats:IvectorExtractorUtteranceStats) → float¶: Returns the part of the acoustic auxf that relates to the gconsts of the Gaussian.

get_acoustic_auxf_mean(utt_stats:IvectorExtractorUtteranceStats, mean:DoubleVectorBase, var:DoubleSpMatrix=default) → float¶: Returns just the part of the acoustic auxf that relates to the speaker-dependent means

get_acoustic_auxf_variance(utt_stats:IvectorExtractorUtteranceStats) → float¶: Returns just the part of the acoustic auxf that relates to the variance of the utt_stats.

get_acoustic_auxf_weight(utt_stats:IvectorExtractorUtteranceStats, mean:DoubleVectorBase, var:DoubleSpMatrix=default) → float¶: Returns the part of the acoustic auxf that relates to the Gaussian-specific weights.

get_auxf(utt_stats:IvectorExtractorUtteranceStats, mean:DoubleVectorBase, var:DoubleSpMatrix=default) → float¶: Returns the data-dependent part of the log-likelihood objective function, summed over frames.

get_ivector_dist_mean(utt_stats:IvectorExtractorUtteranceStats, linear:DoubleVectorBase, quadratic:DoubleSpMatrix)¶: Get the linear and quadratic terms in the distribution over iVectors

get_ivector_dist_prior(utt_stats:IvectorExtractorUtteranceStats, linear:DoubleVectorBase, quadratic:DoubleSpMatrix)¶: Gets the linear and quadratic terms in the distribution over iVectors that arise from the prior.

get_ivector_dist_weight(utt_stats:IvectorExtractorUtteranceStats, mean:DoubleVectorBase, linear:DoubleVectorBase, quadratic:DoubleSpMatrix)¶: Gets the linear and quadratic terms in the distribution over iVectors that arise from the weights

get_ivector_distribution(utt_stats:IvectorExtractorUtteranceStats, mean:DoubleVectorBase, var:DoubleSpMatrix)¶

Gets the distribution over ivectors (or the Gaussian approximation).

Parameters:	utt_stats (`IvectorExtractorUtteranceStats`) – stats for a particular utterance mean (`kaldi.matrix.VectorBase`) – output means ( (var) – class::`kaldi.matrix.packed.SpMatrix`): None if not needed, else must be the correct dimension (`ivector_dim()`)

get_prior_auxf(mean:DoubleVectorBase, var:DoubleSpMatrix=default) → float¶: Returns the prior-related part of the log-likelihood objective function.

ivector_dependent_weights() → bool¶: Calls C++ function bool ::kaldi::IvectorExtractor::IvectorDependentWeights()

ivector_dim() → int¶: Calls C++ function int ::kaldi::IvectorExtractor::IvectorDim()

new_with_params(opts:IvectorExtractorOptions, fgmm:FullGmm) → IvectorExtractor¶: Calls C++ function std::unique_ptr<::kaldi::IvectorExtractor> ::kaldi::IvectorExtractor::IvectorExtractor(::kaldi::IvectorExtractorOptions, ::kaldi::FullGmm)

num_gauss() → int¶: Calls C++ function int ::kaldi::IvectorExtractor::NumGauss()

prior_offset() → float¶: Offset of first dimension.

read(os:istream, binary:bool)¶: Calls C++ function void ::kaldi::IvectorExtractor::Read(::std::basic_istream<char, ::std::char_traits<char> >, bool)

write(os:ostream, binary:bool)¶: Calls C++ function void ::kaldi::IvectorExtractor::Write(::std::basic_ostream<char, ::std::char_traits<char> >, bool)

class kaldi.ivector.IvectorExtractorEstimationOptions¶

CLIF wrapper for ::kaldi::IvectorExtractorEstimationOptions

diagonalize¶: C++ bool IvectorExtractorEstimationOptions.diagonalize

gaussian_min_count¶: C++ double IvectorExtractorEstimationOptions.gaussian_min_count

num_threads¶: C++ ::int32 IvectorExtractorEstimationOptions.num_threads

register(opts:OptionsItf)¶: Calls C++ function void ::kaldi::IvectorExtractorEstimationOptions::Register(::kaldi::OptionsItf *)

variance_floor_factor¶: C++ double IvectorExtractorEstimationOptions.variance_floor_factor

class kaldi.ivector.IvectorExtractorOptions¶

CLIF wrapper for ::kaldi::IvectorExtractorOptions

ivector_dim¶: C++ int IvectorExtractorOptions.ivector_dim

num_iters¶: C++ int IvectorExtractorOptions.num_iters

register(opts:OptionsItf)¶: Calls C++ function void ::kaldi::IvectorExtractorOptions::Register(::kaldi::OptionsItf *)

use_weights¶: C++ bool IvectorExtractorOptions.use_weights

class kaldi.ivector.IvectorExtractorStats¶

CLIF wrapper for ::kaldi::IvectorExtractorStats

acc_stats_for_utterance(extractor:IvectorExtractor, feats:MatrixBase, post:list<list<tuple<int, float>>>)¶: Calls C++ function void ::kaldi::IvectorExtractorStats::AccStatsForUtterance(::kaldi::IvectorExtractor, ::kaldi::MatrixBase<float>, ::kaldi::Posterior)

add(other:IvectorExtractorStats)¶: Calls C++ function void ::kaldi::IvectorExtractorStats::Add(::kaldi::IvectorExtractorStats)

auxf_per_frame() → float¶: Calls C++ function double ::kaldi::IvectorExtractorStats::AuxfPerFrame()

ivector_variance_diagnostic(extractor:IvectorExtractor)¶: Prints the proportion of the variance explained by the Ivector model versus the Gaussians.

new(extractor:IvectorExtractor, stats_opts:IvectorExtractorStatsOptions) → IvectorExtractorStats¶: Calls C++ function std::unique_ptr<::kaldi::IvectorExtractorStats> ::kaldi::IvectorExtractorStats::IvectorExtractorStats(::kaldi::IvectorExtractor, ::kaldi::IvectorExtractorStatsOptions)

read(os:istream, binary:bool)¶: Calls C++ function void ::kaldi::IvectorExtractorStats::Read(::std::basic_istream<char, ::std::char_traits<char> >, bool)

update(opts:IvectorExtractorEstimationOptions, extractor:IvectorExtractor) → float¶: Returns the objf improvement per frame

write(os:ostream, binary:bool)¶: Calls C++ function void ::kaldi::IvectorExtractorStats::Write(::std::basic_ostream<char, ::std::char_traits<char> >, bool)

class kaldi.ivector.IvectorExtractorStatsOptions¶

CLIF wrapper for ::kaldi::IvectorExtractorStatsOptions

cache_size¶: C++ int IvectorExtractorStatsOptions.cache_size

compute_auxf¶: C++ bool IvectorExtractorStatsOptions.compute_auxf

num_samples_for_weights¶: C++ ::int32 IvectorExtractorStatsOptions.num_samples_for_weights

register(opts:OptionsItf)¶: Calls C++ function void ::kaldi::IvectorExtractorStatsOptions::Register(::kaldi::OptionsItf *)

update_variances¶: C++ bool IvectorExtractorStatsOptions.update_variances

class kaldi.ivector.IvectorExtractorUtteranceStats¶

Stats for a particular utterance, i.e., the sufficient stats for estimating an iVector

acc_stats(feats:MatrixBase, post:list<list<tuple<int, float>>>)¶: Calls C++ function void ::kaldi::IvectorExtractorUtteranceStats::AccStats(::kaldi::MatrixBase<float>, ::kaldi::Posterior)

new_with_params(num_gauss:int, feat_dim:int, need_2nd_order_stats:bool) → IvectorExtractorUtteranceStats¶: Calls C++ function std::unique_ptr<::kaldi::IvectorExtractorUtteranceStats> ::kaldi::IvectorExtractorUtteranceStats::IvectorExtractorUtteranceStats(int, int, bool)

num_frames() → float¶: Calls C++ function double ::kaldi::IvectorExtractorUtteranceStats::NumFrames()

scale(scale:float)¶: Calls C++ function void ::kaldi::IvectorExtractorUtteranceStats::Scale(double)

class kaldi.ivector.LogisticRegression¶

CLIF wrapper for ::kaldi::LogisticRegression

get_log_posteriors_matrix(xs:Matrix) → Matrix¶: Calculates the log posterior of the class label given the input xs

get_log_posteriors_vector(x:Vector) → Vector¶: Calculates the log posterior of the class label given the input x.

read(os:istream, binary:bool)¶: Calls C++ function void ::kaldi::LogisticRegression::Read(::std::basic_istream<char, ::std::char_traits<char> >, bool)

scale_priors(prior_scales:Vector)¶: Calls C++ function void ::kaldi::LogisticRegression::ScalePriors(::kaldi::Vector<float>)

train(xs:Matrix, ys:list<int>, conf:LogisticRegressionConfig)¶: xs and ys are the trainning data. Each row of xs is a vector corresponding to the class label in the same row of ys.

write(os:ostream, binary:bool)¶: Calls C++ function void ::kaldi::LogisticRegression::Write(::std::basic_ostream<char, ::std::char_traits<char> >, bool)

class kaldi.ivector.LogisticRegressionConfig¶

CLIF wrapper for ::kaldi::LogisticRegressionConfig

max_steps¶: C++ ::int32 LogisticRegressionConfig.max_steps

mix_up¶: C++ ::int32 LogisticRegressionConfig.mix_up

normalizer¶: C++ double LogisticRegressionConfig.normalizer

power¶: C++ double LogisticRegressionConfig.power

register(opts:OptionsItf)¶: Calls C++ function void ::kaldi::LogisticRegressionConfig::Register(::kaldi::OptionsItf *)

class kaldi.ivector.OnlineIvectorEstimationStats¶

CLIF wrapper for ::kaldi::OnlineIvectorEstimationStats

acc_stats(extractor:IvectorExtractor, feature:VectorBase, gauss_post:list<tuple<int, float>>)¶: Calls C++ function void ::kaldi::OnlineIvectorEstimationStats::AccStats(::kaldi::IvectorExtractor, ::kaldi::VectorBase<float>, ::std::vector< ::std::pair< ::int32, ::kaldi::BaseFloat> >)

acc_stats_sequence(extractor:IvectorExtractor, features:MatrixBase, gauss_post:list<list<tuple<int, float>>>)¶: Calls C++ function void ::kaldi::OnlineIvectorEstimationStats::AccStats(::kaldi::IvectorExtractor, ::kaldi::MatrixBase<float>, ::std::vector< ::std::vector< ::std::pair< ::int32, ::kaldi::BaseFloat> > >)

count() → float¶: Calls C++ function double ::kaldi::OnlineIvectorEstimationStats::Count()

get_ivector(num_cg_iters:int, ivector:DoubleVectorBase)¶: Gets the current estimate of the iVector.

ivector_dim() → int¶: Calls C++ function int ::kaldi::OnlineIvectorEstimationStats::IvectorDim()

new_with_other(other:OnlineIvectorEstimationStats) → OnlineIvectorEstimationStats¶: Calls C++ function std::unique_ptr<::kaldi::OnlineIvectorEstimationStats> ::kaldi::OnlineIvectorEstimationStats::OnlineIvectorEstimationStats(::kaldi::OnlineIvectorEstimationStats)

new_with_params(ivector_dim:int, prior_offset:float, max_count:float) → OnlineIvectorEstimationStats¶: Calls C++ function std::unique_ptr<::kaldi::OnlineIvectorEstimationStats> ::kaldi::OnlineIvectorEstimationStats::OnlineIvectorEstimationStats(int, float, float)

num_frames() → float¶: Calls C++ function double ::kaldi::OnlineIvectorEstimationStats::NumFrames()

objf_change(ivector:DoubleVectorBase) → float¶: Returns the change in objective function per frame from using the default value.

prior_offset() → float¶: Calls C++ function double ::kaldi::OnlineIvectorEstimationStats::PriorOffset()

read(os:istream, binary:bool)¶: Calls C++ function void ::kaldi::OnlineIvectorEstimationStats::Read(::std::basic_istream<char, ::std::char_traits<char> >, bool)

scale(scale:float)¶: Calls C++ function void ::kaldi::OnlineIvectorEstimationStats::Scale(double)

write(os:ostream, binary:bool)¶: Calls C++ function void ::kaldi::OnlineIvectorEstimationStats::Write(::std::basic_ostream<char, ::std::char_traits<char> >, bool)

class kaldi.ivector.Plda¶

CLIF wrapper for ::kaldi::Plda

apply_transform(in_transform:DoubleMatrix)¶: Calls C++ function void ::kaldi::Plda::ApplyTransform(::kaldi::Matrix<double>)

dim() → int¶: Calls C++ function int ::kaldi::Plda::Dim()

from_other(other:Plda) → Plda¶: Calls C++ function std::unique_ptr<::kaldi::Plda> ::kaldi::Plda::Plda(::kaldi::Plda)

log_likelihood_ratio(transformed_train_ivector:DoubleVectorBase, num_train_utts:int, transformed_test_ivector:DoubleVectorBase) → float¶: Calls C++ function double ::kaldi::Plda::LogLikelihoodRatio(::kaldi::VectorBase<double>, int, ::kaldi::VectorBase<double>)

read(os:istream, binary:bool)¶: Calls C++ function void ::kaldi::Plda::Read(::std::basic_istream<char, ::std::char_traits<char> >, bool)

smooth_within_class_covariance(smoothing_factor:float)¶: Calls C++ function void ::kaldi::Plda::SmoothWithinClassCovariance(double)

transform_ivector(config:PldaConfig, ivector:DoubleVectorBase, num_examples:int, transformed_ivector:DoubleVectorBase) → float¶: Calls C++ function double ::kaldi::Plda::TransformIvector(::kaldi::PldaConfig, ::kaldi::VectorBase<double>, int, ::kaldi::VectorBase<double> *)

transform_ivector_single(config:PldaConfig, ivector:VectorBase, num_examples:int, transformed_ivector:VectorBase) → float¶: Calls C++ function float ::kaldi::Plda::TransformIvector(::kaldi::PldaConfig, ::kaldi::VectorBase<float>, int, ::kaldi::VectorBase<float> *)

write(os:ostream, binary:bool)¶: Calls C++ function void ::kaldi::Plda::Write(::std::basic_ostream<char, ::std::char_traits<char> >, bool)

class kaldi.ivector.PldaConfig¶

CLIF wrapper for ::kaldi::PldaConfig

normalize_length¶: C++ bool PldaConfig.normalize_length

register(opts:OptionsItf)¶: Calls C++ function void ::kaldi::PldaConfig::Register(::kaldi::OptionsItf *)

simple_length_norm¶: C++ bool PldaConfig.simple_length_norm

class kaldi.ivector.PldaEstimationConfig¶

CLIF wrapper for ::kaldi::PldaEstimationConfig

num_em_iters¶: C++ ::int32 PldaEstimationConfig.num_em_iters

register(opts:OptionsItf)¶: Calls C++ function void ::kaldi::PldaEstimationConfig::Register(::kaldi::OptionsItf *)

class kaldi.ivector.PldaStats¶

CLIF wrapper for ::kaldi::PldaStats

add_samples(weight:float, group:DoubleMatrix)¶: Calls C++ function void ::kaldi::PldaStats::AddSamples(double, ::kaldi::Matrix<double>)

dim() → int¶: Calls C++ function int ::kaldi::PldaStats::Dim()

init(dim:int)¶: Calls C++ function void ::kaldi::PldaStats::Init(int)

is_sorted() → bool¶: Calls C++ function bool ::kaldi::PldaStats::IsSorted()

sort()¶: Calls C++ function void ::kaldi::PldaStats::Sort()

class kaldi.ivector.PldaUnsupervisedAdaptor¶

CLIF wrapper for ::kaldi::PldaUnsupervisedAdaptor

add_double_stats(weight:float, ivector:DoubleVector)¶: Calls C++ function void ::kaldi::PldaUnsupervisedAdaptor::AddStats(double, ::kaldi::Vector<double>)

add_stats(weight:float, ivector:Vector)¶: Calls C++ function void ::kaldi::PldaUnsupervisedAdaptor::AddStats(double, ::kaldi::Vector<float>)

update_plda(config:PldaUnsupervisedAdaptorConfig, plda:Plda)¶: Calls C++ function void ::kaldi::PldaUnsupervisedAdaptor::UpdatePlda(::kaldi::PldaUnsupervisedAdaptorConfig, ::kaldi::Plda *)

class kaldi.ivector.PldaUnsupervisedAdaptorConfig¶

CLIF wrapper for ::kaldi::PldaUnsupervisedAdaptorConfig

between_covar_scale¶: C++ ::kaldi::BaseFloat PldaUnsupervisedAdaptorConfig.between_covar_scale

mean_diff_scale¶: C++ ::kaldi::BaseFloat PldaUnsupervisedAdaptorConfig.mean_diff_scale

register(opts:OptionsItf)¶: Calls C++ function void ::kaldi::PldaUnsupervisedAdaptorConfig::Register(::kaldi::OptionsItf *)

within_covar_scale¶: C++ ::kaldi::BaseFloat PldaUnsupervisedAdaptorConfig.within_covar_scale

class kaldi.ivector.VadEnergyOptions¶

CLIF wrapper for ::kaldi::VadEnergyOptions

register(opts:OptionsItf)¶: Calls C++ function void ::kaldi::VadEnergyOptions::Register(::kaldi::OptionsItf *)

vad_energy_mean_scale¶: C++ ::kaldi::BaseFloat VadEnergyOptions.vad_energy_mean_scale

vad_energy_threshold¶: C++ ::kaldi::BaseFloat VadEnergyOptions.vad_energy_threshold

vad_frames_context¶: C++ ::int32 VadEnergyOptions.vad_frames_context

vad_proportion_threshold¶: C++ ::kaldi::BaseFloat VadEnergyOptions.vad_proportion_threshold

kaldi.ivector.agglomerative_cluster(costs:Matrix, thresh:float, min_clust:int) → list<int>¶: Calls C++ function void ::kaldi::AgglomerativeCluster(::kaldi::Matrix<float>, float, int, ::std::vector< ::int32>*)

kaldi.ivector.compute_vad_energy(opts:VadEnergyOptions, input_features:MatrixBase) → Vector¶: Calls C++ function void ::kaldi::ComputeVadEnergy(::kaldi::VadEnergyOptions, ::kaldi::MatrixBase<float>, ::kaldi::Vector<float>*)

kaldi.ivector.estimate_ivectors_online(feats:Matrix, post:list<list<tuple<int, float>>>, extractor:IvectorExtractor, ivector_period:int, num_cg_iters:int, max_count:float) -> (objf_improvement:float, ivectors:Matrix)¶: Obtains periodically an estimate of the iVector including all frames up to that point.