kaldi.gmm¶

Functions

`augment_gmm_flags`	Calls C++ function
`get_split_targets`	Calls C++ function
`gmm_flags_to_string`	Calls C++ function
`mle_full_gmm_update`	Calls C++ function
`string_to_gmm_flags`	Calls C++ function
`string_to_sgmm_update_flags`	Calls C++ function
`string_to_sgmm_write_flags`	Calls C++ function

Classes

`AccumDiagGmm`	CLIF wrapper for ::kaldi::AccumDiagGmm
`AccumFullGmm`	CLIF wrapper for ::kaldi::AccumFullGmm
`DiagGmm`	Gaussian Mixture Model with diagonal covariances.
`FullGmm`	Python wrapper for Kaldi::FullGmm<Float>
`FullGmmNormal`	CLIF wrapper for ::kaldi::FullGmmNormal
`GmmUpdateFlags`	An enumeration.
`MapDiagGmmOptions`	CLIF wrapper for ::kaldi::MapDiagGmmOptions
`MleDiagGmmOptions`	CLIF wrapper for ::kaldi::MleDiagGmmOptions
`MleFullGmmOptions`	CLIF wrapper for ::kaldi::MleFullGmmOptions
`SgmmUpdateFlags`	An enumeration.
`SgmmWriteFlags`	An enumeration.

class kaldi.gmm.AccumDiagGmm¶

CLIF wrapper for ::kaldi::AccumDiagGmm

accumulate_for_component(data:VectorBase, comp_index:int, weight:float)¶: Calls C++ function void ::kaldi::AccumDiagGmm::AccumulateForComponent(::kaldi::VectorBase<float>, int, float)

accumulate_from_diag(gmm:DiagGmm, data:VectorBase, frame_posterior:float) → float¶: Calls C++ function float ::kaldi::AccumDiagGmm::AccumulateFromDiag(::kaldi::DiagGmm, ::kaldi::VectorBase<float>, float)

accumulate_from_diag_multi_threaded(gmm:DiagGmm, data:MatrixBase, frame_weights:VectorBase, num_threads:int) → float¶: Calls C++ function float ::kaldi::AccumDiagGmm::AccumulateFromDiagMultiThreaded(::kaldi::DiagGmm, ::kaldi::MatrixBase<float>, ::kaldi::VectorBase<float>, int)

accumulate_from_posteriors(data:VectorBase, gauss_posteriors:VectorBase)¶: Calls C++ function void ::kaldi::AccumDiagGmm::AccumulateFromPosteriors(::kaldi::VectorBase<float>, ::kaldi::VectorBase<float>)

add(scale:float, acc:AccumDiagGmm)¶: Calls C++ function void ::kaldi::AccumDiagGmm::Add(double, ::kaldi::AccumDiagGmm)

add_stats_for_component(comp_id:int, occ:float, x_stats:DoubleVectorBase, x2_stats:DoubleVectorBase)¶: Calls C++ function void ::kaldi::AccumDiagGmm::AddStatsForComponent(int, double, ::kaldi::VectorBase<double>, ::kaldi::VectorBase<double>)

assert_equal(other:AccumDiagGmm)¶: Calls C++ function void ::kaldi::AccumDiagGmm::AssertEqual(::kaldi::AccumDiagGmm)

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

flags_¶: C++ clif_type_56 AccumDiagGmm.Flags()

new(gmm:DiagGmm, flags:int) → AccumDiagGmm¶: Calls C++ function std::unique_ptr<::kaldi::AccumDiagGmm> ::kaldi::AccumDiagGmm::AccumDiagGmm(::kaldi::DiagGmm, unsigned short)

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

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

resize(num_gauss:int, dim:int, flags:int)¶: Calls C++ function void ::kaldi::AccumDiagGmm::Resize(int, int, unsigned short)

resize_with_gmm(gmm:DiagGmm, flags:int)¶: Calls C++ function void ::kaldi::AccumDiagGmm::Resize(::kaldi::DiagGmm, unsigned short)

scale(f:float, flags:int)¶: Calls C++ function void ::kaldi::AccumDiagGmm::Scale(float, unsigned short)

set_zero(flags:int)¶: Calls C++ function void ::kaldi::AccumDiagGmm::SetZero(unsigned short)

smooth_stats(tau:float)¶: Calls C++ function void ::kaldi::AccumDiagGmm::SmoothStats(float)

smooth_with_accum(tau:float, src_acc:AccumDiagGmm)¶: Calls C++ function void ::kaldi::AccumDiagGmm::SmoothWithAccum(float, ::kaldi::AccumDiagGmm)

smooth_with_model(tau:float, src_gmm:DiagGmm)¶: Calls C++ function void ::kaldi::AccumDiagGmm::SmoothWithModel(float, ::kaldi::DiagGmm)

with_other(other:AccumDiagGmm) → AccumDiagGmm¶: Calls C++ function std::unique_ptr<::kaldi::AccumDiagGmm> ::kaldi::AccumDiagGmm::AccumDiagGmm(::kaldi::AccumDiagGmm)

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

class kaldi.gmm.AccumFullGmm¶

CLIF wrapper for ::kaldi::AccumFullGmm

accumulate_for_component(data:VectorBase, comp_index:int, weight:float)¶: Calls C++ function void ::kaldi::AccumFullGmm::AccumulateForComponent(::kaldi::VectorBase<float>, int, float)

accumulate_from_diag(gmm:DiagGmm, data:VectorBase, frame_posterior:float) → float¶: Calls C++ function float ::kaldi::AccumFullGmm::AccumulateFromDiag(::kaldi::DiagGmm, ::kaldi::VectorBase<float>, float)

accumulate_from_full(gmm:FullGmm, data:VectorBase, frame_posterior:float) → float¶: Calls C++ function float ::kaldi::AccumFullGmm::AccumulateFromFull(::kaldi::FullGmm, ::kaldi::VectorBase<float>, float)

accumulate_from_posteriors(data:VectorBase, gauss_posteriors:VectorBase)¶: Calls C++ function void ::kaldi::AccumFullGmm::AccumulateFromPosteriors(::kaldi::VectorBase<float>, ::kaldi::VectorBase<float>)

covariance_accumulator() → list<DoubleSpMatrix>¶: Calls C++ function ::std::vector< ::kaldi::SpMatrix<double> > ::kaldi::AccumFullGmm::covariance_accumulator()

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

flags() → int¶: Calls C++ function unsigned short ::kaldi::AccumFullGmm::Flags()

mean_accumulator() → DoubleMatrix¶: Calls C++ function ::kaldi::Matrix<double> ::kaldi::AccumFullGmm::mean_accumulator()

new_with_full(gmm:FullGmm, flags:int) → AccumFullGmm¶: Calls C++ function std::unique_ptr<::kaldi::AccumFullGmm> ::kaldi::AccumFullGmm::AccumFullGmm(::kaldi::FullGmm, unsigned short)

new_with_other(gmm:AccumFullGmm) → AccumFullGmm¶: Calls C++ function std::unique_ptr<::kaldi::AccumFullGmm> ::kaldi::AccumFullGmm::AccumFullGmm(::kaldi::AccumFullGmm)

new_with_params(num_comp:int, dim:int, flags:int) → AccumFullGmm¶: Calls C++ function std::unique_ptr<::kaldi::AccumFullGmm> ::kaldi::AccumFullGmm::AccumFullGmm(int, int, unsigned short)

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

occupancy() → DoubleVector¶: Calls C++ function ::kaldi::Vector<double> ::kaldi::AccumFullGmm::occupancy()

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

resize(num_comp:int, dim:int, flags:int)¶: Calls C++ function void ::kaldi::AccumFullGmm::Resize(int, int, unsigned short)

resize_var_accumulator(num_comp:int, dim:int)¶: Calls C++ function void ::kaldi::AccumFullGmm::ResizeVarAccumulator(int, int)

resize_with_full(gmm:FullGmm, flags:int)¶: Calls C++ function void ::kaldi::AccumFullGmm::Resize(::kaldi::FullGmm, unsigned short)

scale(f:float, flags:int)¶: Calls C++ function void ::kaldi::AccumFullGmm::Scale(float, unsigned short)

set_zero(flags:int)¶: Calls C++ function void ::kaldi::AccumFullGmm::SetZero(unsigned short)

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

class kaldi.gmm.DiagGmm(nmix=0, dim=0)[source]¶

Gaussian Mixture Model with diagonal covariances.

Parameters:	nmix (int) – Number of Gaussians to mix dim (int) – Dimension of each component

Creates a new DiagGmm with specified number of gaussian mixtures: and dimensions.

Parameters:	nmix (int) – number of gaussian to mix dim (int) – dimension

component_log_likelihood(data:VectorBase, comp_id:int) → float¶

Computes the log-likelihood of a data point given a single Gaussian component.

Parameters:	data (`kaldi.matrix.Vector`) – data point comp_id – component id
Returns:	Log-likehood of input data point for a given component

component_posteriors(data)[source]¶

Computes the posterior probabilities of all Gaussian components given: a data point.

Parameters:	data (VectorBase) – Data point with the same dimension as each component.
Returns:	2-element tuple containing loglike (`float`): Log-likelihood posteriors (`Vector`): Vector with the posterior probabilities
Raises:	ValueError if data is not consistent with gmm dimension.

compute_gconsts() → int¶

Sets the gconsts.

Returns:	Number of gconsts that are invalid e.g. because of zero weights or variances.

copy(src)[source]¶

Copies data from src into this DiagGmm and returns this DiagGmm.

Parameters:	src (FullGmm or DiagGmm) – Source Gmm to copy
Returns:	This DiagGmm after update.

copy_from_diag(diaggmm:DiagGmm)¶: Copies from given DiagGmm

copy_from_full(fullgmm:FullGmm)¶: Copies from given FullGmm

dim() → int¶: Returns the dimensionality of the Gaussian mean vectors.

from_clusterable(gc:GaussClusterable, var_floor:float) → DiagGmm¶: Creates a new DiagGmm from a GaussClusterable.

from_nmix_dim(nmix:int, dim:int) → DiagGmm¶: Creates a new DiagGmm with given number of mixtures and dimension.

from_other(gmm:DiagGmm) → DiagGmm¶: Creates a new DiagGmm from another DiagGmm.

gaussian_selection(data:VectorBase, num_gselect:int) -> (log_like:float, output:list<int>)¶

Gets gaussian selection information for one frame.

Parameters:	data (`kaldi.matrix.Vector`) – data point num_gselect (int) – number of gaussians to select
Returns:	Log-likelihood for the input frame and the best `num_gselect` indices (sorted from best to worst likelihood).

gaussian_selection_matrix(data:MatrixBase, num_gselect:int) -> (log_like:float, output:list<list<int>>)¶

Gets gaussian selection information for a sequence of frames.

Parameters:	data (`kaldi.matrix.Matrix`) – sequence of data points num_gselect (int) – number of gaussians to select
Returns:	Log-likelihood for the input frame and the best `num_gselect` indices (sorted from best to worst likelihood) for each data point.

gaussian_selection_preselect(data:VectorBase, preselect:list<int>, num_gselect:int) -> (log_like:float, output:list<int>)¶

Get gaussian selection information for one frame.

Parameters:	data (`kaldi.matrix.Vector`) – data point preselect (list) – subset of mixture components num_gselect (int) – number of gaussians to select
Returns:	Log-likelihood for the input frame and the best `num_gselect` indices that were preselected (sorted from best to worst likelihood).

gconsts() → Vector¶: Returns gconsts

generate(output:VectorBase)¶

Generates a random data point from this distribution.

Parameters:	output (`kaldi.matrix.Vector`) – Output vector

get_component_mean(gauss:int, out:VectorBase)¶: Gets component mean.

get_component_variance(gauss:int, out:VectorBase)¶: Gets component variance.

get_means() → Matrix¶: Returns component means.

get_vars() → Matrix¶: Returns component variances.

interpolate(rho:float, source:DiagGmm, flags:int=default)¶

Interpolates this model with diagonal GMM

this <- rho * source + (1 - rho) * this

Parameters:	rho (float) – Interpolation weight source (DiagGmm) – Source model flags (int) – Interpolation flags

interpolate_full(rho:float, source:FullGmm, flags:int=default)¶

Interpolates this model with full GMM

this <- rho * source + (1 - rho) * this

Parameters:	rho (float) – Interpolation weight source (FullGmm) – Source model flags (int) – Interpolation flags

inv_vars() → Matrix¶: Returns inverse variances

log_likelihood(data:VectorBase) → float¶

Computes the log-likelihood for a data point

Parameters:	data (`kaldi.matrix.Vector`) – data point

log_likelihoods(data:VectorBase) → Vector¶

Computes the per-component log-likelihoods for a data point

Parameters:	data (`kaldi.matrix.Vector`) – data point

log_likelihoods_matrix(data:MatrixBase) → Matrix¶

Computes the per-component log-likelihoods for a sequence of data points

The row index of the input data matrix and the output log-likelihoods matrix is the frame index.

Parameters:	data (`kaldi.matrix.Matrix`) – sequence of data points

log_likelihoods_preselect(data:VectorBase, indices:list<int>) → Vector¶

Computes the per-component log-likelihoods of a subset of mixture components.

Parameters:	data (`kaldi.matrix.Vector`) – data point indices – list of indices

means_invvars() → Matrix¶: Returns means times inverse variances

merge(target_components:int) → list<int>¶

Merges components

Parameters:	target_components (int) – number of target components
Returns:	The order in which components were merged.

num_gauss() → int¶: Returns the number of mixture components.

perturb(perturb_factor:float)¶

Perturbs components

Component means are perturbed with a random vector multiplied by the pertrubation factor.

Parameters:	perturb_factor (float) – perturbation factor

read(is:istream, binary:bool)¶: Reads gaussian mixture model from input stream.

remove_component(gauss:int, renorm_weights:bool)¶: Removes single component from model.

remove_components(gauss:list<int>, renorm_weights:bool)¶: Removes multiple components from model.

resize(nmix:int, dim:int)¶: Resizes arrays to this dim. Does not initialize data.

set_component_inv_var(gauss:int, in:VectorBase)¶: Sets inv-var for a single component.

set_component_mean(gauss:int, in:VectorBase)¶

Sets mean for a single component.

Internally multiplies with inv-var.

set_component_weight(gauss:int, weight:float)¶: Sets mixture weight for a single component.

set_inv_vars(v:MatrixBase)¶: Sets inverse variances.

set_inv_vars_and_means(invvars:MatrixBase, means:MatrixBase)¶: Sets inverse variances and means.

set_means(m:MatrixBase)¶: Sets means.

set_weights(w:VectorBase)¶: Sets mixture weights.

split(target_components:int, perturb_factor:float) → list<int>¶

Splits components

Parameters:	target_components (int) – number of target components perturb_factor (float) – perturbation factor
Returns:	The order in which components were split.

valid_gconsts() → bool¶: Checks if gconsts are valid

weights() → Vector¶: Returns mixture weights

write(os:ostream, binary:bool)¶: Writes gaussian mixture model to output stream.

class kaldi.gmm.FullGmm(nmix=0, dim=0)[source]¶

Python wrapper for Kaldi::FullGmm<Float>

Provides a more pythonic access to the C++ methods.

Parameters:	nmix (int) – number of gaussian to mix dim (int) – dimension of each gaussian
Raises:	ValueError if nmix or dimension are not positive integers.

Creates a new FullGmm with specified number of gaussian mixtures and dimensions.

Parameters:	nmix (int) – number of gaussian to mix dim (int) – dimension

component_log_likelihood(data:VectorBase, comp_id:int) → float¶

Computes the log-likelihood of a data point given a single Gaussian component.

Parameters:	data (`kaldi.matrix.Vector`) – data point comp_id – component id
Returns:	Log-likehood of input data point for a given component

component_posteriors(data)[source]¶

Computes the posterior probabilities of all Gaussian components given: a data point.

Parameters:	data (VectorBase) – Data point with the same dimension as each component.
Returns:	2-element tuple containing loglike (`float`): Log-likelihood posteriors (`Vector`): Vector with the posterior probabilities
Raises:	ValueError if data is not consistent with gmm dimension.

compute_gconsts() → int¶

Sets the gconsts.

Returns:	Number of gconsts that are invalid e.g. because of zero weights or variances.

copy(src)[source]¶

Copies data from src into this FullGmm and returns this FullGmm.

Parameters:	src (FullGmm or DiagGmm) – Source Gmm to copy
Returns:	This FullGmm after update.

copy_from_full(fullgmm:FullGmm)¶: Copies from given FullGmm

dim() → int¶: Returns the dimensionality of the Gaussian mean vectors.

from_nmix_dim(nmix:int, dim:int) → FullGmm¶: Calls C++ function std::unique_ptr<::kaldi::FullGmm> ::kaldi::FullGmm::FullGmm(int, int)

from_other(gmm:FullGmm) → FullGmm¶: Calls C++ function std::unique_ptr<::kaldi::FullGmm> ::kaldi::FullGmm::FullGmm(::kaldi::FullGmm)

gaussian_selection(data:VectorBase, num_gselect:int) -> (log_like:float, output:list<int>)¶

Gets gaussian selection information for one frame.

Parameters:	data (`kaldi.matrix.Vector`) – data point num_gselect (int) – number of gaussians to select
Returns:	Log-likelihood for the input frame and the best `num_gselect` indices (sorted from best to worst likelihood).

gaussian_selection_preselect(data:VectorBase, preselect:list<int>, num_gselect:int) -> (log_like:float, posteriors:list<int>)¶

Get gaussian selection information for one frame.

Parameters:	data (`kaldi.matrix.Vector`) – data point preselect (list) – subset of mixture components num_gselect (int) – number of gaussians to select
Returns:	Log-likelihood for the input frame and the best `num_gselect` indices that were preselected (sorted from best to worst likelihood).

gconsts() → Vector¶: Returns gconsts

get_component_mean(gauss:int, out:VectorBase)¶: Gets component mean.

get_covars()[source]¶

Returns:	Component Covariances

get_covars_and_means()[source]¶

Returns:	Component Covariances

get_means() → Matrix¶: Returns component means.

interpolate(rho:float, source:FullGmm, flags:int=default)¶

Interpolates this model with other GMM

this <- rho * source + (1 - rho) * this

Parameters:	rho (float) – Interpolation weight source (FullGmm) – Source model flags (int) – Interpolation flags

inv_covars()[source]¶

Returns:	Component inverse covariances

log_likelihood(data:VectorBase) → float¶

Computes the log-likelihood for a data point

Parameters:	data (`kaldi.matrix.Vector`) – data point

log_likelihoods(data:VectorBase) → Vector¶

Computes the per-component log-likelihoods for a data point

Parameters:	data (`kaldi.matrix.Vector`) – data point

log_likelihoods_preselect(data:VectorBase, indices:list<int>) → Vector¶

Computes the per-component log-likelihoods of a subset of mixture components.

Parameters:	data (`kaldi.matrix.Vector`) – data point indices – list of indices

means_invcovars() → Matrix¶: Returns means times inverse covariances

merge(target_components:int) → list<int>¶

Merges components

Parameters:	target_components (int) – number of target components
Returns:	The order in which components were merged.

merge_preselect(target_components:int, preselect_pairs:list<tuple<int, int>>) → float¶

Merges components

This version only considers merging pairs in preselect_pairs. :param target_components: number of target components :type target_components: int :param preselect_pairs: preselected pairs :type preselect_pairs: List[Tuple[int, int]]

Returns:	Delta likelihood.

num_gauss() → int¶: Returns the number of mixture components.

perturb(perturb_factor:float)¶

Perturbs components

Component means are perturbed with a random vector multiplied by the pertrubation factor.

Parameters:	perturb_factor (float) – perturbation factor

read(is:istream, binary:bool)¶: Reads gaussian mixture model from input stream.

remove_component(gauss:int, renorm_weights:bool)¶: Removes single component from model.

remove_components(gauss:list<int>, renorm_weights:bool)¶: Removes multiple components from model.

resize(nmix:int, dim:int)¶: Resizes arrays to this dim. Does not initialize data.

set_inv_covars(v:list<SpMatrix>)¶: Sets inverse covariances.

set_inv_covars_and_means(invcovars:list<SpMatrix>, means:Matrix)¶

Updates both means and (inverse) covariances.

Parameters:	invcovars (list of `SpMatrix`) – List of inverse covariances means (`kaldi.matrix.Matrix`) – matrix of means

set_inv_covars_and_means_inv_covars(invcovars:list<SpMatrix>, means_invcovars:Matrix)¶

Use this if setting both, in the class’s native format.

Parameters:	invcovars (list of `SpMatrix`) – List of inverse covariances means_invcovars (`kaldi.matrix.Matrix`) – matrix of means and invcovars

set_means(means)[source]¶: Sets gmm component means.

set_weights(weights)[source]¶: Sets gmm mixture weights.

split(target_components:int, perturb_factor:float) → list<int>¶

Splits components

Parameters:	target_components (int) – number of target components perturb_factor (float) – perturbation factor
Returns:	The order in which components were split.

weights() → Vector¶: Returns mixture weights

write(os:ostream, binary:bool)¶: Writes gaussian mixture model to output stream.

class kaldi.gmm.FullGmmNormal¶

CLIF wrapper for ::kaldi::FullGmmNormal

copy_from_full(fullgmm:FullGmm)¶: Calls C++ function void ::kaldi::FullGmmNormal::CopyFromFullGmm(::kaldi::FullGmm)

copy_to_full(fullgmm:FullGmm, flags:int=default)¶: Calls C++ function void ::kaldi::FullGmmNormal::CopyToFullGmm(::kaldi::FullGmm *, unsigned short)

means_¶: C++ ::kaldi::Matrix<double> FullGmmNormal.means_

new_with_other(gmm:FullGmm) → FullGmmNormal¶: Calls C++ function std::unique_ptr<::kaldi::FullGmmNormal> ::kaldi::FullGmmNormal::FullGmmNormal(::kaldi::FullGmm)

rand(feats:MatrixBase)¶: Calls C++ function void ::kaldi::FullGmmNormal::Rand(::kaldi::MatrixBase<float> *)

resize(nMix:int, dim:int)¶: Calls C++ function void ::kaldi::FullGmmNormal::Resize(int, int)

vars_¶: C++ ::std::vector< ::kaldi::SpMatrix<double> > FullGmmNormal.vars_

weights_¶: C++ ::kaldi::Vector<double> FullGmmNormal.weights_

class kaldi.gmm.GmmUpdateFlags¶

An enumeration.

ALL = 15¶

MEANS = 1¶

TRANSITIONS = 8¶

VARIANCES = 2¶

WEIGHTS = 4¶

class kaldi.gmm.MapDiagGmmOptions¶

CLIF wrapper for ::kaldi::MapDiagGmmOptions

mean_tau¶: C++ ::kaldi::BaseFloat MapDiagGmmOptions.mean_tau

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

variance_tau¶: C++ ::kaldi::BaseFloat MapDiagGmmOptions.variance_tau

weight_tau¶: C++ ::kaldi::BaseFloat MapDiagGmmOptions.weight_tau

class kaldi.gmm.MleDiagGmmOptions¶

CLIF wrapper for ::kaldi::MleDiagGmmOptions

min_gaussian_occupancy¶: C++ ::kaldi::BaseFloat MleDiagGmmOptions.min_gaussian_occupancy

min_gaussian_weight¶: C++ ::kaldi::BaseFloat MleDiagGmmOptions.min_gaussian_weight

min_variance¶: C++ double MleDiagGmmOptions.min_variance

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

remove_low_count_gaussians¶: C++ bool MleDiagGmmOptions.remove_low_count_gaussians

variance_floor_vector¶: C++ ::kaldi::Vector<double> MleDiagGmmOptions.variance_floor_vector

class kaldi.gmm.MleFullGmmOptions¶

CLIF wrapper for ::kaldi::MleFullGmmOptions

max_condition¶: C++ ::kaldi::BaseFloat MleFullGmmOptions.max_condition

min_gaussian_occupancy¶: C++ ::kaldi::BaseFloat MleFullGmmOptions.min_gaussian_occupancy

min_gaussian_weight¶: C++ ::kaldi::BaseFloat MleFullGmmOptions.min_gaussian_weight

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

remove_low_count_gaussians¶: C++ bool MleFullGmmOptions.remove_low_count_gaussians

variance_floor¶: C++ ::kaldi::BaseFloat MleFullGmmOptions.variance_floor

class kaldi.gmm.SgmmUpdateFlags¶

An enumeration.

ALL = 255¶

COVARIANCE_MATRIX = 8¶

PHONE_PROJECTIONS = 2¶

PHONE_VECTORS = 1¶

PHONE_WEIGHT_PROJECTIONS = 4¶

SPEAKER_PROJECTIONS = 32¶

SPEAKER_WEIGHT_PROJECTIONS = 128¶

SUBSTATE_WEIGHTS = 16¶

TRANSITIONS = 64¶

class kaldi.gmm.SgmmWriteFlags¶

An enumeration.

BACKGROUND_GMMS = 8¶

GLOBAL_PARAMS = 1¶

NORMALIZERS = 4¶

STATE_PARAMS = 2¶

WRITE_ALL = 15¶

kaldi.gmm.augment_gmm_flags(f:int) → int¶: Calls C++ function unsigned short ::kaldi::AugmentGmmFlags(unsigned short)

kaldi.gmm.get_split_targets(state_occs:Vector, target_components:int, power:float, min_count:float) → list<int>¶: Calls C++ function void ::kaldi::GetSplitTargets(::kaldi::Vector<float>, int, float, float, ::std::vector< ::int32>*)

kaldi.gmm.gmm_flags_to_string(gmm_flags:GmmUpdateFlags) → str¶: Calls C++ function ::std::string ::kaldi::GmmFlagsToString(unsigned short)

kaldi.gmm.mle_full_gmm_update(config:MleFullGmmOptions, fullgmm_acc:AccumFullGmm, flags:int, gmm:FullGmm) -> (obj_change_out:float, count_out:float)¶: Calls C++ function void ::kaldi::MleFullGmmUpdate(::kaldi::MleFullGmmOptions, ::kaldi::AccumFullGmm, unsigned short, ::kaldi::FullGmm , float, float*)

kaldi.gmm.string_to_gmm_flags(s:str) → GmmUpdateFlags¶: Calls C++ function unsigned short ::kaldi::StringToGmmFlags(::std::string)

kaldi.gmm.string_to_sgmm_update_flags(s:str) → SgmmUpdateFlags¶: Calls C++ function unsigned short ::kaldi::StringToSgmmUpdateFlags(::std::string)

kaldi.gmm.string_to_sgmm_write_flags(s:str) → SgmmWriteFlags¶: Calls C++ function unsigned short ::kaldi::StringToSgmmWriteFlags(::std::string)

kaldi.gmm.am¶

Functions

`cluster_gaussians_to_ubm`	Calls C++ function
`map_am_diag_gmm_update`	Calls C++ function
`mle_am_diag_gmm_update`	Calls C++ function

Classes

`AccumAmDiagGmm`	CLIF wrapper for ::kaldi::AccumAmDiagGmm
`AmDiagGmm`	CLIF wrapper for ::kaldi::AmDiagGmm
`DecodableAmDiagGmm`	CLIF wrapper for ::kaldi::DecodableAmDiagGmm
`DecodableAmDiagGmmScaled`	CLIF wrapper for ::kaldi::DecodableAmDiagGmmScaled
`DecodableAmDiagGmmUnmapped`	CLIF wrapper for ::kaldi::DecodableAmDiagGmmUnmapped
`UbmClusteringOptions`	CLIF wrapper for ::kaldi::UbmClusteringOptions

class kaldi.gmm.am.AccumAmDiagGmm¶

CLIF wrapper for ::kaldi::AccumAmDiagGmm

accumulate_for_gaussian(model:AmDiagGmm, data:VectorBase, gmm_index:int, gauss_index:int, weight:float)¶: Calls C++ function void ::kaldi::AccumAmDiagGmm::AccumulateForGaussian(::kaldi::AmDiagGmm, ::kaldi::VectorBase<float>, int, int, float)

accumulate_for_gmm(model:AmDiagGmm, data:VectorBase, gmm_index:int, weight:float) → float¶: Calls C++ function float ::kaldi::AccumAmDiagGmm::AccumulateForGmm(::kaldi::AmDiagGmm, ::kaldi::VectorBase<float>, int, float)

accumulate_for_gmm_twofeats(model:AmDiagGmm, data1:VectorBase, data2:VectorBase, gmm_index:int, weight:float) → float¶: Calls C++ function float ::kaldi::AccumAmDiagGmm::AccumulateForGmmTwofeats(::kaldi::AmDiagGmm, ::kaldi::VectorBase<float>, ::kaldi::VectorBase<float>, int, float)

accumulate_from_posteriors(model:AmDiagGmm, data:VectorBase, gmm_index:int, posteriors:VectorBase)¶: Calls C++ function void ::kaldi::AccumAmDiagGmm::AccumulateFromPosteriors(::kaldi::AmDiagGmm, ::kaldi::VectorBase<float>, int, ::kaldi::VectorBase<float>)

add(scale:float, other:AccumAmDiagGmm)¶: Calls C++ function void ::kaldi::AccumAmDiagGmm::Add(float, ::kaldi::AccumAmDiagGmm)

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

get_acc(index:int) → AccumDiagGmm¶: Calls C++ function ::kaldi::AccumDiagGmm * ::kaldi::AccumAmDiagGmm::GetAccPtr(int)

init(model:AmDiagGmm, flags:int)¶: Calls C++ function void ::kaldi::AccumAmDiagGmm::Init(::kaldi::AmDiagGmm, unsigned short)

init_with_dim(model:AmDiagGmm, dim:int, flags:int)¶: Calls C++ function void ::kaldi::AccumAmDiagGmm::Init(::kaldi::AmDiagGmm, int, unsigned short)

num_accs() → int¶: Calls C++ function int ::kaldi::AccumAmDiagGmm::NumAccs()

read(is:istream, binary:bool, add:bool=default)¶: Calls C++ function void ::kaldi::AccumAmDiagGmm::Read(::std::basic_istream<char, ::std::char_traits<char> >, bool, bool)

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

set_zero(flags:int)¶: Calls C++ function void ::kaldi::AccumAmDiagGmm::SetZero(unsigned short)

tot_count() → float¶: Calls C++ function float ::kaldi::AccumAmDiagGmm::TotCount()

tot_log_like() → float¶: Calls C++ function float ::kaldi::AccumAmDiagGmm::TotLogLike()

tot_stats_count() → float¶: Calls C++ function float ::kaldi::AccumAmDiagGmm::TotStatsCount()

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

class kaldi.gmm.am.AmDiagGmm¶

CLIF wrapper for ::kaldi::AmDiagGmm

add_pdf(gmm:DiagGmm)¶: Calls C++ function void ::kaldi::AmDiagGmm::AddPdf(::kaldi::DiagGmm)

compute_gconsts() → int¶: Calls C++ function int ::kaldi::AmDiagGmm::ComputeGconsts()

copy_from_am_diag(other:AmDiagGmm)¶: Calls C++ function void ::kaldi::AmDiagGmm::CopyFromAmDiagGmm(::kaldi::AmDiagGmm)

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

get_gaussian_mean(pdf_index:int, gauss:int, out:VectorBase)¶: Calls C++ function void ::kaldi::AmDiagGmm::GetGaussianMean(int, int, ::kaldi::VectorBase<float> *)

get_gaussian_variance(pdf_index:int, gauss:int, out:VectorBase)¶: Calls C++ function void ::kaldi::AmDiagGmm::GetGaussianVariance(int, int, ::kaldi::VectorBase<float> *)

init(proto:DiagGmm, num_pdfs:int)¶: Calls C++ function void ::kaldi::AmDiagGmm::Init(::kaldi::DiagGmm, int)

log_likelihood(pdf_index:int, data:VectorBase) → float¶: Calls C++ function float ::kaldi::AmDiagGmm::LogLikelihood(int, ::kaldi::VectorBase<float>)

merge_by_count(state_occs:Vector, target_components:int, power:float, min_count:float)¶: Calls C++ function void ::kaldi::AmDiagGmm::MergeByCount(::kaldi::Vector<float>, int, float, float)

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

num_gauss_in_pdf(pdf_index:int) → int¶: Calls C++ function int ::kaldi::AmDiagGmm::NumGaussInPdf(int)

num_pdfs() → int¶: Calls C++ function int ::kaldi::AmDiagGmm::NumPdfs()

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

set_gaussian_mean(pdf_index:int, gauss:int, in:VectorBase)¶: Calls C++ function void ::kaldi::AmDiagGmm::SetGaussianMean(int, int, ::kaldi::VectorBase<float>)

split_by_count(state_occs:Vector, target_components:int, perturb_factor:float, power:float, min_count:float)¶: Calls C++ function void ::kaldi::AmDiagGmm::SplitByCount(::kaldi::Vector<float>, int, float, float, float)

split_pdf(idx:int, target_components:int, perturb_factor:float)¶: Calls C++ function void ::kaldi::AmDiagGmm::SplitPdf(int, int, float)

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

class kaldi.gmm.am.DecodableAmDiagGmm¶

CLIF wrapper for ::kaldi::DecodableAmDiagGmm

is_last_frame(frame:int) → bool¶: Checks if given frame is the last frame.

log_likelihood(frame:int, index:int) → float¶: Returns the log-likehood of the given index for the given frame.

num_frames_ready() → int¶: Returns number of frames ready for decoding.

num_indices() → int¶: Returns number of indices.

trans_model() → TransitionModel¶: Returns the transition model.

class kaldi.gmm.am.DecodableAmDiagGmmScaled¶

CLIF wrapper for ::kaldi::DecodableAmDiagGmmScaled

is_last_frame(frame:int) → bool¶: Checks if given frame is the last frame.

log_likelihood(frame:int, index:int) → float¶: Returns the log-likehood of the given index for the given frame.

num_frames_ready() → int¶: Returns number of frames ready for decoding.

num_indices() → int¶: Returns number of indices.

own_feats(am:AmDiagGmm, tm:TransitionModel, scale:float, log_sum_exp_prune:float, feats:Matrix) → DecodableAmDiagGmmScaled¶: Calls C++ function std::unique_ptr<::kaldi::DecodableAmDiagGmmScaled> ::kaldi::DecodableAmDiagGmmScaled::DecodableAmDiagGmmScaled(::kaldi::AmDiagGmm, ::kaldi::TransitionModel, float, float, ::kaldi::Matrix<float> *)

trans_model() → TransitionModel¶: Returns the transition model.

class kaldi.gmm.am.DecodableAmDiagGmmUnmapped¶

CLIF wrapper for ::kaldi::DecodableAmDiagGmmUnmapped

is_last_frame(frame:int) → bool¶: Checks if given frame is the last frame.

log_likelihood(frame:int, index:int) → float¶: Returns the log-likehood of the given index for the given frame.

num_frames_ready() → int¶: Returns number of frames ready for decoding.

num_indices() → int¶: Returns number of indices.

class kaldi.gmm.am.UbmClusteringOptions¶

CLIF wrapper for ::kaldi::UbmClusteringOptions

check()¶: Calls C++ function void ::kaldi::UbmClusteringOptions::Check()

cluster_varfloor¶: C++ ::kaldi::BaseFloat UbmClusteringOptions.cluster_varfloor

intermediate_num_gauss¶: C++ ::int32 UbmClusteringOptions.intermediate_num_gauss

max_am_gauss¶: C++ ::int32 UbmClusteringOptions.max_am_gauss

reduce_state_factor¶: C++ ::kaldi::BaseFloat UbmClusteringOptions.reduce_state_factor

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

ubm_num_gauss¶: C++ ::int32 UbmClusteringOptions.ubm_num_gauss

kaldi.gmm.am.cluster_gaussians_to_ubm(am:AmDiagGmm, state_occs:Vector, opts:UbmClusteringOptions, ubm_out:DiagGmm)¶: Calls C++ function void ::kaldi::ClusterGaussiansToUbm(::kaldi::AmDiagGmm, ::kaldi::Vector<float>, ::kaldi::UbmClusteringOptions, ::kaldi::DiagGmm *)

kaldi.gmm.am.map_am_diag_gmm_update(config:MapDiagGmmOptions, diag_gmm_acc:AccumAmDiagGmm, flags:int, am_gmm:AmDiagGmm) -> (obj_change_out:float, count_out:float)¶: Calls C++ function void ::kaldi::MapAmDiagGmmUpdate(::kaldi::MapDiagGmmOptions, ::kaldi::AccumAmDiagGmm, unsigned short, ::kaldi::AmDiagGmm , float, float*)

kaldi.gmm.am.mle_am_diag_gmm_update(config:MleDiagGmmOptions, diag_gmm_acc:AccumAmDiagGmm, flags:int, am_gmm:AmDiagGmm) -> (obj_change_out:float, count_out:float)¶: Calls C++ function void ::kaldi::MleAmDiagGmmUpdate(::kaldi::MleDiagGmmOptions, ::kaldi::AccumAmDiagGmm, unsigned short, ::kaldi::AmDiagGmm , float, float*)