This version of DecodableAmNnet is intended for a version of the decoder that processes different utterances with multiple threads. More...

#include <decodable-am-nnet.h>

Inheritance diagram for DecodableAmNnetParallel:

[legend]

Collaboration diagram for DecodableAmNnetParallel:

[legend]

Public Member Functions
	DecodableAmNnetParallel (const TransitionModel &trans_model, const AmNnet &am_nnet, const CuMatrix< BaseFloat > *feats, bool pad_input=true, BaseFloat prob_scale=1.0)

void	Compute ()

virtual BaseFloat	LogLikelihood (int32 frame, int32 transition_id)
	Returns the log likelihood, which will be negated in the decoder. More...

int32	NumFramesReady () const
	The call NumFramesReady() will return the number of frames currently available for this decodable object. More...

virtual int32	NumIndices () const
	Returns the number of states in the acoustic model (they will be indexed one-based, i.e. More...

virtual bool	IsLastFrame (int32 frame) const
	Returns true if this is the last frame. More...

	~DecodableAmNnetParallel ()

Public Member Functions inherited from DecodableInterface
virtual	~DecodableInterface ()

Protected Member Functions
	KALDI_DISALLOW_COPY_AND_ASSIGN (DecodableAmNnetParallel)

Protected Attributes
const TransitionModel &	trans_model_

const AmNnet &	am_nnet_

CuMatrix< BaseFloat >	log_probs_

const CuMatrix< BaseFloat > *	feats_

bool	pad_input_

BaseFloat	prob_scale_

Detailed Description

This version of DecodableAmNnet is intended for a version of the decoder that processes different utterances with multiple threads.

It needs to do the computation in a different place than the initializer, since the initializer gets called in the main thread of the program.

Definition at line 105 of file decodable-am-nnet.h.

Constructor & Destructor Documentation

◆ DecodableAmNnetParallel()

DecodableAmNnetParallel	(	const TransitionModel &	trans_model,
		const AmNnet &	am_nnet,
		const CuMatrix< BaseFloat > *	feats,
		bool	pad_input = `true`,
		BaseFloat	prob_scale = `1.0`
	)

inline

Definition at line 107 of file decodable-am-nnet.h.

References KALDI_ASSERT.

                                  :
       trans_model_(trans_model), am_nnet_(am_nnet), feats_(feats),
       pad_input_(pad_input), prob_scale_(prob_scale) {
     KALDI_ASSERT(feats_ != NULL);
   }

◆ ~DecodableAmNnetParallel()

~DecodableAmNnetParallel ( )

inline

Definition at line 166 of file decodable-am-nnet.h.

                              {
     delete feats_;
   }

Member Function Documentation

◆ Compute()

void Compute ( )

inline

Definition at line 118 of file decodable-am-nnet.h.

References MatrixBase< Real >::AddVecToRows(), MatrixBase< Real >::ApplyFloor(), MatrixBase< Real >::ApplyLog(), KALDI_ASSERT, DecodableAmNnet::log_probs_, kaldi::nnet2::NnetComputation(), TransitionModel::NumPdfs(), Matrix< Real >::Resize(), MatrixBase< Real >::Scale(), and DecodableAmNnet::trans_model_.

                  {
     log_probs_.Resize(feats_->NumRows(), trans_model_.NumPdfs());
     // the following function is declared in nnet-compute.h
     NnetComputation(am_nnet_.GetNnet(), *feats_,
                     pad_input_, &log_probs_);
     log_probs_.ApplyFloor(1.0e-20); // Avoid log of zero which leads to NaN.
     log_probs_.ApplyLog();
     CuVector<BaseFloat> priors(am_nnet_.Priors());
     KALDI_ASSERT(priors.Dim() == trans_model_.NumPdfs() &&
                  "Priors in neural network not set up.");
     priors.ApplyLog();
     // subtract log-prior (divide by prior)
     log_probs_.AddVecToRows(-1.0, priors);
     // apply probability scale.
     log_probs_.Scale(prob_scale_);
     delete feats_;
     feats_ = NULL;
   }

◆ IsLastFrame()

virtual bool IsLastFrame ( int32 frame ) const

inlinevirtual

Returns true if this is the last frame.

Frames are zero-based, so the first frame is zero. IsLastFrame(-1) will return false, unless the file is empty (which is a case that I'm not sure all the code will handle, so be careful). Caution: the behavior of this function in an online setting is being changed somewhat. In future it may return false in cases where we haven't yet decided to terminate decoding, but later true if we decide to terminate decoding. The plan in future is to rely more on NumFramesReady(), and in future, IsLastFrame() would always return false in an online-decoding setting, and would only return true in a decoding-from-matrix setting where we want to allow the last delta or LDA features to be flushed out for compatibility with the baseline setup.

Implements DecodableInterface.

Definition at line 162 of file decodable-am-nnet.h.

References KALDI_ASSERT, and DecodableAmNnet::NumFramesReady().

                                               {
     KALDI_ASSERT(frame < NumFramesReady());
     return (frame == NumFramesReady() - 1);
   }

◆ KALDI_DISALLOW_COPY_AND_ASSIGN()

KALDI_DISALLOW_COPY_AND_ASSIGN ( DecodableAmNnetParallel )

protected

◆ LogLikelihood()

virtual BaseFloat LogLikelihood	(	int32	frame,
		int32	index
	)

inlinevirtual

Returns the log likelihood, which will be negated in the decoder.

The "frame" starts from zero. You should verify that NumFramesReady() > frame before calling this.

Implements DecodableInterface.

Definition at line 139 of file decodable-am-nnet.h.

References DecodableAmNnet::log_probs_, DecodableAmNnet::trans_model_, and TransitionModel::TransitionIdToPdfFast().

                                                                     {
     if (feats_) Compute(); // this function sets feats_ to NULL.
     return log_probs_(frame,
                       trans_model_.TransitionIdToPdfFast(transition_id));
   }

◆ NumFramesReady()

int32 NumFramesReady ( ) const

inlinevirtual

The call NumFramesReady() will return the number of frames currently available for this decodable object.

This is for use in setups where you don't want the decoder to block while waiting for input. This is newly added as of Jan 2014, and I hope, going forward, to rely on this mechanism more than IsLastFrame to know when to stop decoding.

Reimplemented from DecodableInterface.

Definition at line 145 of file decodable-am-nnet.h.

References DecodableAmNnet::log_probs_, and MatrixBase< Real >::NumRows().

                                {
     if (feats_) {
       if (pad_input_) return feats_->NumRows();
       else {
         int32 ans = feats_->NumRows() - am_nnet_.GetNnet().LeftContext() -
             am_nnet_.GetNnet().RightContext();
         if (ans < 0) ans = 0;
         return ans;
       }
     } else {
       return log_probs_.NumRows();
     }
   }