#include <online-feat-input.h>

Inheritance diagram for OnlineFeInput< E >:

Collaboration diagram for OnlineFeInput< E >:

[legend]

Public Member Functions
	OnlineFeInput (OnlineAudioSourceItf au_src, E fe, const int32 frame_size, const int32 frame_shift, const bool snip_edges=true)

virtual int32	Dim () const

virtual bool	Compute (Matrix< BaseFloat > *output)

Public Member Functions inherited from OnlineFeatInputItf
virtual	~OnlineFeatInputItf ()

Private Member Functions
	KALDI_DISALLOW_COPY_AND_ASSIGN (OnlineFeInput)

Private Attributes
OnlineAudioSourceItf *	source_

E *	extractor_

const int32	frame_size_

const int32	frame_shift_

Vector< BaseFloat >	wave_

Vector< BaseFloat >	wave_remainder_

FrameExtractionOptions	frame_opts_

Detailed Description

template<class E>
class kaldi::OnlineFeInput< E >

Definition at line 272 of file online-feat-input.h.

Constructor & Destructor Documentation

◆ OnlineFeInput()

OnlineFeInput	(	OnlineAudioSourceItf *	au_src,
		E *	fe,
		const int32	frame_size,
		const int32	frame_shift,
		const bool	snip_edges = `true`
	)

Definition at line 300 of file online-feat-input.h.

References FrameExtractionOptions::frame_length_ms, OnlineFeInput< E >::frame_opts_, FrameExtractionOptions::frame_shift_ms, FrameExtractionOptions::samp_freq, and FrameExtractionOptions::snip_edges.

     : source_(au_src), extractor_(fe),
       frame_size_(frame_size), frame_shift_(frame_shift) {
       // we need a FrameExtractionOptions to call NumFrames()
       // 1000 is just a fake sample rate which equates ms and samples
       frame_opts_.samp_freq = 1000;
       frame_opts_.frame_shift_ms = frame_shift;
       frame_opts_.frame_length_ms = frame_size;
       frame_opts_.snip_edges = snip_edges;
 }

Member Function Documentation

◆ Compute()

bool Compute ( Matrix< BaseFloat > * output )

virtual

Implements OnlineFeatInputItf.

Definition at line 314 of file online-feat-input.h.

References OnlineFeInput< E >::extractor_, OnlineFeInput< E >::frame_opts_, OnlineFeInput< E >::frame_shift_, OnlineFeInput< E >::frame_size_, KALDI_ASSERT, KALDI_WARN, kaldi::NumFrames(), MatrixBase< Real >::NumRows(), VectorBase< Real >::Range(), OnlineAudioSourceItf::Read(), Matrix< Real >::Resize(), OnlineFeInput< E >::source_, and OnlineFeInput< E >::wave_remainder_.

                                                    {
   MatrixIndexT nvec = output->NumRows(); // the number of output vectors
   if (nvec <= 0) {
     KALDI_WARN << "No feature vectors requested?!";
     return true;
   }
 
   // Prepare the input audio samples
   int32 samples_req = frame_size_ + (nvec - 1) * frame_shift_;
   Vector<BaseFloat> read_samples(samples_req);
 
   bool ans = source_->Read(&read_samples);
 
   Vector<BaseFloat> all_samples(wave_remainder_.Dim() + read_samples.Dim());
   all_samples.Range(0, wave_remainder_.Dim()).CopyFromVec(wave_remainder_);
   all_samples.Range(wave_remainder_.Dim(), read_samples.Dim()).
       CopyFromVec(read_samples);
 
   // Extract the features
   if (all_samples.Dim() >= frame_size_) {
     // extract waveform remainder before calling Compute()
     int32 num_frames = NumFrames(all_samples.Dim(), frame_opts_);
     // offset is the amount at the start that has been extracted.
     int32 offset = num_frames * frame_shift_;
     int32 remaining_len = all_samples.Dim() - offset;
     wave_remainder_.Resize(remaining_len);
     KALDI_ASSERT(remaining_len >= 0);
     if (remaining_len > 0)
       wave_remainder_.CopyFromVec(SubVector<BaseFloat>(all_samples, offset, remaining_len));
     extractor_->Compute(all_samples, 1.0, output);
   } else {
     output->Resize(0, 0);
     wave_remainder_ = all_samples;
   }
 
   return ans;
 }