[src] Add batched xvector computation (#3643)

Author: danpovey

src/nnet3/nnet-general-component.cc

@@ -573,7 +573,7 @@ void StatisticsPoolingComponent::InitFromConfig(ConfigLine *cfl) {
 StatisticsPoolingComponent::StatisticsPoolingComponent():
  input_dim_(-1), input_period_(1), left_context_(-1), right_context_(-1),
  num_log_count_features_(0), output_stddevs_(false),
- variance_floor_(1.0e-10) { }
+ variance_floor_(1.0e-10), require_direct_input_(false) { }
 
 
 StatisticsPoolingComponent::StatisticsPoolingComponent(
@@ -582,7 +582,8 @@ StatisticsPoolingComponent::StatisticsPoolingComponent(
  left_context_(other.left_context_), right_context_(other.right_context_),
  num_log_count_features_(other.num_log_count_features_),
  output_stddevs_(other.output_stddevs_),
- variance_floor_(1.0e-10) {
+ variance_floor_(other.variance_floor_),
+ require_direct_input_(other.require_direct_input_) {
  Check();
 }
 
@@ -614,6 +615,9 @@ void StatisticsPoolingComponent::Read(std::istream &is, bool binary) {
  ExpectToken(is, binary, "<VarianceFloor>");
  ReadBasicType(is, binary, &variance_floor_);
  ExpectToken(is, binary, "</StatisticsPoolingComponent>");
+ require_direct_input_ = false; // This is not written to disk, it's only used
+ // temporarily, in memory (see
+ // nnet3-xvector-compute-batched.cc).
  Check();
 }
 

src/nnet3/nnet-general-component.h

@@ -331,7 +331,8 @@ class StatisticsExtractionComponentPrecomputedIndexes:
  or whatever, instead of just component-name, because its output is only defined at multiples
  of its input-period.
 
- The output of StatisticsPoolingComponent will only be defined if at least one input was defined.
+ The output of StatisticsPoolingComponent will only be defined if at least one
+ input was defined.
  */
 class StatisticsPoolingComponent: public Component {
  public:
@@ -396,6 +397,11 @@ class StatisticsPoolingComponent: public Component {
  const std::vector<Index> &output_indexes,
  bool need_backprop) const;
 
+ // Used in computing the 'real' context of networks involving this component;
+ // with the default value of false, the left/right context will always appear
+ // to be 0.
+ void SetRequireDirectInput(bool b) { require_direct_input_ = b; }
+
  private:
  // Checks that the parameters are valid.
  void Check() const;
@@ -411,6 +417,13 @@ class StatisticsPoolingComponent: public Component {
  int32 num_log_count_features_;
  bool output_stddevs_;
  BaseFloat variance_floor_;
+ // If require_direct_input_ is set to true, in order for a particular 't'
+ // value to be available at the output of this component, it will require that
+ // 't' value to be computable at the input. This is used in computing the
+ // "real" left/right context of the network, but this member isn't currently
+ // written to disk and will default to false when read.
+ bool require_direct_input_;
+
 };
 
 class StatisticsPoolingComponentPrecomputedIndexes:

src/nnet3/nnet-utils.cc

@@ -300,6 +300,15 @@ void SetNnetAsGradient(Nnet *nnet) {
  }
 }
 
+void SetRequireDirectInput(bool b, Nnet *nnet) {
+ for (int32 c = 0; c < nnet->NumComponents(); c++) {
+ Component *comp = nnet->GetComponent(c);
+ if (dynamic_cast<StatisticsPoolingComponent*>(comp) != NULL)
+ dynamic_cast<StatisticsPoolingComponent*>(comp)->SetRequireDirectInput(b);
+ }
+}
+
+
 void ScaleNnet(BaseFloat scale, Nnet *nnet) {
  if (scale == 1.0) return;
  else {
@@ -724,7 +733,7 @@ class SvdApplier {
  << " components to FixedAffineComponent.";
  }
 
- // This function finds the minimum index of 
+ // This function finds the minimum index of
  // the Descending order sorted [input_vector],
  // over a range of indices from [lower] to [upper] index,
  // for which the sum of elements upto the found min. index is greater
@@ -743,7 +752,7 @@ class SvdApplier {
  }
  return (i+1);
  }
- 
+
 // Here we perform SVD based refactorig of an input Affine component.
 // After applying SVD , we sort the Singularity values in descending order,
 // and take the subset of values which contribute to energy_threshold times
@@ -777,7 +786,7 @@ class SvdApplier {
  if (energy_threshold_ > 0) {
  BaseFloat min_singular_sum = energy_threshold_ * s2_sum_orig;
  bottleneck_dim_ = GetReducedDimension(s2, 0, s2.Dim()-1, min_singular_sum);
- } 
+ }
  SubVector<BaseFloat> this_part(s2, 0, bottleneck_dim_);
  BaseFloat s2_sum_reduced = this_part.Sum();
  BaseFloat shrinkage_ratio =

src/nnet3/nnet-utils.h

@@ -118,6 +118,14 @@ void ScaleNnet(BaseFloat scale, Nnet *nnet);
 /// learning_rate_ to 1 for each UpdatableComponent in nnet
 void SetNnetAsGradient(Nnet *nnet);
 
+
+/// Calls the corresponding function in any component of type
+/// StatisticsPoolingComponent; used as a way to compute the 'real' left-right
+/// context of networks including SatisticsPoolingComponent, which will give you
+/// the minimum chunk size they can consume.
+void SetRequireDirectInput(bool b, Nnet *nnet);
+
+
 /// Does *dest += alpha * src (affects nnet parameters and
 /// stored stats).
 void AddNnet(const Nnet &src, BaseFloat alpha, Nnet *dest);

src/nnet3bin/Makefile

@@ -19,6 +19,7 @@ BINFILES = nnet3-init nnet3-info nnet3-get-egs nnet3-copy-egs nnet3-subset-egs \
  nnet3-discriminative-subset-egs nnet3-get-egs-simple \
  nnet3-discriminative-compute-from-egs nnet3-latgen-faster-looped \
  nnet3-egs-augment-image nnet3-xvector-get-egs nnet3-xvector-compute \
+ nnet3-xvector-compute-batched \
  nnet3-latgen-grammar nnet3-compute-batch nnet3-latgen-faster-batch \
  cuda-gpu-available cuda-compiled
 
@@ -36,4 +37,5 @@ ADDLIBS = ../nnet3/kaldi-nnet3.a ../chain/kaldi-chain.a \
  ../tree/kaldi-tree.a ../util/kaldi-util.a ../matrix/kaldi-matrix.a \
  ../base/kaldi-base.a
 
+
 include ../makefiles/default_rules.mk