add min cutoff & top n cutoff

deep_speech_2/deploy.py

@@ -18,7 +18,7 @@ import time
 parser = argparse.ArgumentParser(description=__doc__)
 parser.add_argument(
     "--num_samples",
-    default=4,
+    default=10,
     type=int,
     help="Number of samples for inference. (default: %(default)s)")
 parser.add_argument(
@@ -95,12 +95,12 @@ parser.add_argument(
     help="Path for language model. (default: %(default)s)")
 parser.add_argument(
     "--alpha",
-    default=0.26,
+    default=1.5,
     type=float,
     help="Parameter associated with language model. (default: %(default)f)")
 parser.add_argument(
     "--beta",
-    default=0.1,
+    default=0.3,
     type=float,
     help="Parameter associated with word count. (default: %(default)f)")
 parser.add_argument(
@@ -109,6 +109,12 @@ parser.add_argument(
     type=float,
     help="The cutoff probability of pruning"
     "in beam search. (default: %(default)f)")
+parser.add_argument(
+    "--cutoff_top_n",
+    default=40,
+    type=int,
+    help="The cutoff number of pruning"
+    "in beam search. (default: %(default)f)")
 args = parser.parse_args()
 
 
@@ -184,6 +190,7 @@ def infer():
                 vocabulary=data_generator.vocab_list,
                 blank_id=len(data_generator.vocab_list),
                 cutoff_prob=args.cutoff_prob,
+                cutoff_top_n=args.cutoff_top_n,
                 ext_scoring_func=ext_scorer, )
             batch_beam_results += [beam_result]
     else:
@@ -194,6 +201,7 @@ def infer():
             blank_id=len(data_generator.vocab_list),
             num_processes=args.num_processes_beam_search,
             cutoff_prob=args.cutoff_prob,
+            cutoff_top_n=args.cutoff_top_n,
             ext_scoring_func=ext_scorer, )
 
     for i, beam_result in enumerate(batch_beam_results):

deep_speech_2/deploy/ctc_decoders.cpp

@@ -62,6 +62,7 @@ std::vector<std::pair<double, std::string> >
                             std::vector<std::string> vocabulary,
                             int blank_id,
                             double cutoff_prob,
+                            int cutoff_top_n,
                             Scorer *ext_scorer)
 {
     // dimension check
@@ -116,12 +117,24 @@ std::vector<std::pair<double, std::string> >
             prob_idx.push_back(std::pair<int, double>(i, prob[i]));
         }
 
+       float min_cutoff = -NUM_FLT_INF;
+       bool full_beam = false;
+       if (ext_scorer != nullptr) {
+           int num_prefixes = std::min((int)prefixes.size(), beam_size);
+           std::sort(prefixes.begin(), prefixes.begin() + num_prefixes,
+                     prefix_compare);
+           min_cutoff = prefixes[num_prefixes-1]->_score + log(prob[blank_id])
+                        - std::max(0.0, ext_scorer->beta);
+           full_beam = (num_prefixes == beam_size);
+       }
+
         // pruning of vacobulary
         int cutoff_len = prob.size();
-        if (cutoff_prob < 1.0) {
+        if (cutoff_prob < 1.0 || cutoff_top_n < prob.size()) {
             std::sort(prob_idx.begin(),
                       prob_idx.end(),
                       pair_comp_second_rev<int, double>);
+            if (cutoff_prob < 1.0) {
                 double cum_prob = 0.0;
                 cutoff_len = 0;
                 for (int i=0; i<prob_idx.size(); i++) {
@@ -129,6 +142,8 @@ std::vector<std::pair<double, std::string> >
                     cutoff_len += 1;
                     if (cum_prob >= cutoff_prob) break;
                 }
+            }
+            cutoff_len = std::min(cutoff_len, cutoff_top_n);
             prob_idx = std::vector<std::pair<int, double> >( prob_idx.begin(),
                             prob_idx.begin() + cutoff_len);
         }
@@ -138,15 +153,17 @@ std::vector<std::pair<double, std::string> >
             log_prob_idx.push_back(std::pair<int, float>
                   (prob_idx[i].first, log(prob_idx[i].second + NUM_FLT_MIN)));
         }
-
         // loop over chars
         for (int index = 0; index < log_prob_idx.size(); index++) {
             auto c = log_prob_idx[index].first;
             float log_prob_c = log_prob_idx[index].second;
-            //float log_probs_prev;
 
             for (int i = 0; i < prefixes.size() && i<beam_size; i++) {
                 auto prefix = prefixes[i];
+
+                if (full_beam && log_prob_c + prefix->_score < min_cutoff) {
+                    break;
+                }
                 // blank
                 if (c == blank_id) {
                     prefix->_log_prob_b_cur = log_sum_exp(
@@ -178,7 +195,7 @@ std::vector<std::pair<double, std::string> >
                         (c == space_id || ext_scorer->is_character_based()) ) {
                         PathTrie *prefix_to_score = nullptr;
 
-                        // don't score the space
+                        // skip scoring the space
                         if (ext_scorer->is_character_based()) {
                             prefix_to_score = prefix_new;
                         } else {
@@ -202,10 +219,10 @@ std::vector<std::pair<double, std::string> >
         } // end of loop over chars
 
         prefixes.clear();
-        // update log probabilities
+        // update log probs
         root.iterate_to_vec(prefixes);
 
-        // sort prefixes by score
+        // preserve top beam_size prefixes
         if (prefixes.size() >= beam_size) {
             std::nth_element(prefixes.begin(),
                     prefixes.begin() + beam_size,
@@ -218,18 +235,20 @@ std::vector<std::pair<double, std::string> >
         }
     }
 
+    // compute aproximate ctc score as the return score
     for (size_t i = 0; i < beam_size && i < prefixes.size(); i++) {
         double approx_ctc = prefixes[i]->_score;
 
-        // remove word insert:
+        if (ext_scorer != nullptr) {
             std::vector<int> output;
             prefixes[i]->get_path_vec(output);
             size_t prefix_length = output.size();
+            auto words = ext_scorer->split_labels(output);
+           // remove word insert
+            approx_ctc = approx_ctc - prefix_length * ext_scorer->beta;
             // remove language model weight:
-        if (ext_scorer != nullptr) {
-           // auto words = split_labels(output);
-           // approx_ctc = approx_ctc - path_length * ext_scorer->beta;
-           // approx_ctc -= (_lm->get_sent_log_prob(words)) * ext_scorer->alpha;
+            approx_ctc -= (ext_scorer->get_sent_log_prob(words))
+                          * ext_scorer->alpha;
         }
 
         prefixes[i]->_approx_ctc = approx_ctc;
@@ -253,11 +272,9 @@ std::vector<std::pair<double, std::string> >
         for (int j = 0; j < output.size(); j++) {
             output_str += vocabulary[output[j]];
         }
-        std::pair<double, std::string> output_pair(space_prefixes[i]->_score,
-                                                   output_str);
-        output_vecs.emplace_back(
-            output_pair
-        );
+        std::pair<double, std::string>
+            output_pair(-space_prefixes[i]->_approx_ctc, output_str);
+        output_vecs.emplace_back(output_pair);
     }
 
     return output_vecs;
@@ -272,6 +289,7 @@ std::vector<std::vector<std::pair<double, std::string> > >
                 int blank_id,
                 int num_processes,
                 double cutoff_prob,
+                int cutoff_top_n,
                 Scorer *ext_scorer
                 ) {
     if (num_processes <= 0) {
@@ -295,7 +313,8 @@ std::vector<std::vector<std::pair<double, std::string> > >
     for (int i = 0; i < batch_size; i++) {
         res.emplace_back(
                 pool.enqueue(ctc_beam_search_decoder, probs_split[i],
-                    beam_size, vocabulary, blank_id, cutoff_prob, ext_scorer)
+                    beam_size, vocabulary, blank_id, cutoff_prob,
+                    cutoff_top_n, ext_scorer)
             );
     }
     // get decoding results

deep_speech_2/deploy/ctc_decoders.h

@@ -39,6 +39,7 @@ std::vector<std::pair<double, std::string> >
                             std::vector<std::string> vocabulary,
                             int blank_id,
                             double cutoff_prob=1.0,
+                            int cutoff_top_n=40,
                             Scorer *ext_scorer=NULL
                             );
 
@@ -66,6 +67,7 @@ std::vector<std::vector<std::pair<double, std::string>>>
                             int blank_id,
                             int num_processes,
                             double cutoff_prob=1.0,
+                            int cutoff_top_n=40,
                             Scorer *ext_scorer=NULL
                             );
 

deep_speech_2/deploy/scorer.h

@@ -50,6 +50,7 @@ public:
     void fill_dictionary(bool add_space);
     // set char map
     void set_char_map(std::vector<std::string> char_list);
+    std::vector<std::string> split_labels(const std::vector<int> &labels);
     // expose to decoder
     double alpha;
     double beta;
@@ -60,7 +61,6 @@ protected:
     void load_LM(const char* filename);
     double get_log_prob(const std::vector<std::string>& words);
     std::string vec2str(const std::vector<int> &input);
-    std::vector<std::string> split_labels(const std::vector<int> &labels);
 
 private:
     void* _language_model;

deep_speech_2/deploy/swig_decoders_wrapper.py

@@ -43,6 +43,7 @@ def ctc_beam_search_decoder(probs_seq,
                             vocabulary,
                             blank_id,
                             cutoff_prob=1.0,
+                            cutoff_top_n=40,
                             ext_scoring_func=None):
     """Wrapper for the CTC Beam Search Decoder.
 
@@ -59,6 +60,10 @@ def ctc_beam_search_decoder(probs_seq,
     :param cutoff_prob: Cutoff probability in pruning,
                         default 1.0, no pruning.
     :type cutoff_prob: float
+    :param cutoff_top_n: Cutoff number in pruning, only top cutoff_top_n
+                        characters with highest probs in vocabulary will be
+                        used in beam search, default 40.
+    :type cutoff_top_n: int
     :param ext_scoring_func: External scoring function for
                             partially decoded sentence, e.g. word count
                             or language model.
@@ -67,9 +72,9 @@ def ctc_beam_search_decoder(probs_seq,
              results, in descending order of the probability.
     :rtype: list
     """
-    return swig_decoders.ctc_beam_search_decoder(probs_seq.tolist(), beam_size,
-                                                 vocabulary, blank_id,
-                                                 cutoff_prob, ext_scoring_func)
+    return swig_decoders.ctc_beam_search_decoder(
+        probs_seq.tolist(), beam_size, vocabulary, blank_id, cutoff_prob,
+        cutoff_top_n, ext_scoring_func)
 
 
 def ctc_beam_search_decoder_batch(probs_split,
@@ -78,6 +83,7 @@ def ctc_beam_search_decoder_batch(probs_split,
                                   blank_id,
                                   num_processes,
                                   cutoff_prob=1.0,
+                                  cutoff_top_n=40,
                                   ext_scoring_func=None):
     """Wrapper for the batched CTC beam search decoder.
 
@@ -92,11 +98,15 @@ def ctc_beam_search_decoder_batch(probs_split,
     :type blank_id: int
     :param num_processes: Number of parallel processes.
     :type num_processes: int
-    :param cutoff_prob: Cutoff probability in pruning,
+    :param cutoff_prob: Cutoff probability in vocabulary pruning,
                         default 1.0, no pruning.
+    :type cutoff_prob: float
+    :param cutoff_top_n: Cutoff number in pruning, only top cutoff_top_n
+                        characters with highest probs in vocabulary will be
+                        used in beam search, default 40.
+    :type cutoff_top_n: int
     :param num_processes: Number of parallel processes.
     :type num_processes: int
-    :type cutoff_prob: float
     :param ext_scoring_func: External scoring function for
                             partially decoded sentence, e.g. word count
                             or language model.
@@ -109,4 +119,4 @@ def ctc_beam_search_decoder_batch(probs_split,
 
     return swig_decoders.ctc_beam_search_decoder_batch(
         probs_split, beam_size, vocabulary, blank_id, num_processes,
-        cutoff_prob, ext_scoring_func)
+        cutoff_prob, cutoff_top_n, ext_scoring_func)