Merge branch 'speechx' into speechx_refactor

speechx/speechx/asr/decoder/ctc_prefix_beam_search_decoder_main.cc

@@ -12,9 +12,10 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "decoder/ctc_prefix_beam_search_decoder.h"
+
 #include "absl/strings/str_split.h"
 #include "base/common.h"
+#include "decoder/ctc_prefix_beam_search_decoder.h"
 #include "frontend/audio/data_cache.h"
 #include "fst/symbol-table.h"
 #include "kaldi/util/table-types.h"
@@ -118,9 +119,9 @@ int main(int argc, char* argv[]) {
                     ori_feature_len - chunk_idx * chunk_stride, chunk_size);
             }
             if (this_chunk_size < receptive_field_length) {
-                LOG(WARNING) << "utt: " << utt << " skip last "
+                LOG(WARNING)
-                             << this_chunk_size << " frames, expect is "
+                    << "utt: " << utt << " skip last " << this_chunk_size
-                             << receptive_field_length;
+                    << " frames, expect is " << receptive_field_length;
                 break;
             }
 

speechx/speechx/asr/nnet/CMakeLists.txt

@@ -1,3 +1,4 @@
+
 set(srcs decodable.cc nnet_producer.cc)
 
 if(USING_DS2)

speechx/speechx/asr/nnet/decodable.cc

@@ -134,4 +134,4 @@ void Decodable::AttentionRescoring(const std::vector<std::vector<int>>& hyps,
     VLOG(1) << "Attention Rescoring cost:  " << timer.Elapsed() << " sec.";
 }
 
-}  // namespace ppspeech
+}  // namespace ppspeech

speechx/speechx/asr/nnet/decodable.h

@@ -62,12 +62,14 @@ class Decodable : public kaldi::DecodableInterface {
 
     int32 TokenId2NnetId(int32 token_id);
 
+
     // for offline test
     void Acceptlikelihood(const kaldi::Matrix<kaldi::BaseFloat>& likelihood);
 
   private:
     std::shared_ptr<NnetProducer> nnet_producer_;
 
+
     // the frame is nnet prob frame rather than audio feature frame
     // nnet frame subsample the feature frame
     // eg: 35 frame features output 8 frame inferences
@@ -79,7 +81,6 @@ class Decodable : public kaldi::DecodableInterface {
     int32 current_log_post_subsampled_offset_;
     int32 num_chunk_computed_;
     std::vector<kaldi::BaseFloat> framelikelihood_;
-
     kaldi::BaseFloat acoustic_scale_;
 };
 

speechx/speechx/asr/recognizer/u2_recognizer.cc

@@ -50,6 +50,7 @@ U2Recognizer::U2Recognizer(const U2RecognizerResource& resource)
 void U2Recognizer::Reset() {
     global_frame_offset_ = 0;
     input_finished_ = false;
+
     num_frames_ = 0;
     result_.clear();
 

speechx/speechx/asr/server/CMakeLists.txt

@@ -1 +1,2 @@
 #add_subdirectory(websocket)
+