format code

speechx/speechx/decoder/ctc_prefix_beam_search_decoder.cc

@@ -118,11 +118,13 @@ void CTCPrefixBeamSearch::AdvanceDecoding(
         std::vector<float> topk_score;
         std::vector<int32_t> topk_index;
         TopK(logp_t, first_beam_size, &topk_score, &topk_index);
-        VLOG(2) << "topk: " << num_frame_decoded_ << " " <<  *std::max_element(logp_t.begin(), logp_t.end()) << " " << topk_score[0];
-        for (int i = 0; i < topk_score.size(); i++){
-             VLOG(2) << "topk: " << num_frame_decoded_ << " " << topk_score[i];
+        VLOG(2) << "topk: " << num_frame_decoded_ << " "
+                << *std::max_element(logp_t.begin(), logp_t.end()) << " "
+                << topk_score[0];
+        for (int i = 0; i < topk_score.size(); i++) {
+            VLOG(2) << "topk: " << num_frame_decoded_ << " " << topk_score[i];
         }
-       
+
         // 2. token passing
         for (int i = 0; i < topk_index.size(); ++i) {
             int id = topk_index[i];
@@ -303,15 +305,16 @@ void CTCPrefixBeamSearch::UpdateOutputs(
     outputs_.emplace_back(output);
 }
 
-void CTCPrefixBeamSearch::FinalizeSearch() { 
-    UpdateFinalContext(); 
-    
+void CTCPrefixBeamSearch::FinalizeSearch() {
+    UpdateFinalContext();
+
     VLOG(2) << "num_frame_decoded_: " << num_frame_decoded_;
     int cnt = 0;
-    for (int i = 0; i < hypotheses_.size(); i ++){
-        VLOG(2) << "hyp " << cnt << " len: " << hypotheses_[i].size() << " ctc score: " << likelihood_[i];
-        for (int j = 0; j < hypotheses_[i].size(); j ++){
-            VLOG(2) <<  hypotheses_[i][j];
+    for (int i = 0; i < hypotheses_.size(); i++) {
+        VLOG(2) << "hyp " << cnt << " len: " << hypotheses_[i].size()
+                << " ctc score: " << likelihood_[i];
+        for (int j = 0; j < hypotheses_[i].size(); j++) {
+            VLOG(2) << hypotheses_[i][j];
         }
     }
 }

speechx/speechx/decoder/ctc_prefix_beam_search_decoder.h

@@ -13,7 +13,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-// modified from https://github.com/wenet-e2e/wenet/blob/main/runtime/core/decoder/ctc_prefix_beam_search.cc
+// modified from
+// https://github.com/wenet-e2e/wenet/blob/main/runtime/core/decoder/ctc_prefix_beam_search.cc
 
 #pragma once
 

speechx/speechx/decoder/ctc_prefix_beam_search_score.h

@@ -13,7 +13,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-// modified from https://github.com/wenet-e2e/wenet/blob/main/runtime/core/decoder/ctc_prefix_beam_search.h
+// modified from
+// https://github.com/wenet-e2e/wenet/blob/main/runtime/core/decoder/ctc_prefix_beam_search.h
 
 #pragma once
 

speechx/speechx/decoder/param.h

@@ -20,7 +20,9 @@
 
 // feature
 DEFINE_bool(use_fbank, false, "False for fbank; or linear feature");
-DEFINE_bool(fill_zero, false, "fill zero at last chunk, when chunk < chunk_size");
+DEFINE_bool(fill_zero,
+            false,
+            "fill zero at last chunk, when chunk < chunk_size");
 // DEFINE_bool(to_float32, true, "audio convert to pcm32. True for linear
 // feature, or fbank");
 DEFINE_int32(num_bins, 161, "num bins of mel");

speechx/speechx/frontend/audio/assembler.cc

@@ -16,9 +16,9 @@
 
 namespace ppspeech {
 
+using kaldi::BaseFloat;
 using kaldi::Vector;
 using kaldi::VectorBase;
-using kaldi::BaseFloat;
 using std::unique_ptr;
 
 Assembler::Assembler(AssemblerOptions opts,
@@ -51,9 +51,11 @@ bool Assembler::Compute(Vector<BaseFloat>* feats) {
         Vector<BaseFloat> feature;
         bool result = base_extractor_->Read(&feature);
         if (result == false || feature.Dim() == 0) {
-            VLOG(1) << "result: " << result << " feature dim: " << feature.Dim();
+            VLOG(1) << "result: " << result
+                    << " feature dim: " << feature.Dim();
             if (IsFinished() == false) {
-                VLOG(1) << "finished reading feature. cache size: " << feature_cache_.size();
+                VLOG(1) << "finished reading feature. cache size: "
+                        << feature_cache_.size();
                 return false;
             } else {
                 VLOG(1) << "break";
@@ -69,7 +71,8 @@ bool Assembler::Compute(Vector<BaseFloat>* feats) {
     }
 
     if (feature_cache_.size() < receptive_filed_length_) {
-        VLOG(1) << "feature_cache less than receptive_filed_lenght. " << feature_cache_.size() << ": " << receptive_filed_length_;
+        VLOG(1) << "feature_cache less than receptive_filed_lenght. "
+                << feature_cache_.size() << ": " << receptive_filed_length_;
         return false;
     }
 
@@ -81,7 +84,8 @@ bool Assembler::Compute(Vector<BaseFloat>* feats) {
         }
     }
 
-    int32 this_chunk_size = std::min(static_cast<int32>(feature_cache_.size()), frame_chunk_size_);
+    int32 this_chunk_size =
+        std::min(static_cast<int32>(feature_cache_.size()), frame_chunk_size_);
     feats->Resize(dim_ * this_chunk_size);
     VLOG(1) << "read " << this_chunk_size << " feat.";
 
@@ -89,7 +93,7 @@ bool Assembler::Compute(Vector<BaseFloat>* feats) {
     while (counter < this_chunk_size) {
         Vector<BaseFloat>& val = feature_cache_.front();
         CHECK(val.Dim() == dim_) << val.Dim();
-      
+
         int32 start = counter * dim_;
         feats->Range(start, dim_).CopyFromVec(val);
 
@@ -99,7 +103,7 @@ bool Assembler::Compute(Vector<BaseFloat>* feats) {
 
         // val is reference, so we should pop here
         feature_cache_.pop();
-  
+
         counter++;
     }
     CHECK(feature_cache_.size() == cache_size_);
@@ -108,11 +112,11 @@ bool Assembler::Compute(Vector<BaseFloat>* feats) {
 }
 
 
- void Assembler::Reset() { 
+void Assembler::Reset() {
     std::queue<kaldi::Vector<kaldi::BaseFloat>> empty;
     std::swap(feature_cache_, empty);
     nframes_ = 0;
-    base_extractor_->Reset(); 
+    base_extractor_->Reset();
 }
 
 }  // namespace ppspeech

speechx/speechx/frontend/audio/assembler.h

@@ -25,7 +25,8 @@ struct AssemblerOptions {
     int32 receptive_filed_length{1};
     int32 subsampling_rate{1};
     int32 nnet_decoder_chunk{1};
-    bool fill_zero{false}; // whether fill zero when last chunk is not equal to frame_chunk_size_
+    bool fill_zero{false};  // whether fill zero when last chunk is not equal to
+                            // frame_chunk_size_
 };
 
 class Assembler : public FrontendInterface {
@@ -62,7 +63,7 @@ class Assembler : public FrontendInterface {
     std::queue<kaldi::Vector<kaldi::BaseFloat>> feature_cache_;
     std::unique_ptr<FrontendInterface> base_extractor_;
 
-    int32 nframes_; // num frame computed
+    int32 nframes_;  // num frame computed
     DISALLOW_COPY_AND_ASSIGN(Assembler);
 };
 

speechx/speechx/frontend/audio/audio_cache.cc

@@ -13,13 +13,14 @@
 // limitations under the License.
 
 #include "frontend/audio/audio_cache.h"
+
 #include "kaldi/base/timer.h"
 
 namespace ppspeech {
 
 using kaldi::BaseFloat;
-using kaldi::VectorBase;
 using kaldi::Vector;
+using kaldi::VectorBase;
 
 AudioCache::AudioCache(int buffer_size, bool to_float32)
     : finished_(false),
@@ -85,8 +86,8 @@ bool AudioCache::Read(Vector<BaseFloat>* waves) {
     offset_ = (offset_ + chunk_size) % ring_buffer_.size();
 
     nsamples_ += chunk_size;
-    VLOG(1) << "nsamples readed: " <<  nsamples_;
-    
+    VLOG(1) << "nsamples readed: " << nsamples_;
+
     ready_feed_condition_.notify_one();
     return true;
 }

speechx/speechx/frontend/audio/audio_cache.h

@@ -62,7 +62,7 @@ class AudioCache : public FrontendInterface {
     kaldi::int32 timeout_;  // millisecond
     bool to_float32_;       // int16 -> float32. used in linear_spectrogram
 
-    int32 nsamples_; // number samples readed.
+    int32 nsamples_;  // number samples readed.
     DISALLOW_COPY_AND_ASSIGN(AudioCache);
 };
 

speechx/speechx/frontend/audio/feature_cache.cc

@@ -16,12 +16,12 @@
 
 namespace ppspeech {
 
-using kaldi::Vector;
-using kaldi::VectorBase;
 using kaldi::BaseFloat;
-using std::vector;
 using kaldi::SubVector;
+using kaldi::Vector;
+using kaldi::VectorBase;
 using std::unique_ptr;
+using std::vector;
 
 FeatureCache::FeatureCache(FeatureCacheOptions opts,
                            unique_ptr<FrontendInterface> base_extractor) {

speechx/speechx/frontend/audio/feature_cache.h

@@ -77,7 +77,7 @@ class FeatureCache : public FrontendInterface {
     std::condition_variable ready_feed_condition_;
     std::condition_variable ready_read_condition_;
 
-    int32 nframe_; // num of feature computed
+    int32 nframe_;  // num of feature computed
     DISALLOW_COPY_AND_ASSIGN(FeatureCache);
 };
 

speechx/speechx/frontend/audio/feature_pipeline.h

@@ -91,8 +91,7 @@ struct FeaturePipelineOptions {
                   << opts.assembler_opts.receptive_filed_length;
         LOG(INFO) << "nnet chunk size: "
                   << opts.assembler_opts.nnet_decoder_chunk;
-        LOG(INFO) << "frontend fill zeros: "
-                  << opts.assembler_opts.fill_zero;
+        LOG(INFO) << "frontend fill zeros: " << opts.assembler_opts.fill_zero;
         return opts;
     }
 };

speechx/speechx/nnet/decodable.cc

@@ -79,7 +79,8 @@ bool Decodable::AdvanceChunk() {
     int32& vocab_dim = out.vocab_dim;
     Vector<BaseFloat>& logprobs = out.logprobs;
 
-    VLOG(2) << "Forward out " << logprobs.Dim() / vocab_dim  << " decoder frames.";
+    VLOG(2) << "Forward out " << logprobs.Dim() / vocab_dim
+            << " decoder frames.";
     // cache nnet outupts
     nnet_out_cache_.Resize(logprobs.Dim() / vocab_dim, vocab_dim);
     nnet_out_cache_.CopyRowsFromVec(logprobs);
@@ -127,7 +128,9 @@ bool Decodable::FrameLikelihood(int32 frame, vector<BaseFloat>* likelihood) {
         (*likelihood)[idx] =
             nnet_out_cache_(frame - frame_offset_, idx) * acoustic_scale_;
 
-        VLOG(4) << "nnet out: " << frame  << " offset:" << frame_offset_  << " " << nnet_out_cache_.NumRows() << " logprob: " <<  nnet_out_cache_(frame - frame_offset_, idx);
+        VLOG(4) << "nnet out: " << frame << " offset:" << frame_offset_ << " "
+                << nnet_out_cache_.NumRows()
+                << " logprob: " << nnet_out_cache_(frame - frame_offset_, idx);
     }
     return true;
 }

speechx/speechx/nnet/u2_nnet.cc

@@ -13,7 +13,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-// modified from https://github.com/wenet-e2e/wenet/blob/main/runtime/core/decoder/asr_model.cc
+// modified from
+// https://github.com/wenet-e2e/wenet/blob/main/runtime/core/decoder/asr_model.cc
 
 #include "nnet/u2_nnet.h"
 
@@ -129,7 +130,7 @@ U2Nnet::U2Nnet(const U2Nnet& other) {
     forward_attention_decoder_ = other.forward_attention_decoder_;
     ctc_activation_ = other.ctc_activation_;
 
-    offset_ = other.offset_; 
+    offset_ = other.offset_;
 
     // copy model ptr
     model_ = other.model_;
@@ -626,8 +627,10 @@ void U2Nnet::AttentionRescoring(const std::vector<std::vector<int>>& hyps,
         // combinded left-to-right and right-to-lfet score
         (*rescoring_score)[i] =
             score * (1 - reverse_weight) + r_score * reverse_weight;
-        VLOG(1) << "hyp " << i << " " << hyp.size() << " score: " << score << " r_score: " << r_score
-                << " reverse_weight: " << reverse_weight << " final score: " << (*rescoring_score)[i];
+        VLOG(1) << "hyp " << i << " " << hyp.size() << " score: " << score
+                << " r_score: " << r_score
+                << " reverse_weight: " << reverse_weight
+                << " final score: " << (*rescoring_score)[i];
     }
 }
 

speechx/speechx/nnet/u2_nnet.h

@@ -13,7 +13,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-// modified from https://github.com/wenet-e2e/wenet/blob/main/runtime/core/decoder/asr_model.h
+// modified from
+// https://github.com/wenet-e2e/wenet/blob/main/runtime/core/decoder/asr_model.h
 #pragma once
 
 #include "base/common.h"

speechx/speechx/recognizer/u2_recognizer.cc

@@ -190,12 +190,15 @@ void U2Recognizer::AttentionRescoring() {
     // combine ctc score and rescoring score
     for (size_t i = 0; i < num_hyps; i++) {
         VLOG(1) << "hyp " << i << " rescoring_score: " << rescoring_score[i]
-                << " ctc_score: " << result_[i].score << " rescoring_weight: " <<  opts_.decoder_opts.rescoring_weight << " ctc_weight: " <<  opts_.decoder_opts.ctc_weight;
+                << " ctc_score: " << result_[i].score
+                << " rescoring_weight: " << opts_.decoder_opts.rescoring_weight
+                << " ctc_weight: " << opts_.decoder_opts.ctc_weight;
         result_[i].score =
             opts_.decoder_opts.rescoring_weight * rescoring_score[i] +
             opts_.decoder_opts.ctc_weight * result_[i].score;
 
-        VLOG(1) << "hyp: " << result_[0].sentence << " score: " << result_[0].score;
+        VLOG(1) << "hyp: " << result_[0].sentence
+                << " score: " << result_[0].score;
     }
 
     std::sort(result_.begin(), result_.end(), DecodeResult::CompareFunc);

speechx/speechx/recognizer/u2_recognizer.h

@@ -96,13 +96,14 @@ struct U2RecognizerResource {
         U2RecognizerResource resource;
         resource.vocab_path = FLAGS_vocab_path;
         resource.acoustic_scale = FLAGS_acoustic_scale;
-        LOG(INFO) << "vocab path: " <<  resource.vocab_path;
-        LOG(INFO) << "acoustic_scale: " <<  resource.acoustic_scale;
+        LOG(INFO) << "vocab path: " << resource.vocab_path;
+        LOG(INFO) << "acoustic_scale: " << resource.acoustic_scale;
 
         resource.feature_pipeline_opts =
             ppspeech::FeaturePipelineOptions::InitFromFlags();
         resource.feature_pipeline_opts.assembler_opts.fill_zero = false;
-        LOG(INFO) << "u2 need fill zero be false: " << resource.feature_pipeline_opts.assembler_opts.fill_zero;
+        LOG(INFO) << "u2 need fill zero be false: "
+                  << resource.feature_pipeline_opts.assembler_opts.fill_zero;
         resource.model_opts = ppspeech::ModelOptions::InitFromFlags();
         resource.decoder_opts = ppspeech::DecodeOptions::InitFromFlags();
         return resource;

speechx/speechx/recognizer/u2_recognizer_main.cc

@@ -78,7 +78,8 @@ int main(int argc, char* argv[]) {
                 recognizer.SetFinished();
             }
             recognizer.Decode();
-            LOG(INFO) << "Pratial result: " << cnt << " " << recognizer.GetPartialResult();
+            LOG(INFO) << "Pratial result: " << cnt << " "
+                      << recognizer.GetPartialResult();
 
             // no overlap
             sample_offset += cur_chunk_size;
@@ -88,7 +89,7 @@ int main(int argc, char* argv[]) {
 
         // second pass decoding
         recognizer.Rescoring();
-    
+
         std::string result = recognizer.GetFinalResult();
 
         recognizer.Reset();

speechx/speechx/utils/math.cc

@@ -79,10 +79,10 @@ void TopK(const std::vector<T>& data,
     int cur = values->size() - 1;
     while (!pq.empty()) {
         const auto& item = pq.top();
-    
+
         (*values)[cur] = item.first;
         (*indices)[cur] = item.second;
-         
+
         // item if reference, must pop here
         pq.pop();