fix #2013; and format

examples/wenetspeech/asr1/local/extract_meta.py

@@ -13,7 +13,6 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-
 import argparse
 import json
 import os

paddlespeech/__init__.py

@@ -14,7 +14,3 @@
 import _locale
 
 _locale._getdefaultlocale = (lambda *args: ['en_US', 'utf8'])
-
-
-
-

paddlespeech/cli/base_commands.py

@@ -145,4 +145,3 @@ for com, info in _commands.items():
         name='paddlespeech.{}'.format(com),
         description=info[0],
         cls='paddlespeech.cli.{}.{}'.format(com, info[1]))
-        

paddlespeech/cli/cls/infer.py

@@ -21,12 +21,12 @@ from typing import Union
 import numpy as np
 import paddle
 import yaml
+from paddleaudio import load
+from paddleaudio.features import LogMelSpectrogram
 
 from ..executor import BaseExecutor
 from ..log import logger
 from ..utils import stats_wrapper
-from paddleaudio import load
-from paddleaudio.features import LogMelSpectrogram
 
 __all__ = ['CLSExecutor']
 

paddlespeech/cli/vector/infer.py

@@ -22,13 +22,13 @@ from typing import Union
 
 import paddle
 import soundfile
+from paddleaudio.backends import load as load_audio
+from paddleaudio.compliance.librosa import melspectrogram
 from yacs.config import CfgNode
 
 from ..executor import BaseExecutor
 from ..log import logger
 from ..utils import stats_wrapper
-from paddleaudio.backends import load as load_audio
-from paddleaudio.compliance.librosa import melspectrogram
 from paddlespeech.vector.io.batch import feature_normalize
 from paddlespeech.vector.modules.sid_model import SpeakerIdetification
 

paddlespeech/resource/model_alias.py

@@ -22,8 +22,7 @@ model_alias = {
     # -------------- ASR --------------
     # ---------------------------------
     "deepspeech2offline": ["paddlespeech.s2t.models.ds2:DeepSpeech2Model"],
-    "deepspeech2online":
+    "deepspeech2online": ["paddlespeech.s2t.models.ds2:DeepSpeech2Model"],
-    ["paddlespeech.s2t.models.ds2:DeepSpeech2Model"],
     "conformer": ["paddlespeech.s2t.models.u2:U2Model"],
     "conformer_online": ["paddlespeech.s2t.models.u2:U2Model"],
     "transformer": ["paddlespeech.s2t.models.u2:U2Model"],

paddlespeech/s2t/decoders/scorers/ctc_prefix_score.py

@@ -76,7 +76,8 @@ class CTCPrefixScorePD():
         last_ids = [yi[-1] for yi in y]  # last output label ids
         n_bh = len(last_ids)  # batch * hyps
         n_hyps = n_bh // self.batch  # assuming each utterance has the same # of hyps
-        self.scoring_num = paddle.shape(scoring_ids)[-1] if scoring_ids is not None else 0
+        self.scoring_num = paddle.shape(scoring_ids)[
+            -1] if scoring_ids is not None else 0
         # prepare state info
         if state is None:
             r_prev = paddle.full(

paddlespeech/s2t/exps/deepspeech2/model.py

@@ -22,11 +22,9 @@ import numpy as np
 import paddle
 from paddle import distributed as dist
 from paddle import inference
-from paddle.io import DataLoader
-from paddlespeech.s2t.io.dataloader import BatchDataLoader
 
 from paddlespeech.s2t.frontend.featurizer.text_featurizer import TextFeaturizer
-from paddlespeech.s2t.io.dataset import ManifestDataset
+from paddlespeech.s2t.io.dataloader import BatchDataLoader
 from paddlespeech.s2t.models.ds2 import DeepSpeech2InferModel
 from paddlespeech.s2t.models.ds2 import DeepSpeech2Model
 from paddlespeech.s2t.training.gradclip import ClipGradByGlobalNormWithLog
@@ -238,8 +236,7 @@ class DeepSpeech2Tester(DeepSpeech2Trainer):
     def __init__(self, config, args):
         super().__init__(config, args)
         self._text_featurizer = TextFeaturizer(
-            unit_type=config.unit_type,
+            unit_type=config.unit_type, vocab=config.vocab_filepath)
-            vocab=config.vocab_filepath)
         self.vocab_list = self._text_featurizer.vocab_list
 
     def ordid2token(self, texts, texts_len):
@@ -248,7 +245,8 @@ class DeepSpeech2Tester(DeepSpeech2Trainer):
         for text, n in zip(texts, texts_len):
             n = n.numpy().item()
             ids = text[:n]
-            trans.append(self._text_featurizer.defeaturize(ids.numpy().tolist()))
+            trans.append(
+                self._text_featurizer.defeaturize(ids.numpy().tolist()))
         return trans
 
     def compute_metrics(self,

paddlespeech/s2t/models/ds2/__init__.py

@@ -11,10 +11,11 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+import sys
+
 from .deepspeech2 import DeepSpeech2InferModel
 from .deepspeech2 import DeepSpeech2Model
 from paddlespeech.s2t.utils import dynamic_pip_install
-import sys
 
 try:
     import paddlespeech_ctcdecoders

paddlespeech/s2t/models/ds2/deepspeech2.py

@@ -372,11 +372,15 @@ class DeepSpeech2InferModel(DeepSpeech2Model):
     def __init__(self, *args, **kwargs):
         super().__init__(*args, **kwargs)
 
-    def forward(self, audio_chunk, audio_chunk_lens, chunk_state_h_box=None,
+    def forward(self,
+                audio_chunk,
+                audio_chunk_lens,
+                chunk_state_h_box=None,
                 chunk_state_c_box=None):
         if self.encoder.rnn_direction == "forward":
             eouts_chunk, eouts_chunk_lens, final_state_h_box, final_state_c_box = self.encoder(
-                audio_chunk, audio_chunk_lens, chunk_state_h_box, chunk_state_c_box)
+                audio_chunk, audio_chunk_lens, chunk_state_h_box,
+                chunk_state_c_box)
             probs_chunk = self.decoder.softmax(eouts_chunk)
             return probs_chunk, eouts_chunk_lens, final_state_h_box, final_state_c_box
         elif self.encoder.rnn_direction == "bidirect":
@@ -392,8 +396,8 @@ class DeepSpeech2InferModel(DeepSpeech2Model):
                 self,
                 input_spec=[
                     paddle.static.InputSpec(
-                        shape=[None, None,
+                        shape=[None, None, self.encoder.feat_size
-                               self.encoder.feat_size],  #[B, chunk_size, feat_dim]
+                               ],  #[B, chunk_size, feat_dim]
                         dtype='float32'),
                     paddle.static.InputSpec(shape=[None],
                                             dtype='int64'),  # audio_length, [B]

paddlespeech/s2t/models/lm/transformer.py

@@ -90,7 +90,7 @@ class TransformerLM(nn.Layer, LMInterface, BatchScorerInterface):
 
     def _target_mask(self, ys_in_pad):
         ys_mask = ys_in_pad != 0
-        m = subsequent_mask(paddle.shape(ys_mask)[-1])).unsqueeze(0)
+        m = subsequent_mask(paddle.shape(ys_mask)[-1]).unsqueeze(0)
         return ys_mask.unsqueeze(-2) & m
 
     def forward(self, x: paddle.Tensor, t: paddle.Tensor

paddlespeech/s2t/models/u2/updater.py

@@ -11,7 +11,6 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-
 from contextlib import nullcontext
 
 import paddle

paddlespeech/s2t/modules/ctc.py

@@ -11,6 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+import sys
 from typing import Union
 
 import paddle
@@ -22,7 +23,6 @@ from paddlespeech.s2t.modules.align import Linear
 from paddlespeech.s2t.modules.loss import CTCLoss
 from paddlespeech.s2t.utils import ctc_utils
 from paddlespeech.s2t.utils.log import Log
-import sys
 
 logger = Log(__name__).getlog()
 

paddlespeech/s2t/utils/tensor_utils.py

@@ -82,7 +82,8 @@ def pad_sequence(sequences: List[paddle.Tensor],
     max_size = paddle.shape(sequences[0])
     # (TODO Hui Zhang): slice not supprot `end==start`
     # trailing_dims = max_size[1:]
-    trailing_dims = tuple(max_size[1:].numpy().tolist()) if sequences[0].ndim >= 2 else ()
+    trailing_dims = tuple(
+        max_size[1:].numpy().tolist()) if sequences[0].ndim >= 2 else ()
     max_len = max([s.shape[0] for s in sequences])
     if batch_first:
         out_dims = (len(sequences), max_len) + trailing_dims

paddlespeech/server/engine/asr/online/asr_engine.py

@@ -55,7 +55,7 @@ class PaddleASRConnectionHanddler:
         self.config = asr_engine.config  # server config
         self.model_config = asr_engine.executor.config
         self.asr_engine = asr_engine
-    
+
         # model_type, sample_rate and text_feature is shared for deepspeech2 and conformer
         self.model_type = self.asr_engine.executor.model_type
         self.sample_rate = self.asr_engine.executor.sample_rate
@@ -191,7 +191,7 @@ class PaddleASRConnectionHanddler:
         self.num_frames = 0
 
         ## endpoint
-        self.endpoint_state = False # True for detect endpoint
+        self.endpoint_state = False  # True for detect endpoint
 
         ## conformer
         self.model_reset()
@@ -503,11 +503,13 @@ class PaddleASRConnectionHanddler:
 
         # endpoint
         if not is_finished:
+
             def contain_nonsilence():
                 return len(self.hyps) > 0 and len(self.hyps[0]) > 0
 
             decoding_something = contain_nonsilence()
-            if self.endpointer.endpoint_detected(ctc_probs.numpy(), decoding_something):
+            if self.endpointer.endpoint_detected(ctc_probs.numpy(),
+                                                 decoding_something):
                 self.endpoint_state = True
                 logger.info(f"Endpoint is detected at {self.num_frames} frame.")
 
@@ -869,7 +871,6 @@ class ASREngine(BaseEngine):
         logger.info("Initialize ASR server engine successfully.")
         return True
 
-
     def new_handler(self):
         """New handler from model.
 

paddlespeech/server/engine/asr/online/ctc_endpoint.py

@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 from dataclasses import dataclass
-from typing import List
+
 import numpy as np
 
 from paddlespeech.cli.log import logger
@@ -76,12 +76,11 @@ class OnlineCTCEndpoint:
             decoding_something or (not rule.must_contain_nonsilence)
         ) and trailine_silence >= rule.min_trailing_silence and utterance_length >= rule.min_utterance_length
         if (ans):
-            logger.info(
+            logger.info(f"Endpoint Rule: {rule_name} activated: {rule}")
-                f"Endpoint Rule: {rule_name} activated: {rule}"
-            )
         return ans
 
-    def endpoint_detected(self, ctc_log_probs: np.ndarray,
+    def endpoint_detected(self,
+                          ctc_log_probs: np.ndarray,
                           decoding_something: bool) -> bool:
         """detect endpoint.
 

paddlespeech/server/engine/tts/online/python/tts_engine.py

@@ -42,7 +42,6 @@ class TTSServerExecutor(TTSExecutor):
         self.task_resource = CommonTaskResource(
             task='tts', model_format='dynamic', inference_mode='online')
 
-
     def get_model_info(self,
                        field: str,
                        model_name: str,

paddlespeech/server/ws/asr_api.py

@@ -19,7 +19,6 @@ from fastapi import WebSocketDisconnect
 from starlette.websockets import WebSocketState as WebSocketState
 
 from paddlespeech.cli.log import logger
-from paddlespeech.server.engine.asr.online.asr_engine import PaddleASRConnectionHanddler
 from paddlespeech.server.engine.engine_pool import get_engine_pool
 router = APIRouter()
 
@@ -106,7 +105,7 @@ async def websocket_endpoint(websocket: WebSocket):
                     logger.info("endpoint: detected and rescoring.")
                     connection_handler.rescoring()
                     word_time_stamp = connection_handler.get_word_time_stamp()
-                
+
                 asr_results = connection_handler.get_result()
 
                 if connection_handler.endpoint_state:
@@ -124,7 +123,7 @@ async def websocket_endpoint(websocket: WebSocket):
                         }
                         await websocket.send_json(resp)
                         break
- 
+
                 # return the current partial result
                 # if the engine create the vad instance, this connection will have many partial results 
                 resp = {'result': asr_results}

paddlespeech/t2s/exps/synthesize.py

@@ -140,10 +140,7 @@ def parse_args():
         ],
         help='Choose acoustic model type of tts task.')
     parser.add_argument(
-        '--am_config',
+        '--am_config', type=str, default=None, help='Config of acoustic model.')
-        type=str,
-        default=None,
-        help='Config of acoustic model.')
     parser.add_argument(
         '--am_ckpt',
         type=str,
@@ -179,10 +176,7 @@ def parse_args():
         ],
         help='Choose vocoder type of tts task.')
     parser.add_argument(
-        '--voc_config',
+        '--voc_config', type=str, default=None, help='Config of voc.')
-        type=str,
-        default=None,
-        help='Config of voc.')
     parser.add_argument(
         '--voc_ckpt', type=str, default=None, help='Checkpoint file of voc.')
     parser.add_argument(

paddlespeech/t2s/exps/synthesize_e2e.py

@@ -174,10 +174,7 @@ def parse_args():
         ],
         help='Choose acoustic model type of tts task.')
     parser.add_argument(
-        '--am_config',
+        '--am_config', type=str, default=None, help='Config of acoustic model.')
-        type=str,
-        default=None,
-        help='Config of acoustic model.')
     parser.add_argument(
         '--am_ckpt',
         type=str,
@@ -220,10 +217,7 @@ def parse_args():
         ],
         help='Choose vocoder type of tts task.')
     parser.add_argument(
-        '--voc_config',
+        '--voc_config', type=str, default=None, help='Config of voc.')
-        type=str,
-        default=None,
-        help='Config of voc.')
     parser.add_argument(
         '--voc_ckpt', type=str, default=None, help='Checkpoint file of voc.')
     parser.add_argument(

paddlespeech/t2s/exps/voice_cloning.py

@@ -131,10 +131,7 @@ def parse_args():
         choices=['fastspeech2_aishell3', 'tacotron2_aishell3'],
         help='Choose acoustic model type of tts task.')
     parser.add_argument(
-        '--am_config',
+        '--am_config', type=str, default=None, help='Config of acoustic model.')
-        type=str,
-        default=None,
-        help='Config of acoustic model.')
     parser.add_argument(
         '--am_ckpt',
         type=str,
@@ -160,10 +157,7 @@ def parse_args():
         help='Choose vocoder type of tts task.')
 
     parser.add_argument(
-        '--voc_config',
+        '--voc_config', type=str, default=None, help='Config of voc.')
-        type=str,
-        default=None,
-        help='Config of voc.')
     parser.add_argument(
         '--voc_ckpt', type=str, default=None, help='Checkpoint file of voc.')
     parser.add_argument(

paddlespeech/t2s/models/vits/__init__.py

@@ -12,4 +12,4 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 from .vits import *
-from .vits_updater import *
+from .vits_updater import *

paddlespeech/t2s/models/vits/vits_updater.py

@@ -56,7 +56,8 @@ class VITSUpdater(StandardUpdater):
         self.models: Dict[str, Layer] = models
         # self.model = model
 
-        self.model = model._layers if isinstance(model, paddle.DataParallel) else model
+        self.model = model._layers if isinstance(model,
+                                                 paddle.DataParallel) else model
 
         self.optimizers = optimizers
         self.optimizer_g: Optimizer = optimizers['generator']
@@ -225,7 +226,8 @@ class VITSEvaluator(StandardEvaluator):
         models = {"main": model}
         self.models: Dict[str, Layer] = models
         # self.model = model
-        self.model = model._layers if isinstance(model, paddle.DataParallel) else model
+        self.model = model._layers if isinstance(model,
+                                                 paddle.DataParallel) else model
 
         self.criterions = criterions
         self.criterion_mel = criterions['mel']

paddlespeech/t2s/modules/losses.py

@@ -971,18 +971,18 @@ class FeatureMatchLoss(nn.Layer):
 
         return feat_match_loss
 
+
 # loss for VITS
 class KLDivergenceLoss(nn.Layer):
     """KL divergence loss."""
 
     def forward(
-        self,
+            self,
-        z_p: paddle.Tensor,
+            z_p: paddle.Tensor,
-        logs_q: paddle.Tensor,
+            logs_q: paddle.Tensor,
-        m_p: paddle.Tensor,
+            m_p: paddle.Tensor,
-        logs_p: paddle.Tensor,
+            logs_p: paddle.Tensor,
-        z_mask: paddle.Tensor,
+            z_mask: paddle.Tensor, ) -> paddle.Tensor:
-    ) -> paddle.Tensor:
         """Calculate KL divergence loss.
 
         Args:
@@ -1002,8 +1002,8 @@ class KLDivergenceLoss(nn.Layer):
         logs_p = paddle.cast(logs_p, 'float32')
         z_mask = paddle.cast(z_mask, 'float32')
         kl = logs_p - logs_q - 0.5
-        kl += 0.5 * ((z_p - m_p) ** 2) * paddle.exp(-2.0 * logs_p)
+        kl += 0.5 * ((z_p - m_p)**2) * paddle.exp(-2.0 * logs_p)
         kl = paddle.sum(kl * z_mask)
         loss = kl / paddle.sum(z_mask)
 
-        return loss
+        return loss

speechx/examples/README.md

@@ -25,4 +25,3 @@ netron exp/deepspeech2_online/checkpoints/avg_1.jit.pdmodel  --port 8022 --host
 > Reminder: Only for developer, make sure you know what's it.
 
 * codelab - for speechx developer, using for test.
-

speechx/examples/ds2_ol/README.md

@@ -3,4 +3,4 @@
 ## Examples
 
 * `websocket` - Streaming ASR with websocket for deepspeech2_aishell.    
-* `aishell` - Streaming Decoding under aishell dataset, for local WER test.    
+* `aishell` - Streaming Decoding under aishell dataset, for local WER test.    

speechx/speechx/codelab/README.md

@@ -4,4 +4,3 @@
 > Reminder: Only for developer.
 
 * codelab - for speechx developer, using for test.
-

speechx/speechx/decoder/ctc_prefix_beam_search_decoder_main.cc

@@ -91,8 +91,8 @@ int main(int argc, char* argv[]) {
     std::shared_ptr<ppspeech::Decodable> decodable(
         new ppspeech::Decodable(nnet, raw_data));
 
-     int32 chunk_size = FLAGS_receptive_field_length
+    int32 chunk_size = FLAGS_receptive_field_length +
-        + (FLAGS_nnet_decoder_chunk - 1) * FLAGS_downsampling_rate;
+                       (FLAGS_nnet_decoder_chunk - 1) * FLAGS_downsampling_rate;
     int32 chunk_stride = FLAGS_downsampling_rate * FLAGS_nnet_decoder_chunk;
     int32 receptive_field_length = FLAGS_receptive_field_length;
     LOG(INFO) << "chunk size (frame): " << chunk_size;

speechx/speechx/decoder/ctc_tlg_decoder.cc

@@ -64,7 +64,7 @@ std::string TLGDecoder::GetPartialResult() {
         std::string word = word_symbol_table_->Find(words_id[idx]);
         words += word;
     }
-    return words; 
+    return words;
 }
 
 std::string TLGDecoder::GetFinalBestPath() {

speechx/speechx/decoder/param.h

@@ -82,7 +82,7 @@ FeaturePipelineOptions InitFeaturePipelineOptions() {
     opts.assembler_opts.subsampling_rate = FLAGS_downsampling_rate;
     opts.assembler_opts.receptive_filed_length = FLAGS_receptive_field_length;
     opts.assembler_opts.nnet_decoder_chunk = FLAGS_nnet_decoder_chunk;
-    
+
     return opts;
 }
 

speechx/speechx/decoder/tlg_decoder_main.cc

@@ -93,8 +93,8 @@ int main(int argc, char* argv[]) {
     std::shared_ptr<ppspeech::Decodable> decodable(
         new ppspeech::Decodable(nnet, raw_data, FLAGS_acoustic_scale));
 
-    int32 chunk_size = FLAGS_receptive_field_length
+    int32 chunk_size = FLAGS_receptive_field_length +
-        + (FLAGS_nnet_decoder_chunk - 1) * FLAGS_downsampling_rate;
+                       (FLAGS_nnet_decoder_chunk - 1) * FLAGS_downsampling_rate;
     int32 chunk_stride = FLAGS_downsampling_rate * FLAGS_nnet_decoder_chunk;
     int32 receptive_field_length = FLAGS_receptive_field_length;
     LOG(INFO) << "chunk size (frame): " << chunk_size;

speechx/speechx/frontend/audio/assembler.cc

@@ -24,7 +24,8 @@ using std::unique_ptr;
 Assembler::Assembler(AssemblerOptions opts,
                      unique_ptr<FrontendInterface> base_extractor) {
     frame_chunk_stride_ = opts.subsampling_rate * opts.nnet_decoder_chunk;
-    frame_chunk_size_ = (opts.nnet_decoder_chunk - 1) * opts.subsampling_rate + opts.receptive_filed_length;
+    frame_chunk_size_ = (opts.nnet_decoder_chunk - 1) * opts.subsampling_rate +
+                        opts.receptive_filed_length;
     receptive_filed_length_ = opts.receptive_filed_length;
     base_extractor_ = std::move(base_extractor);
     dim_ = base_extractor_->Dim();
@@ -50,8 +51,8 @@ bool Assembler::Compute(Vector<BaseFloat>* feats) {
         Vector<BaseFloat> feature;
         result = base_extractor_->Read(&feature);
         if (result == false || feature.Dim() == 0) {
-          if (IsFinished() == false) return false;
+            if (IsFinished() == false) return false;
-          break;
+            break;
         }
         feature_cache_.push(feature);
     }
@@ -61,22 +62,22 @@ bool Assembler::Compute(Vector<BaseFloat>* feats) {
     }
 
     while (feature_cache_.size() < frame_chunk_size_) {
-       Vector<BaseFloat> feature(dim_, kaldi::kSetZero); 
+        Vector<BaseFloat> feature(dim_, kaldi::kSetZero);
-       feature_cache_.push(feature);
+        feature_cache_.push(feature);
     }
 
-    int32 counter = 0; 
+    int32 counter = 0;
     int32 cache_size = frame_chunk_size_ - frame_chunk_stride_;
     int32 elem_dim = base_extractor_->Dim();
     while (counter < frame_chunk_size_) {
-      Vector<BaseFloat>& val = feature_cache_.front();
+        Vector<BaseFloat>& val = feature_cache_.front();
-      int32 start = counter * elem_dim;
+        int32 start = counter * elem_dim;
-      feats->Range(start, elem_dim).CopyFromVec(val);
+        feats->Range(start, elem_dim).CopyFromVec(val);
-      if (frame_chunk_size_ - counter <= cache_size ) {
+        if (frame_chunk_size_ - counter <= cache_size) {
-          feature_cache_.push(val);
+            feature_cache_.push(val);
-      }
+        }
-      feature_cache_.pop();
+        feature_cache_.pop();
-      counter++;
+        counter++;
     }
 
     return result;

speechx/speechx/frontend/audio/assembler.h

@@ -25,7 +25,7 @@ struct AssemblerOptions {
     int32 receptive_filed_length;
     int32 subsampling_rate;
     int32 nnet_decoder_chunk;
-    
+
     AssemblerOptions()
         : receptive_filed_length(1),
           subsampling_rate(1),
@@ -47,15 +47,11 @@ class Assembler : public FrontendInterface {
     // feat dim
     virtual size_t Dim() const { return dim_; }
 
-    virtual void SetFinished() {
+    virtual void SetFinished() { base_extractor_->SetFinished(); }
-        base_extractor_->SetFinished();
-    }
 
     virtual bool IsFinished() const { return base_extractor_->IsFinished(); }
 
-    virtual void Reset() {
+    virtual void Reset() { base_extractor_->Reset(); }
-        base_extractor_->Reset();
-    }
 
   private:
     bool Compute(kaldi::Vector<kaldi::BaseFloat>* feats);

speechx/speechx/frontend/audio/audio_cache.h

@@ -30,7 +30,7 @@ class AudioCache : public FrontendInterface {
 
     virtual bool Read(kaldi::Vector<kaldi::BaseFloat>* waves);
 
-    // the audio dim is 1, one sample, which is useless, 
+    // the audio dim is 1, one sample, which is useless,
     // so we return size_(cache samples) instead.
     virtual size_t Dim() const { return size_; }
 

speechx/speechx/frontend/audio/fbank.cc

@@ -29,19 +29,19 @@ using kaldi::Matrix;
 using std::vector;
 
 FbankComputer::FbankComputer(const Options& opts)
-    : opts_(opts),
+    : opts_(opts), computer_(opts) {}
-    computer_(opts) {}
 
 int32 FbankComputer::Dim() const {
     return opts_.mel_opts.num_bins + (opts_.use_energy ? 1 : 0);
 }
 
 bool FbankComputer::NeedRawLogEnergy() {
-    return opts_.use_energy && opts_.raw_energy; 
+    return opts_.use_energy && opts_.raw_energy;
 }
 
 // Compute feat
-bool FbankComputer::Compute(Vector<BaseFloat>* window, Vector<BaseFloat>* feat) {
+bool FbankComputer::Compute(Vector<BaseFloat>* window,
+                            Vector<BaseFloat>* feat) {
     RealFft(window, true);
     kaldi::ComputePowerSpectrum(window);
     const kaldi::MelBanks& mel_bank = *(computer_.GetMelBanks(1.0));

speechx/speechx/frontend/audio/feature_cache.cc

@@ -72,9 +72,9 @@ bool FeatureCache::Compute() {
     bool result = base_extractor_->Read(&feature);
     if (result == false || feature.Dim() == 0) return false;
 
-    int32 num_chunk = feature.Dim() / dim_ ;
+    int32 num_chunk = feature.Dim() / dim_;
     for (int chunk_idx = 0; chunk_idx < num_chunk; ++chunk_idx) {
-        int32 start = chunk_idx *  dim_;
+        int32 start = chunk_idx * dim_;
         Vector<BaseFloat> feature_chunk(dim_);
         SubVector<BaseFloat> tmp(feature.Data() + start, dim_);
         feature_chunk.CopyFromVec(tmp);

speechx/speechx/frontend/audio/feature_cache.h

@@ -22,9 +22,7 @@ namespace ppspeech {
 struct FeatureCacheOptions {
     int32 max_size;
     int32 timeout;  // ms
-    FeatureCacheOptions()
+    FeatureCacheOptions() : max_size(kint16max), timeout(1) {}
-        : max_size(kint16max),
-          timeout(1) {}
 };
 
 class FeatureCache : public FrontendInterface {

speechx/speechx/frontend/audio/feature_common.h

@@ -23,11 +23,11 @@ template <class F>
 class StreamingFeatureTpl : public FrontendInterface {
   public:
     typedef typename F::Options Options;
-    StreamingFeatureTpl(const Options& opts, 
+    StreamingFeatureTpl(const Options& opts,
                         std::unique_ptr<FrontendInterface> base_extractor);
     virtual void Accept(const kaldi::VectorBase<kaldi::BaseFloat>& waves);
     virtual bool Read(kaldi::Vector<kaldi::BaseFloat>* feats);
-    
+
     // the dim_ is the dim of single frame feature
     virtual size_t Dim() const { return computer_.Dim(); }
 
@@ -39,8 +39,9 @@ class StreamingFeatureTpl : public FrontendInterface {
         base_extractor_->Reset();
         remained_wav_.Resize(0);
     }
+
   private:
-    bool Compute(const kaldi::Vector<kaldi::BaseFloat>& waves, 
+    bool Compute(const kaldi::Vector<kaldi::BaseFloat>& waves,
                  kaldi::Vector<kaldi::BaseFloat>* feats);
     Options opts_;
     std::unique_ptr<FrontendInterface> base_extractor_;

speechx/speechx/frontend/audio/feature_common_inl.h

@@ -16,16 +16,15 @@
 namespace ppspeech {
 
 template <class F>
-StreamingFeatureTpl<F>::StreamingFeatureTpl(const Options& opts, 
+StreamingFeatureTpl<F>::StreamingFeatureTpl(
-                        std::unique_ptr<FrontendInterface> base_extractor):
+    const Options& opts, std::unique_ptr<FrontendInterface> base_extractor)
-                        opts_(opts),
+    : opts_(opts), computer_(opts), window_function_(opts.frame_opts) {
-                        computer_(opts),
-                        window_function_(opts.frame_opts) {
     base_extractor_ = std::move(base_extractor);
 }
 
 template <class F>
-void StreamingFeatureTpl<F>::Accept(const kaldi::VectorBase<kaldi::BaseFloat>& waves) {
+void StreamingFeatureTpl<F>::Accept(
+    const kaldi::VectorBase<kaldi::BaseFloat>& waves) {
     base_extractor_->Accept(waves);
 }
 
@@ -58,8 +57,9 @@ bool StreamingFeatureTpl<F>::Read(kaldi::Vector<kaldi::BaseFloat>* feats) {
 
 // Compute feat
 template <class F>
-bool StreamingFeatureTpl<F>::Compute(const kaldi::Vector<kaldi::BaseFloat>& waves,
+bool StreamingFeatureTpl<F>::Compute(
-                                     kaldi::Vector<kaldi::BaseFloat>* feats) {
+    const kaldi::Vector<kaldi::BaseFloat>& waves,
+    kaldi::Vector<kaldi::BaseFloat>* feats) {
     const kaldi::FrameExtractionOptions& frame_opts =
         computer_.GetFrameOptions();
     int32 num_samples = waves.Dim();
@@ -84,9 +84,11 @@ bool StreamingFeatureTpl<F>::Compute(const kaldi::Vector<kaldi::BaseFloat>& wave
                              &window,
                              need_raw_log_energy ? &raw_log_energy : NULL);
 
-        kaldi::Vector<kaldi::BaseFloat> this_feature(computer_.Dim(), kaldi::kUndefined);
+        kaldi::Vector<kaldi::BaseFloat> this_feature(computer_.Dim(),
+                                                     kaldi::kUndefined);
         computer_.Compute(&window, &this_feature);
-        kaldi::SubVector<kaldi::BaseFloat> output_row(feats->Data() + frame * Dim(), Dim());
+        kaldi::SubVector<kaldi::BaseFloat> output_row(
+            feats->Data() + frame * Dim(), Dim());
         output_row.CopyFromVec(this_feature);
     }
     return true;

speechx/speechx/frontend/audio/feature_pipeline.h

@@ -16,6 +16,7 @@
 
 #pragma once
 
+#include "frontend/audio/assembler.h"
 #include "frontend/audio/audio_cache.h"
 #include "frontend/audio/data_cache.h"
 #include "frontend/audio/fbank.h"
@@ -23,7 +24,6 @@
 #include "frontend/audio/frontend_itf.h"
 #include "frontend/audio/linear_spectrogram.h"
 #include "frontend/audio/normalizer.h"
-#include "frontend/audio/assembler.h"
 
 namespace ppspeech {
 

speechx/speechx/frontend/audio/linear_spectrogram.cc

@@ -28,22 +28,21 @@ using kaldi::VectorBase;
 using kaldi::Matrix;
 using std::vector;
 
-LinearSpectrogramComputer::LinearSpectrogramComputer(
+LinearSpectrogramComputer::LinearSpectrogramComputer(const Options& opts)
-    const Options& opts)
     : opts_(opts) {
     kaldi::FeatureWindowFunction feature_window_function(opts.frame_opts);
     int32 window_size = opts.frame_opts.WindowSize();
     frame_length_ = window_size;
     dim_ = window_size / 2 + 1;
-    BaseFloat hanning_window_energy = kaldi::VecVec(feature_window_function.window,
+    BaseFloat hanning_window_energy = kaldi::VecVec(
-                                          feature_window_function.window);
+        feature_window_function.window, feature_window_function.window);
     int32 sample_rate = opts.frame_opts.samp_freq;
     scale_ = 2.0 / (hanning_window_energy * sample_rate);
 }
 
 // Compute spectrogram feat
 bool LinearSpectrogramComputer::Compute(Vector<BaseFloat>* window,
-                                Vector<BaseFloat>* feat) {
+                                        Vector<BaseFloat>* feat) {
     window->Resize(frame_length_, kaldi::kCopyData);
     RealFft(window, true);
     kaldi::ComputePowerSpectrum(window);

speechx/speechx/nnet/nnet_forward_main.cc

@@ -14,8 +14,8 @@
 
 #include "base/flags.h"
 #include "base/log.h"
-#include "frontend/audio/data_cache.h"
 #include "frontend/audio/assembler.h"
+#include "frontend/audio/data_cache.h"
 #include "kaldi/util/table-types.h"
 #include "nnet/decodable.h"
 #include "nnet/paddle_nnet.h"
@@ -75,8 +75,8 @@ int main(int argc, char* argv[]) {
     std::shared_ptr<ppspeech::Decodable> decodable(
         new ppspeech::Decodable(nnet, raw_data, FLAGS_acoustic_scale));
 
-    int32 chunk_size = FLAGS_receptive_field_length 
+    int32 chunk_size = FLAGS_receptive_field_length +
-        + (FLAGS_nnet_decoder_chunk - 1) * FLAGS_downsampling_rate;
+                       (FLAGS_nnet_decoder_chunk - 1) * FLAGS_downsampling_rate;
     int32 chunk_stride = FLAGS_downsampling_rate * FLAGS_nnet_decoder_chunk;
     int32 receptive_field_length = FLAGS_receptive_field_length;
     LOG(INFO) << "chunk size (frame): " << chunk_size;
@@ -130,7 +130,9 @@ int main(int argc, char* argv[]) {
             vector<kaldi::BaseFloat> prob;
             while (decodable->FrameLikelihood(frame_idx, &prob)) {
                 kaldi::Vector<kaldi::BaseFloat> vec_tmp(prob.size());
-                std::memcpy(vec_tmp.Data(), prob.data(), sizeof(kaldi::BaseFloat)*prob.size());
+                std::memcpy(vec_tmp.Data(),
+                            prob.data(),
+                            sizeof(kaldi::BaseFloat) * prob.size());
                 prob_vec.push_back(vec_tmp);
                 frame_idx++;
             }
@@ -142,7 +144,8 @@ int main(int argc, char* argv[]) {
             KALDI_LOG << " the nnet prob of " << utt << " is empty";
             continue;
         }
-        kaldi::Matrix<kaldi::BaseFloat> result(prob_vec.size(),prob_vec[0].Dim());
+        kaldi::Matrix<kaldi::BaseFloat> result(prob_vec.size(),
+                                               prob_vec[0].Dim());
         for (int32 row_idx = 0; row_idx < prob_vec.size(); ++row_idx) {
             for (int32 col_idx = 0; col_idx < prob_vec[0].Dim(); ++col_idx) {
                 result(row_idx, col_idx) = prob_vec[row_idx](col_idx);

speechx/speechx/protocol/websocket/websocket_client.h

@@ -40,8 +40,8 @@ class WebSocketClient {
     void SendEndSignal();
     void SendDataEnd();
     bool Done() const { return done_; }
-    std::string GetResult() const { return result_; } 
+    std::string GetResult() const { return result_; }
-    std::string GetPartialResult() const { return partial_result_;}
+    std::string GetPartialResult() const { return partial_result_; }
 
   private:
     void Connect();

speechx/speechx/protocol/websocket/websocket_server.cc

@@ -76,9 +76,10 @@ void ConnectionHandler::OnSpeechData(const beast::flat_buffer& buffer) {
     recognizer_->Accept(pcm_data);
 
     std::string partial_result = recognizer_->GetPartialResult();
-    
+
-    json::value rv = {
+    json::value rv = {{"status", "ok"},
-        {"status", "ok"}, {"type", "partial_result"}, {"result", partial_result}};
+                      {"type", "partial_result"},
+                      {"result", partial_result}};
     ws_.text(true);
     ws_.write(asio::buffer(json::serialize(rv)));
 }