msb-public
/
PaddleSpeech
mirror of https://github.com/PaddlePaddle/PaddleSpeech
1
0
Code Issues Projects Releases Wiki Activity

fixed overload , test=doc

Browse Source
pull/1475/head
WilliamZhang06 2 years ago
parent b8f16ac9b0
commit 7ebe904e20
4 changed files with 3 additions and 195 deletions
Show Stats Download Patch File Download Diff File

2
paddlespeech/server/conf/asr/asr.yaml
Unescape View File

@ -4,4 +4,4 @@ sample_rate: 16000
cfg_path: # [optional]
ckpt_path: # [optional]
decode_method: 'attention_rescoring'
force_yes: False
force_yes: True

4
paddlespeech/server/conf/asr/asr_pd.yaml
Unescape View File

@ -6,13 +6,13 @@
# am choices=['deepspeech2offline_aishell'] TODO
##################################################################
model_type: 'deepspeech2offline_aishell'
am_model: # the pdmodel file of am static model [optional]
am_model: # the pdmodel file of am static model [optional]
am_params: # the pdiparams file of am static model [optional]
lang: 'zh'
sample_rate: 16000
cfg_path:
decode_method:
force_yes:
force_yes: True
am_predictor_conf:
use_gpu: True

97
paddlespeech/server/engine/asr/paddleinference/asr_engine.py
Unescape View File

@ -60,108 +60,11 @@ pretrained_models = {
}
class ASRServerExecutor(ASRExecutor):
def __init__(self):
super().__init__()
pass
def _check(self, audio_file: str, sample_rate: int, force_yes: bool):
self.sample_rate = sample_rate
if self.sample_rate != 16000 and self.sample_rate != 8000:
logger.error("please input --sr 8000 or --sr 16000")
return False
logger.info("checking the audio file format......")
try:
audio, audio_sample_rate = soundfile.read(
audio_file, dtype="int16", always_2d=True)
except Exception as e:
logger.exception(e)
logger.error(
"can not open the audio file, please check the audio file format is 'wav'. \n \
you can try to use sox to change the file format.\n \
For example: \n \
sample rate: 16k \n \
sox input_audio.xx --rate 16k --bits 16 --channels 1 output_audio.wav \n \
sample rate: 8k \n \
sox input_audio.xx --rate 8k --bits 16 --channels 1 output_audio.wav \n \
")
logger.info("The sample rate is %d" % audio_sample_rate)
if audio_sample_rate != self.sample_rate:
logger.warning("The sample rate of the input file is not {}.\n \
The program will resample the wav file to {}.\n \
If the result does not meet your expectations，\n \
Please input the 16k 16 bit 1 channel wav file. \
".format(self.sample_rate, self.sample_rate))
self.change_format = True
else:
logger.info("The audio file format is right")
self.change_format = False
return True
def preprocess(self, model_type: str, input: Union[str, os.PathLike]):
"""
Input preprocess and return paddle.Tensor stored in self.input.
Input content can be a text(tts), a file(asr, cls) or a streaming(not supported yet).
"""
audio_file = input
# Get the object for feature extraction
if "deepspeech2online" in model_type or "deepspeech2offline" in model_type:
audio, _ = self.collate_fn_test.process_utterance(
audio_file=audio_file, transcript=" ")
audio_len = audio.shape[0]
audio = paddle.to_tensor(audio, dtype='float32')
audio_len = paddle.to_tensor(audio_len)
audio = paddle.unsqueeze(audio, axis=0)
self._inputs["audio"] = audio
self._inputs["audio_len"] = audio_len
logger.info(f"audio feat shape: {audio.shape}")
elif "conformer" in model_type or "transformer" in model_type or "wenetspeech" in model_type:
logger.info("get the preprocess conf")
preprocess_conf = self.config.preprocess_config
preprocess_args = {"train": False}
preprocessing = Transformation(preprocess_conf)
logger.info("read the audio file")
audio, audio_sample_rate = soundfile.read(
audio_file, dtype="int16", always_2d=True)
if self.change_format:
if audio.shape[1] >= 2:
audio = audio.mean(axis=1, dtype=np.int16)
else:
audio = audio[:, 0]
# pcm16 -> pcm 32
audio = self._pcm16to32(audio)
audio = librosa.resample(audio, audio_sample_rate,
self.sample_rate)
audio_sample_rate = self.sample_rate
# pcm32 -> pcm 16
audio = self._pcm32to16(audio)
else:
audio = audio[:, 0]
logger.info(f"audio shape: {audio.shape}")
# fbank
audio = preprocessing(audio, **preprocess_args)
audio_len = paddle.to_tensor(audio.shape[0])
audio = paddle.to_tensor(audio, dtype='float32').unsqueeze(axis=0)
self._inputs["audio"] = audio
self._inputs["audio_len"] = audio_len
logger.info(f"audio feat shape: {audio.shape}")
else:
raise Exception("wrong type")
def _init_from_path(self,
model_type: str='wenetspeech',
am_model: Optional[os.PathLike]=None,

95
paddlespeech/server/engine/asr/python/asr_engine.py
Unescape View File

@ -38,101 +38,6 @@ class ASRServerExecutor(ASRExecutor):
super().__init__()
pass
def _check(self, audio_file: str, sample_rate: int, force_yes: bool):
self.sample_rate = sample_rate
if self.sample_rate != 16000 and self.sample_rate != 8000:
logger.error("please input --sr 8000 or --sr 16000")
return False
logger.info("checking the audio file format......")
try:
audio, audio_sample_rate = soundfile.read(
audio_file, dtype="int16", always_2d=True)
except Exception as e:
logger.exception(e)
logger.error(
"can not open the audio file, please check the audio file format is 'wav'. \n \
you can try to use sox to change the file format.\n \
For example: \n \
sample rate: 16k \n \
sox input_audio.xx --rate 16k --bits 16 --channels 1 output_audio.wav \n \
sample rate: 8k \n \
sox input_audio.xx --rate 8k --bits 16 --channels 1 output_audio.wav \n \
")
logger.info("The sample rate is %d" % audio_sample_rate)
if audio_sample_rate != self.sample_rate:
logger.warning("The sample rate of the input file is not {}.\n \
The program will resample the wav file to {}.\n \
If the result does not meet your expectations，\n \
Please input the 16k 16 bit 1 channel wav file. \
".format(self.sample_rate, self.sample_rate))
self.change_format = True
else:
logger.info("The audio file format is right")
self.change_format = False
return True
def preprocess(self, model_type: str, input: Union[str, os.PathLike]):
"""
Input preprocess and return paddle.Tensor stored in self.input.
Input content can be a text(tts), a file(asr, cls) or a streaming(not supported yet).
"""
audio_file = input
# Get the object for feature extraction
if "deepspeech2online" in model_type or "deepspeech2offline" in model_type:
audio, _ = self.collate_fn_test.process_utterance(
audio_file=audio_file, transcript=" ")
audio_len = audio.shape[0]
audio = paddle.to_tensor(audio, dtype='float32')
audio_len = paddle.to_tensor(audio_len)
audio = paddle.unsqueeze(audio, axis=0)
# vocab_list = collate_fn_test.vocab_list
self._inputs["audio"] = audio
self._inputs["audio_len"] = audio_len
logger.info(f"audio feat shape: {audio.shape}")
elif "conformer" in model_type or "transformer" in model_type or "wenetspeech" in model_type:
logger.info("get the preprocess conf")
preprocess_conf = self.config.preprocess_config
preprocess_args = {"train": False}
preprocessing = Transformation(preprocess_conf)
logger.info("read the audio file")
audio, audio_sample_rate = soundfile.read(
audio_file, dtype="int16", always_2d=True)
if self.change_format:
if audio.shape[1] >= 2:
audio = audio.mean(axis=1, dtype=np.int16)
else:
audio = audio[:, 0]
# pcm16 -> pcm 32
audio = self._pcm16to32(audio)
audio = librosa.resample(audio, audio_sample_rate,
self.sample_rate)
audio_sample_rate = self.sample_rate
# pcm32 -> pcm 16
audio = self._pcm32to16(audio)
else:
audio = audio[:, 0]
logger.info(f"audio shape: {audio.shape}")
# fbank
audio = preprocessing(audio, **preprocess_args)
audio_len = paddle.to_tensor(audio.shape[0])
audio = paddle.to_tensor(audio, dtype='float32').unsqueeze(axis=0)
self._inputs["audio"] = audio
self._inputs["audio_len"] = audio_len
logger.info(f"audio feat shape: {audio.shape}")
else:
raise Exception("wrong type")
class ASREngine(BaseEngine):
"""ASR server engine

Write Preview
Loading…
Cancel
Save