EngineX-Iluvatar/enginex-mr_series-sherpa-onnx
Archived
15
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

Add C++ runtime for non-streaming faster conformer transducer from NeMo. (#854)

Browse Source
This commit is contained in:
Fangjun Kuang
2024-05-10 12:15:39 +08:00
committed by GitHub
parent 5d8c35e44e
commit 17cd3a5f01
31 changed files with 1093 additions and 153 deletions
Download Patch File Download Diff File Expand all files Collapse all files

68
python-api-examples/offline-nemo-ctc-decode-files.py Executable file
View File

@@ -0,0 +1,68 @@
#!/usr/bin/env python3
"""
This file shows how to use a non-streaming CTC model from NeMo
to decode files.
Please download model files from
https://github.com/k2-fsa/sherpa-onnx/releases/tag/asr-models
The example model supports 10 languages and it is converted from
https://catalog.ngc.nvidia.com/orgs/nvidia/teams/nemo/models/stt_multilingual_fastconformer_hybrid_large_pc
"""
from pathlib import Path
import sherpa_onnx
import soundfile as sf
def create_recognizer():
model = "./sherpa-onnx-nemo-fast-conformer-ctc-be-de-en-es-fr-hr-it-pl-ru-uk-20k/model.onnx"
tokens = "./sherpa-onnx-nemo-fast-conformer-ctc-be-de-en-es-fr-hr-it-pl-ru-uk-20k/tokens.txt"
test_wav = "./sherpa-onnx-nemo-fast-conformer-ctc-be-de-en-es-fr-hr-it-pl-ru-uk-20k/test_wavs/de-german.wav"
# test_wav = "./sherpa-onnx-nemo-fast-conformer-ctc-be-de-en-es-fr-hr-it-pl-ru-uk-20k/test_wavs/en-english.wav"
# test_wav = "./sherpa-onnx-nemo-fast-conformer-ctc-be-de-en-es-fr-hr-it-pl-ru-uk-20k/test_wavs/es-spanish.wav"
# test_wav = "./sherpa-onnx-nemo-fast-conformer-ctc-be-de-en-es-fr-hr-it-pl-ru-uk-20k/test_wavs/fr-french.wav"
# test_wav = "./sherpa-onnx-nemo-fast-conformer-ctc-be-de-en-es-fr-hr-it-pl-ru-uk-20k/test_wavs/hr-croatian.wav"
# test_wav = "./sherpa-onnx-nemo-fast-conformer-ctc-be-de-en-es-fr-hr-it-pl-ru-uk-20k/test_wavs/it-italian.wav"
# test_wav = "./sherpa-onnx-nemo-fast-conformer-ctc-be-de-en-es-fr-hr-it-pl-ru-uk-20k/test_wavs/po-polish.wav"
# test_wav = "./sherpa-onnx-nemo-fast-conformer-ctc-be-de-en-es-fr-hr-it-pl-ru-uk-20k/test_wavs/ru-russian.wav"
# test_wav = "./sherpa-onnx-nemo-fast-conformer-ctc-be-de-en-es-fr-hr-it-pl-ru-uk-20k/test_wavs/uk-ukrainian.wav"
if not Path(model).is_file() or not Path(test_wav).is_file():
raise ValueError(
"""Please download model files from
https://github.com/k2-fsa/sherpa-onnx/releases/tag/asr-models
"""
)
return (
sherpa_onnx.OfflineRecognizer.from_nemo_ctc(
model=model,
tokens=tokens,
debug=True,
),
test_wav,
)
def main():
recognizer, wave_filename = create_recognizer()
audio, sample_rate = sf.read(wave_filename, dtype="float32", always_2d=True)
audio = audio[:, 0] # only use the first channel
# audio is a 1-D float32 numpy array normalized to the range [-1, 1]
# sample_rate does not need to be 16000 Hz
stream = recognizer.create_stream()
stream.accept_waveform(sample_rate, audio)
recognizer.decode_stream(stream)
print(wave_filename)
print(stream.result)
if __name__ == "__main__":
main()

73
python-api-examples/offline-nemo-transducer-decode-files.py Executable file
View File

@@ -0,0 +1,73 @@
#!/usr/bin/env python3
"""
This file shows how to use a non-streaming transducer model from NeMo
to decode files.
Please download model files from
https://github.com/k2-fsa/sherpa-onnx/releases/tag/asr-models
The example model supports 10 languages and it is converted from
https://catalog.ngc.nvidia.com/orgs/nvidia/teams/nemo/models/stt_multilingual_fastconformer_hybrid_large_pc
"""
from pathlib import Path
import sherpa_onnx
import soundfile as sf
def create_recognizer():
encoder = "./sherpa-onnx-nemo-fast-conformer-transducer-be-de-en-es-fr-hr-it-pl-ru-uk-20k/encoder.onnx"
decoder = "./sherpa-onnx-nemo-fast-conformer-transducer-be-de-en-es-fr-hr-it-pl-ru-uk-20k/decoder.onnx"
joiner = "./sherpa-onnx-nemo-fast-conformer-transducer-be-de-en-es-fr-hr-it-pl-ru-uk-20k/joiner.onnx"
tokens = "./sherpa-onnx-nemo-fast-conformer-transducer-be-de-en-es-fr-hr-it-pl-ru-uk-20k/tokens.txt"
test_wav = "./sherpa-onnx-nemo-fast-conformer-transducer-be-de-en-es-fr-hr-it-pl-ru-uk-20k/test_wavs/de-german.wav"
# test_wav = "./sherpa-onnx-nemo-fast-conformer-transducer-be-de-en-es-fr-hr-it-pl-ru-uk-20k/test_wavs/en-english.wav"
# test_wav = "./sherpa-onnx-nemo-fast-conformer-transducer-be-de-en-es-fr-hr-it-pl-ru-uk-20k/test_wavs/es-spanish.wav"
# test_wav = "./sherpa-onnx-nemo-fast-conformer-transducer-be-de-en-es-fr-hr-it-pl-ru-uk-20k/test_wavs/fr-french.wav"
# test_wav = "./sherpa-onnx-nemo-fast-conformer-transducer-be-de-en-es-fr-hr-it-pl-ru-uk-20k/test_wavs/hr-croatian.wav"
# test_wav = "./sherpa-onnx-nemo-fast-conformer-transducer-be-de-en-es-fr-hr-it-pl-ru-uk-20k/test_wavs/it-italian.wav"
# test_wav = "./sherpa-onnx-nemo-fast-conformer-transducer-be-de-en-es-fr-hr-it-pl-ru-uk-20k/test_wavs/po-polish.wav"
# test_wav = "./sherpa-onnx-nemo-fast-conformer-transducer-be-de-en-es-fr-hr-it-pl-ru-uk-20k/test_wavs/ru-russian.wav"
# test_wav = "./sherpa-onnx-nemo-fast-conformer-transducer-be-de-en-es-fr-hr-it-pl-ru-uk-20k/test_wavs/uk-ukrainian.wav"
if not Path(encoder).is_file() or not Path(test_wav).is_file():
raise ValueError(
"""Please download model files from
https://github.com/k2-fsa/sherpa-onnx/releases/tag/asr-models
"""
)
return (
sherpa_onnx.OfflineRecognizer.from_transducer(
encoder=encoder,
decoder=decoder,
joiner=joiner,
tokens=tokens,
model_type="nemo_transducer",
debug=True,
),
test_wav,
)
def main():
recognizer, wave_filename = create_recognizer()
audio, sample_rate = sf.read(wave_filename, dtype="float32", always_2d=True)
audio = audio[:, 0] # only use the first channel
# audio is a 1-D float32 numpy array normalized to the range [-1, 1]
# sample_rate does not need to be 16000 Hz
stream = recognizer.create_stream()
stream.accept_waveform(sample_rate, audio)
recognizer.decode_stream(stream)
print(wave_filename)
print(stream.result)
if __name__ == "__main__":
main()