EngineX-Iluvatar/enginex_bi_series-sherpa-onnx
Archived
16
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

Add Python APIs for WeNet CTC models (#428)

Browse Source
This commit is contained in:
Fangjun Kuang
2023-11-16 14:20:41 +08:00
committed by GitHub
parent fac4f6bc7c
commit 049fb9f451
13 changed files with 538 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

38
python-api-examples/offline-decode-files.py
View File

@@ -59,7 +59,16 @@ python3 ./python-api-examples/offline-decode-files.py \
./sherpa-onnx-whisper-base.en/test_wavs/1.wav \
./sherpa-onnx-whisper-base.en/test_wavs/8k.wav
(5) For tdnn models of the yesno recipe from icefall
(5) For CTC models from WeNet
python3 ./python-api-examples/offline-decode-files.py \
--wenet-ctc=./sherpa-onnx-zh-wenet-wenetspeech/model.onnx \
--tokens=./sherpa-onnx-zh-wenet-wenetspeech/tokens.txt \
./sherpa-onnx-zh-wenet-wenetspeech/test_wavs/0.wav \
./sherpa-onnx-zh-wenet-wenetspeech/test_wavs/1.wav \
./sherpa-onnx-zh-wenet-wenetspeech/test_wavs/8k.wav
(6) For tdnn models of the yesno recipe from icefall
python3 ./python-api-examples/offline-decode-files.py \
--sample-rate=8000 \
@@ -154,6 +163,13 @@ def get_args():
help="Path to the model.onnx from NeMo CTC",
)
parser.add_argument(
"--wenet-ctc",
default="",
type=str,
help="Path to the model.onnx from WeNet CTC",
)
parser.add_argument(
"--tdnn-model",
default="",
@@ -254,6 +270,7 @@ def assert_file_exists(filename: str):
"https://k2-fsa.github.io/sherpa/onnx/pretrained_models/index.html to download it"
)
def read_wave(wave_filename: str) -> Tuple[np.ndarray, int]:
"""
Args:
@@ -287,6 +304,7 @@ def main():
if args.encoder:
assert len(args.paraformer) == 0, args.paraformer
assert len(args.nemo_ctc) == 0, args.nemo_ctc
assert len(args.wenet_ctc) == 0, args.wenet_ctc
assert len(args.whisper_encoder) == 0, args.whisper_encoder
assert len(args.whisper_decoder) == 0, args.whisper_decoder
assert len(args.tdnn_model) == 0, args.tdnn_model
@@ -310,6 +328,7 @@ def main():
)
elif args.paraformer:
assert len(args.nemo_ctc) == 0, args.nemo_ctc
assert len(args.wenet_ctc) == 0, args.wenet_ctc
assert len(args.whisper_encoder) == 0, args.whisper_encoder
assert len(args.whisper_decoder) == 0, args.whisper_decoder
assert len(args.tdnn_model) == 0, args.tdnn_model
@@ -326,6 +345,7 @@ def main():
debug=args.debug,
)
elif args.nemo_ctc:
assert len(args.wenet_ctc) == 0, args.wenet_ctc
assert len(args.whisper_encoder) == 0, args.whisper_encoder
assert len(args.whisper_decoder) == 0, args.whisper_decoder
assert len(args.tdnn_model) == 0, args.tdnn_model
@@ -341,6 +361,22 @@ def main():
decoding_method=args.decoding_method,
debug=args.debug,
)
elif args.wenet_ctc:
assert len(args.whisper_encoder) == 0, args.whisper_encoder
assert len(args.whisper_decoder) == 0, args.whisper_decoder
assert len(args.tdnn_model) == 0, args.tdnn_model
assert_file_exists(args.wenet_ctc)
recognizer = sherpa_onnx.OfflineRecognizer.from_wenet_ctc(
model=args.wenet_ctc,
tokens=args.tokens,
num_threads=args.num_threads,
sample_rate=args.sample_rate,
feature_dim=args.feature_dim,
decoding_method=args.decoding_method,
debug=args.debug,
)
elif args.whisper_encoder:
assert len(args.tdnn_model) == 0, args.tdnn_model
assert_file_exists(args.whisper_encoder)