Spaces:
Sleeping
Sleeping
update
Browse files- examples/wenet/infer.py +0 -1
- examples/wenet/toolbox_infer.py +22 -9
examples/wenet/infer.py
CHANGED
|
@@ -58,7 +58,6 @@ def main():
|
|
| 58 |
decoding_method="greedy_search",
|
| 59 |
num_active_paths=2,
|
| 60 |
)
|
| 61 |
-
|
| 62 |
recognizer = sherpa.OfflineRecognizer(config)
|
| 63 |
|
| 64 |
signal, sample_rate = librosa.load(args.filename, sr=args.sample_rate)
|
|
|
|
| 58 |
decoding_method="greedy_search",
|
| 59 |
num_active_paths=2,
|
| 60 |
)
|
|
|
|
| 61 |
recognizer = sherpa.OfflineRecognizer(config)
|
| 62 |
|
| 63 |
signal, sample_rate = librosa.load(args.filename, sr=args.sample_rate)
|
examples/wenet/toolbox_infer.py
CHANGED
|
@@ -69,24 +69,37 @@ def main():
|
|
| 69 |
nn_model_file = local_model_dir / m_dict["nn_model_file"]
|
| 70 |
tokens_file = local_model_dir / m_dict["tokens_file"]
|
| 71 |
|
| 72 |
-
recognizer = models.load_recognizer(
|
| 73 |
-
|
| 74 |
-
|
| 75 |
-
|
| 76 |
-
|
| 77 |
-
|
| 78 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 79 |
decoding_method="greedy_search",
|
| 80 |
num_active_paths=2,
|
| 81 |
)
|
|
|
|
| 82 |
|
| 83 |
s = recognizer.create_stream()
|
| 84 |
-
|
| 85 |
s.accept_wave_file(
|
| 86 |
temp_file.as_posix()
|
| 87 |
)
|
| 88 |
recognizer.decode_stream(s)
|
| 89 |
-
|
| 90 |
text = s.result.text.strip()
|
| 91 |
text = text.lower()
|
| 92 |
print("text: {}".format(text))
|
|
|
|
| 69 |
nn_model_file = local_model_dir / m_dict["nn_model_file"]
|
| 70 |
tokens_file = local_model_dir / m_dict["tokens_file"]
|
| 71 |
|
| 72 |
+
# recognizer = models.load_recognizer(
|
| 73 |
+
# repo_id=m_dict["repo_id"],
|
| 74 |
+
# nn_model_file=nn_model_file.as_posix(),
|
| 75 |
+
# tokens_file=tokens_file.as_posix(),
|
| 76 |
+
# sub_folder=m_dict["sub_folder"],
|
| 77 |
+
# local_model_dir=local_model_dir,
|
| 78 |
+
# recognizer_type=m_dict["recognizer_type"],
|
| 79 |
+
# decoding_method="greedy_search",
|
| 80 |
+
# num_active_paths=2,
|
| 81 |
+
# )
|
| 82 |
+
|
| 83 |
+
feat_config = sherpa.FeatureConfig(normalize_samples=False)
|
| 84 |
+
feat_config.fbank_opts.frame_opts.samp_freq = args.sample_rate
|
| 85 |
+
feat_config.fbank_opts.mel_opts.num_bins = 80
|
| 86 |
+
feat_config.fbank_opts.frame_opts.dither = 0
|
| 87 |
+
|
| 88 |
+
config = sherpa.OfflineRecognizerConfig(
|
| 89 |
+
nn_model=nn_model_file.as_posix(),
|
| 90 |
+
tokens=tokens_file.as_posix(),
|
| 91 |
+
use_gpu=False,
|
| 92 |
+
feat_config=feat_config,
|
| 93 |
decoding_method="greedy_search",
|
| 94 |
num_active_paths=2,
|
| 95 |
)
|
| 96 |
+
recognizer = sherpa.OfflineRecognizer(config)
|
| 97 |
|
| 98 |
s = recognizer.create_stream()
|
|
|
|
| 99 |
s.accept_wave_file(
|
| 100 |
temp_file.as_posix()
|
| 101 |
)
|
| 102 |
recognizer.decode_stream(s)
|
|
|
|
| 103 |
text = s.result.text.strip()
|
| 104 |
text = text.lower()
|
| 105 |
print("text: {}".format(text))
|