csukuangfj
commited on
Commit
·
b11b38b
1
Parent(s):
24c2ef7
Add Thai
Browse files- examples.py +24 -0
- model.py +52 -0
- test_wavs/thai/0.wav +0 -0
- test_wavs/thai/1.wav +0 -0
- test_wavs/thai/2.wav +0 -0
- test_wavs/thai/trans.txt +3 -0
examples.py
CHANGED
@@ -72,6 +72,14 @@ examples = [
|
|
72 |
"No",
|
73 |
"./test_wavs/russian/russian-i-love-you.wav",
|
74 |
],
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
75 |
[
|
76 |
"Russian",
|
77 |
"alphacep/vosk-model-ru",
|
@@ -437,4 +445,20 @@ examples = [
|
|
437 |
"No",
|
438 |
"./test_wavs/korean/3.wav",
|
439 |
],
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
440 |
]
|
|
|
72 |
"No",
|
73 |
"./test_wavs/russian/russian-i-love-you.wav",
|
74 |
],
|
75 |
+
[
|
76 |
+
"Thai",
|
77 |
+
"yfyeung/icefall-asr-gigaspeech2-th-zipformer-2024-06-20",
|
78 |
+
"greedy_search",
|
79 |
+
4,
|
80 |
+
"No",
|
81 |
+
"./test_wavs/thai/0.wav",
|
82 |
+
],
|
83 |
[
|
84 |
"Russian",
|
85 |
"alphacep/vosk-model-ru",
|
|
|
445 |
"No",
|
446 |
"./test_wavs/korean/3.wav",
|
447 |
],
|
448 |
+
[
|
449 |
+
"Thai",
|
450 |
+
"yfyeung/icefall-asr-gigaspeech2-th-zipformer-2024-06-20",
|
451 |
+
"greedy_search",
|
452 |
+
4,
|
453 |
+
"No",
|
454 |
+
"./test_wavs/thai/1.wav",
|
455 |
+
],
|
456 |
+
[
|
457 |
+
"Thai",
|
458 |
+
"yfyeung/icefall-asr-gigaspeech2-th-zipformer-2024-06-20",
|
459 |
+
"greedy_search",
|
460 |
+
4,
|
461 |
+
"No",
|
462 |
+
"./test_wavs/thai/2.wav",
|
463 |
+
],
|
464 |
]
|
model.py
CHANGED
@@ -212,6 +212,10 @@ def get_pretrained_model(
|
|
212 |
return korean_models[repo_id](
|
213 |
repo_id, decoding_method=decoding_method, num_active_paths=num_active_paths
|
214 |
)
|
|
|
|
|
|
|
|
|
215 |
else:
|
216 |
raise ValueError(f"Unsupported repo_id: {repo_id}")
|
217 |
|
@@ -293,6 +297,48 @@ def _get_aishell2_pretrained_model(
|
|
293 |
return recognizer
|
294 |
|
295 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
296 |
@lru_cache(maxsize=10)
|
297 |
def _get_zrjin_cantonese_pre_trained_model(
|
298 |
repo_id: str, decoding_method: str, num_active_paths: int
|
@@ -1365,6 +1411,10 @@ korean_models = {
|
|
1365 |
"k2-fsa/sherpa-onnx-streaming-zipformer-korean-2024-06-16": _get_streaming_zipformer_pre_trained_model,
|
1366 |
}
|
1367 |
|
|
|
|
|
|
|
|
|
1368 |
|
1369 |
all_models = {
|
1370 |
**chinese_models,
|
@@ -1379,6 +1429,7 @@ all_models = {
|
|
1379 |
**french_models,
|
1380 |
**russian_models,
|
1381 |
**korean_models,
|
|
|
1382 |
}
|
1383 |
|
1384 |
language_to_models = {
|
@@ -1395,4 +1446,5 @@ language_to_models = {
|
|
1395 |
"French": list(french_models.keys()),
|
1396 |
"Russian": list(russian_models.keys()),
|
1397 |
"Korean": list(korean_models.keys()),
|
|
|
1398 |
}
|
|
|
212 |
return korean_models[repo_id](
|
213 |
repo_id, decoding_method=decoding_method, num_active_paths=num_active_paths
|
214 |
)
|
215 |
+
elif repo_id in thai_models:
|
216 |
+
return thai_models[repo_id](
|
217 |
+
repo_id, decoding_method=decoding_method, num_active_paths=num_active_paths
|
218 |
+
)
|
219 |
else:
|
220 |
raise ValueError(f"Unsupported repo_id: {repo_id}")
|
221 |
|
|
|
297 |
return recognizer
|
298 |
|
299 |
|
300 |
+
@lru_cache(maxsize=10)
|
301 |
+
def _get_yifan_thai_pretrained_model(
|
302 |
+
repo_id: str, decoding_method: str, num_active_paths: int
|
303 |
+
) -> sherpa_onnx.OfflineRecognizer:
|
304 |
+
assert repo_id in (
|
305 |
+
"yfyeung/icefall-asr-gigaspeech2-th-zipformer-2024-06-20",
|
306 |
+
), repo_id
|
307 |
+
|
308 |
+
encoder_model = _get_nn_model_filename(
|
309 |
+
repo_id=repo_id,
|
310 |
+
filename="encoder-epoch-12-avg-5.int8.onnx",
|
311 |
+
subfolder="exp",
|
312 |
+
)
|
313 |
+
|
314 |
+
decoder_model = _get_nn_model_filename(
|
315 |
+
repo_id=repo_id,
|
316 |
+
filename="decoder-epoch-12-avg-5.onnx",
|
317 |
+
subfolder="exp",
|
318 |
+
)
|
319 |
+
|
320 |
+
joiner_model = _get_nn_model_filename(
|
321 |
+
repo_id=repo_id,
|
322 |
+
filename="joiner-epoch-12-avg-5.int8.onnx",
|
323 |
+
subfolder="exp",
|
324 |
+
)
|
325 |
+
|
326 |
+
tokens = _get_token_filename(repo_id=repo_id, subfolder="data/lang_bpe_2000")
|
327 |
+
|
328 |
+
recognizer = sherpa_onnx.OfflineRecognizer.from_transducer(
|
329 |
+
tokens=tokens,
|
330 |
+
encoder=encoder_model,
|
331 |
+
decoder=decoder_model,
|
332 |
+
joiner=joiner_model,
|
333 |
+
num_threads=2,
|
334 |
+
sample_rate=16000,
|
335 |
+
feature_dim=80,
|
336 |
+
decoding_method=decoding_method,
|
337 |
+
)
|
338 |
+
|
339 |
+
return recognizer
|
340 |
+
|
341 |
+
|
342 |
@lru_cache(maxsize=10)
|
343 |
def _get_zrjin_cantonese_pre_trained_model(
|
344 |
repo_id: str, decoding_method: str, num_active_paths: int
|
|
|
1411 |
"k2-fsa/sherpa-onnx-streaming-zipformer-korean-2024-06-16": _get_streaming_zipformer_pre_trained_model,
|
1412 |
}
|
1413 |
|
1414 |
+
thai_models = {
|
1415 |
+
"yfyeung/icefall-asr-gigaspeech2-th-zipformer-2024-06-20": _get_yifan_thai_pretrained_model,
|
1416 |
+
}
|
1417 |
+
|
1418 |
|
1419 |
all_models = {
|
1420 |
**chinese_models,
|
|
|
1429 |
**french_models,
|
1430 |
**russian_models,
|
1431 |
**korean_models,
|
1432 |
+
**thai_models,
|
1433 |
}
|
1434 |
|
1435 |
language_to_models = {
|
|
|
1446 |
"French": list(french_models.keys()),
|
1447 |
"Russian": list(russian_models.keys()),
|
1448 |
"Korean": list(korean_models.keys()),
|
1449 |
+
"Thai": list(thai_models.keys()),
|
1450 |
}
|
test_wavs/thai/0.wav
ADDED
Binary file (144 kB). View file
|
|
test_wavs/thai/1.wav
ADDED
Binary file (137 kB). View file
|
|
test_wavs/thai/2.wav
ADDED
Binary file (316 kB). View file
|
|
test_wavs/thai/trans.txt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
0 ก็เดี๋ยวเกมในนัดต่อไปต้องไปเจอกับทางอินโดนีเซียนะครับ
|
2 |
+
1 ก็ไม่ได้เน้นเรื่องของผลการแข่งขันอยู่แล้วครับเหมือนที่คาร์ลอสเซซาร์นั้นได้บอกไว้นะครับ
|
3 |
+
2 เกมในเกมที่แล้วเนี่ยตอนพักครึ่งหลังเนี่ยเหมือนคาร์ลอสจะบอกว่าจริงจริงจะไม่ส่งมูฮัมหมัดลงด้วยซ้ําแล้วนะครับแต่ว่าเหมือนกับท้ายเกมเนี่ยส่งไปด้วยความมั่นใจแล้วโอ้โหประตูที่สาม
|