File size: 32,006 Bytes
6b31279
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0c9d7b7
d60566f
 
 
9c3d4ca
b0eec9a
d60566f
6b31279
0c9d7b7
 
 
 
f59be55
0c9d7b7
d633661
 
 
 
6b31279
 
 
 
d633661
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b0eec9a
d633661
b0eec9a
 
 
 
 
 
 
 
 
 
 
 
d633661
b0eec9a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d633661
 
 
 
 
 
 
 
 
 
 
 
e6d227e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b0eec9a
e6d227e
 
 
 
 
 
b0eec9a
 
 
 
 
 
d633661
 
e6d227e
 
b0eec9a
d633661
b0eec9a
 
39b3b3e
0c9d7b7
 
 
 
d633661
39b3b3e
0c9d7b7
 
 
39b3b3e
0c9d7b7
 
 
39b3b3e
0c9d7b7
 
 
fa00390
0c9d7b7
 
 
 
 
 
 
 
 
 
 
e6d227e
 
 
 
756d91b
 
 
 
62e20af
 
 
 
39b3b3e
 
 
 
 
 
 
 
 
6b31279
39b3b3e
 
 
6b31279
39b3b3e
 
6b31279
39b3b3e
 
 
 
 
6b31279
39b3b3e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0c9d7b7
 
 
 
 
39b3b3e
 
 
 
 
ee6ba22
39b3b3e
0c9d7b7
39b3b3e
 
 
0c9d7b7
 
 
 
 
 
 
 
 
 
 
 
 
 
39b3b3e
 
0c9d7b7
 
 
 
c342af5
62e20af
 
 
 
c342af5
 
 
 
62e20af
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
16be569
62e20af
 
 
 
 
 
 
 
 
 
 
 
 
27c18ec
6fb68e3
27c18ec
 
5cb3bb4
4a82406
27c18ec
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
39b3b3e
0c9d7b7
 
 
 
 
39b3b3e
 
ee6ba22
39b3b3e
0c9d7b7
ee6ba22
 
6b31279
0c9d7b7
 
 
 
 
 
 
 
 
 
 
 
 
 
6b31279
09d9587
0c9d7b7
 
 
 
09d9587
39b3b3e
4abdf19
0c9d7b7
 
 
 
39b3b3e
451aa75
0ae65b0
fc7e843
39b3b3e
0c2bc86
18d7e37
f59be55
 
 
4abdf19
9fa23bb
ccb04a3
ee6ba22
39b3b3e
0c2bc86
18d7e37
 
 
 
0c2bc86
 
451aa75
 
 
 
 
 
0ae65b0
 
 
 
 
 
fc7e843
 
 
 
 
 
f59be55
 
 
 
 
 
 
0c9d7b7
39b3b3e
0c2bc86
39b3b3e
4abdf19
 
9fa23bb
 
 
 
4abdf19
 
 
0c9d7b7
 
 
 
 
 
 
 
 
 
 
 
 
39b3b3e
 
0c9d7b7
 
 
 
39b3b3e
 
0c9d7b7
 
 
 
 
39b3b3e
 
ee6ba22
39b3b3e
0c9d7b7
39b3b3e
09d9587
 
0c9d7b7
 
 
 
 
 
 
 
 
 
 
 
 
 
39b3b3e
 
0c9d7b7
 
 
 
39b3b3e
 
9b54310
0c9d7b7
 
 
 
39b3b3e
 
9b54310
ee6ba22
39b3b3e
9b54310
 
 
 
 
 
 
0c9d7b7
39b3b3e
9b54310
09d9587
9b54310
0c9d7b7
 
 
 
 
 
 
 
 
 
 
 
 
09d9587
39b3b3e
0c9d7b7
 
 
 
39b3b3e
f80684a
0c9d7b7
 
 
 
 
f80684a
e491b4f
f80684a
ee6ba22
f80684a
e491b4f
 
 
 
 
0c9d7b7
f80684a
e491b4f
f80684a
0c9d7b7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f80684a
 
0c9d7b7
 
 
 
f80684a
b1459d7
0c9d7b7
 
 
 
 
b1459d7
 
 
 
0c9d7b7
b1459d7
 
 
0c9d7b7
 
 
 
 
 
 
 
 
 
 
 
 
 
b1459d7
 
0c9d7b7
 
 
 
b1459d7
6f26bbb
0c9d7b7
 
 
 
 
6f26bbb
5681f66
6f26bbb
 
 
5681f66
 
 
 
 
0c9d7b7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5681f66
0c9d7b7
 
 
 
6f26bbb
0c9d7b7
 
 
 
 
 
 
6f26bbb
 
0c9d7b7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6f26bbb
e6d227e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c342af5
e6d227e
 
 
 
62e20af
e6d227e
 
 
 
 
 
 
 
 
b0eec9a
 
 
 
 
d633661
b0eec9a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9c3d4ca
 
b0eec9a
 
 
 
 
 
 
 
d633661
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e618c84
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a461307
e618c84
 
 
 
 
39b3b3e
d633661
f80684a
e618c84
e491b4f
39b3b3e
 
b1459d7
376cd19
0c9d7b7
2f9ed4b
39b3b3e
 
 
27c18ec
 
 
7fc713f
39b3b3e
0ae65b0
fc7e843
4abdf19
 
 
 
f59be55
 
 
4abdf19
9fa23bb
ccb04a3
0c9d7b7
39b3b3e
 
 
9b54310
ce37a1c
39b3b3e
 
6f26bbb
5681f66
1f16e2f
6f26bbb
 
0c9d7b7
 
 
 
 
 
 
 
e6d227e
 
 
 
b0eec9a
 
 
 
 
62e20af
 
 
 
 
39b3b3e
 
 
 
3425b9c
6f26bbb
0c9d7b7
 
e6d227e
62e20af
39b3b3e
 
 
f80684a
 
 
3425b9c
6f26bbb
0c9d7b7
 
e6d227e
62e20af
39b3b3e
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
# Copyright      2022  Xiaomi Corp.        (authors: Fangjun Kuang)
#
# See LICENSE for clarification regarding multiple authors
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import os
from functools import lru_cache
from typing import Union

import torch
import torchaudio
from huggingface_hub import hf_hub_download

os.system(
    "cp -v /home/user/.local/lib/python3.8/site-packages/k2/lib/*.so /home/user/.local/lib/python3.8/site-packages/sherpa/lib/"
)

import k2  # noqa
import sherpa
import sherpa_onnx
import numpy as np
from typing import Tuple
import wave

sample_rate = 16000


def read_wave(wave_filename: str) -> Tuple[np.ndarray, int]:
    """
    Args:
      wave_filename:
        Path to a wave file. It should be single channel and each sample should
        be 16-bit. Its sample rate does not need to be 16kHz.
    Returns:
      Return a tuple containing:
       - A 1-D array of dtype np.float32 containing the samples, which are
       normalized to the range [-1, 1].
       - sample rate of the wave file
    """

    with wave.open(wave_filename) as f:
        assert f.getnchannels() == 1, f.getnchannels()
        assert f.getsampwidth() == 2, f.getsampwidth()  # it is in bytes
        num_samples = f.getnframes()
        samples = f.readframes(num_samples)
        samples_int16 = np.frombuffer(samples, dtype=np.int16)
        samples_float32 = samples_int16.astype(np.float32)

        samples_float32 = samples_float32 / 32768
        return samples_float32, f.getframerate()


def decode_offline_recognizer(
    recognizer: sherpa.OfflineRecognizer,
    filename: str,
) -> str:
    s = recognizer.create_stream()

    s.accept_wave_file(filename)
    recognizer.decode_stream(s)

    text = s.result.text.strip()
    return text.lower()


def decode_online_recognizer(
    recognizer: sherpa.OnlineRecognizer,
    filename: str,
) -> str:
    samples, actual_sample_rate = torchaudio.load(filename)
    assert sample_rate == actual_sample_rate, (
        sample_rate,
        actual_sample_rate,
    )
    samples = samples[0].contiguous()

    s = recognizer.create_stream()

    tail_padding = torch.zeros(int(sample_rate * 0.3), dtype=torch.float32)
    s.accept_waveform(sample_rate, samples)
    s.accept_waveform(sample_rate, tail_padding)
    s.input_finished()

    while recognizer.is_ready(s):
        recognizer.decode_stream(s)

    text = recognizer.get_result(s).text
    return text.strip().lower()


def decode_offline_recognizer_sherpa_onnx(
    recognizer: sherpa_onnx.OfflineRecognizer,
    filename: str,
) -> str:
    s = recognizer.create_stream()
    samples, sample_rate = read_wave(filename)
    s.accept_waveform(sample_rate, samples)
    recognizer.decode_stream(s)

    return s.result.text.lower()


def decode_online_recognizer_sherpa_onnx(
    recognizer: sherpa_onnx.OnlineRecognizer,
    filename: str,
) -> str:
    s = recognizer.create_stream()
    samples, sample_rate = read_wave(filename)
    s.accept_waveform(sample_rate, samples)

    tail_paddings = np.zeros(int(0.3 * sample_rate), dtype=np.float32)
    s.accept_waveform(sample_rate, tail_paddings)
    s.input_finished()

    while recognizer.is_ready(s):
        recognizer.decode_stream(s)

    return recognizer.get_result(s).lower()


def decode(
    recognizer: Union[
        sherpa.OfflineRecognizer,
        sherpa.OnlineRecognizer,
        sherpa_onnx.OfflineRecognizer,
        sherpa_onnx.OnlineRecognizer,
    ],
    filename: str,
) -> str:
    if isinstance(recognizer, sherpa.OfflineRecognizer):
        return decode_offline_recognizer(recognizer, filename)
    elif isinstance(recognizer, sherpa.OnlineRecognizer):
        return decode_online_recognizer(recognizer, filename)
    elif isinstance(recognizer, sherpa_onnx.OfflineRecognizer):
        return decode_offline_recognizer_sherpa_onnx(recognizer, filename)
    elif isinstance(recognizer, sherpa_onnx.OnlineRecognizer):
        return decode_online_recognizer_sherpa_onnx(recognizer, filename)
    else:
        raise ValueError(f"Unknown recognizer type {type(recognizer)}")


@lru_cache(maxsize=30)
def get_pretrained_model(
    repo_id: str,
    decoding_method: str,
    num_active_paths: int,
) -> Union[sherpa.OfflineRecognizer, sherpa.OnlineRecognizer]:
    if repo_id in chinese_models:
        return chinese_models[repo_id](
            repo_id, decoding_method=decoding_method, num_active_paths=num_active_paths
        )
    elif repo_id in english_models:
        return english_models[repo_id](
            repo_id, decoding_method=decoding_method, num_active_paths=num_active_paths
        )
    elif repo_id in chinese_english_mixed_models:
        return chinese_english_mixed_models[repo_id](
            repo_id, decoding_method=decoding_method, num_active_paths=num_active_paths
        )
    elif repo_id in tibetan_models:
        return tibetan_models[repo_id](
            repo_id, decoding_method=decoding_method, num_active_paths=num_active_paths
        )
    elif repo_id in arabic_models:
        return arabic_models[repo_id](
            repo_id, decoding_method=decoding_method, num_active_paths=num_active_paths
        )
    elif repo_id in german_models:
        return german_models[repo_id](
            repo_id, decoding_method=decoding_method, num_active_paths=num_active_paths
        )
    elif repo_id in french_models:
        return french_models[repo_id](
            repo_id, decoding_method=decoding_method, num_active_paths=num_active_paths
        )
    elif repo_id in japanese_models:
        return japanese_models[repo_id](
            repo_id, decoding_method=decoding_method, num_active_paths=num_active_paths
        )
    elif repo_id in russian_models:
        return russian_models[repo_id](
            repo_id, decoding_method=decoding_method, num_active_paths=num_active_paths
        )
    else:
        raise ValueError(f"Unsupported repo_id: {repo_id}")


def _get_nn_model_filename(
    repo_id: str,
    filename: str,
    subfolder: str = "exp",
) -> str:
    nn_model_filename = hf_hub_download(
        repo_id=repo_id,
        filename=filename,
        subfolder=subfolder,
    )
    return nn_model_filename


def _get_bpe_model_filename(
    repo_id: str,
    filename: str = "bpe.model",
    subfolder: str = "data/lang_bpe_500",
) -> str:
    bpe_model_filename = hf_hub_download(
        repo_id=repo_id,
        filename=filename,
        subfolder=subfolder,
    )
    return bpe_model_filename


def _get_token_filename(
    repo_id: str,
    filename: str = "tokens.txt",
    subfolder: str = "data/lang_char",
) -> str:
    token_filename = hf_hub_download(
        repo_id=repo_id,
        filename=filename,
        subfolder=subfolder,
    )
    return token_filename


@lru_cache(maxsize=10)
def _get_aishell2_pretrained_model(
    repo_id: str,
    decoding_method: str,
    num_active_paths: int,
) -> sherpa.OfflineRecognizer:
    assert repo_id in [
        # context-size 1
        "yuekai/icefall-asr-aishell2-pruned-transducer-stateless5-A-2022-07-12",  # noqa
        # context-size 2
        "yuekai/icefall-asr-aishell2-pruned-transducer-stateless5-B-2022-07-12",  # noqa
    ], repo_id

    nn_model = _get_nn_model_filename(
        repo_id=repo_id,
        filename="cpu_jit.pt",
    )
    tokens = _get_token_filename(repo_id=repo_id)

    feat_config = sherpa.FeatureConfig()
    feat_config.fbank_opts.frame_opts.samp_freq = sample_rate
    feat_config.fbank_opts.mel_opts.num_bins = 80
    feat_config.fbank_opts.frame_opts.dither = 0

    config = sherpa.OfflineRecognizerConfig(
        nn_model=nn_model,
        tokens=tokens,
        use_gpu=False,
        feat_config=feat_config,
        decoding_method=decoding_method,
        num_active_paths=num_active_paths,
    )

    recognizer = sherpa.OfflineRecognizer(config)

    return recognizer


@lru_cache(maxsize=10)
def _get_russian_pre_trained_model(
    repo_id: str, decoding_method: str, num_active_paths: int
) -> sherpa_onnx.OfflineRecognizer:
    assert repo_id in (
        "alphacep/vosk-model-ru",
        "alphacep/vosk-model-small-ru",
    ), repo_id

    if repo_id == "alphacep/vosk-model-ru":
        model_dir = "am-onnx"
    elif repo_id == "alphacep/vosk-model-small-ru":
        model_dir = "am"

    encoder_model = _get_nn_model_filename(
        repo_id=repo_id,
        filename="encoder.onnx",
        subfolder=model_dir,
    )

    decoder_model = _get_nn_model_filename(
        repo_id=repo_id,
        filename="decoder.onnx",
        subfolder=model_dir,
    )

    joiner_model = _get_nn_model_filename(
        repo_id=repo_id,
        filename="joiner.onnx",
        subfolder=model_dir,
    )

    tokens = _get_token_filename(repo_id=repo_id, subfolder="lang")

    recognizer = sherpa_onnx.OfflineRecognizer.from_transducer(
        tokens=tokens,
        encoder=encoder_model,
        decoder=decoder_model,
        joiner=joiner_model,
        num_threads=2,
        sample_rate=16000,
        feature_dim=80,
        decoding_method=decoding_method,
    )

    return recognizer


@lru_cache(maxsize=10)
def _get_whisper_model(
    repo_id: str, decoding_method: str, num_active_paths: int
) -> sherpa_onnx.OfflineRecognizer:
    name = repo_id.split("-")[1]
    assert name in ("tiny.en", "base.en", "small.en", "medium.en"), repo_id
    full_repo_id = "csukuangfj/sherpa-onnx-whisper-" + name
    encoder = _get_nn_model_filename(
        repo_id=full_repo_id,
        filename=f"{name}-encoder.int8.ort",
        subfolder=".",
    )

    decoder = _get_nn_model_filename(
        repo_id=full_repo_id,
        filename=f"{name}-decoder.int8.ort",
        subfolder=".",
    )

    tokens = _get_token_filename(
        repo_id=full_repo_id, subfolder=".", filename=f"{name}-tokens.txt"
    )

    recognizer = sherpa_onnx.OfflineRecognizer.from_whisper(
        encoder=encoder,
        decoder=decoder,
        tokens=tokens,
        num_threads=2,
    )

    return recognizer


@lru_cache(maxsize=10)
def _get_gigaspeech_pre_trained_model(
    repo_id: str,
    decoding_method: str,
    num_active_paths: int,
) -> sherpa.OfflineRecognizer:
    assert repo_id in [
        "wgb14/icefall-asr-gigaspeech-pruned-transducer-stateless2",
    ], repo_id

    nn_model = _get_nn_model_filename(
        repo_id=repo_id,
        filename="cpu_jit-iter-3488000-avg-20.pt",
    )
    tokens = "./giga-tokens.txt"

    feat_config = sherpa.FeatureConfig()
    feat_config.fbank_opts.frame_opts.samp_freq = sample_rate
    feat_config.fbank_opts.mel_opts.num_bins = 80
    feat_config.fbank_opts.frame_opts.dither = 0

    config = sherpa.OfflineRecognizerConfig(
        nn_model=nn_model,
        tokens=tokens,
        use_gpu=False,
        feat_config=feat_config,
        decoding_method=decoding_method,
        num_active_paths=num_active_paths,
    )

    recognizer = sherpa.OfflineRecognizer(config)

    return recognizer


@lru_cache(maxsize=10)
def _get_english_model(
    repo_id: str,
    decoding_method: str,
    num_active_paths: int,
) -> sherpa.OfflineRecognizer:
    assert repo_id in [
        "WeijiZhuang/icefall-asr-librispeech-pruned-transducer-stateless8-2022-12-02",  # noqa
        "yfyeung/icefall-asr-multidataset-pruned_transducer_stateless7-2023-05-04",  # noqa
        "yfyeung/icefall-asr-finetune-mux-pruned_transducer_stateless7-2023-05-19",  # noqa
        "csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13",  # noqa
        "csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless7-2022-11-11",  # noqa
        "csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless8-2022-11-14",  # noqa
        "Zengwei/icefall-asr-librispeech-zipformer-large-2023-05-16",  # noqa
        "Zengwei/icefall-asr-librispeech-zipformer-2023-05-15",  # noqa
        "Zengwei/icefall-asr-librispeech-zipformer-small-2023-05-16",  # noqa
        "videodanchik/icefall-asr-tedlium3-conformer-ctc2",
        "pkufool/icefall_asr_librispeech_conformer_ctc",
        "WayneWiser/icefall-asr-librispeech-conformer-ctc2-jit-bpe-500-2022-07-21",
    ], repo_id

    filename = "cpu_jit.pt"
    if (
        repo_id
        == "csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless7-2022-11-11"
    ):
        filename = "cpu_jit-torch-1.10.0.pt"

    if (
        repo_id
        == "WeijiZhuang/icefall-asr-librispeech-pruned-transducer-stateless8-2022-12-02"
    ):
        filename = "cpu_jit-torch-1.10.pt"

    if (
        repo_id
        == "yfyeung/icefall-asr-multidataset-pruned_transducer_stateless7-2023-05-04"
    ):
        filename = "cpu_jit-epoch-30-avg-4.pt"

    if (
        repo_id
        == "yfyeung/icefall-asr-finetune-mux-pruned_transducer_stateless7-2023-05-19"
    ):
        filename = "cpu_jit-epoch-20-avg-5.pt"

    if repo_id in (
        "Zengwei/icefall-asr-librispeech-zipformer-large-2023-05-16",
        "Zengwei/icefall-asr-librispeech-zipformer-2023-05-15",
        "Zengwei/icefall-asr-librispeech-zipformer-small-2023-05-16",
    ):
        filename = "jit_script.pt"

    nn_model = _get_nn_model_filename(
        repo_id=repo_id,
        filename=filename,
    )
    subfolder = "data/lang_bpe_500"

    if repo_id in (
        "videodanchik/icefall-asr-tedlium3-conformer-ctc2",
        "pkufool/icefall_asr_librispeech_conformer_ctc",
    ):
        subfolder = "data/lang_bpe"

    tokens = _get_token_filename(repo_id=repo_id, subfolder=subfolder)

    feat_config = sherpa.FeatureConfig()
    feat_config.fbank_opts.frame_opts.samp_freq = sample_rate
    feat_config.fbank_opts.mel_opts.num_bins = 80
    feat_config.fbank_opts.frame_opts.dither = 0

    config = sherpa.OfflineRecognizerConfig(
        nn_model=nn_model,
        tokens=tokens,
        use_gpu=False,
        feat_config=feat_config,
        decoding_method=decoding_method,
        num_active_paths=num_active_paths,
    )

    recognizer = sherpa.OfflineRecognizer(config)

    return recognizer


@lru_cache(maxsize=10)
def _get_wenetspeech_pre_trained_model(
    repo_id: str,
    decoding_method: str,
    num_active_paths: int,
):
    assert repo_id in [
        "luomingshuang/icefall_asr_wenetspeech_pruned_transducer_stateless2",
    ], repo_id

    nn_model = _get_nn_model_filename(
        repo_id=repo_id,
        filename="cpu_jit_epoch_10_avg_2_torch_1.7.1.pt",
    )
    tokens = _get_token_filename(repo_id=repo_id)

    feat_config = sherpa.FeatureConfig()
    feat_config.fbank_opts.frame_opts.samp_freq = sample_rate
    feat_config.fbank_opts.mel_opts.num_bins = 80
    feat_config.fbank_opts.frame_opts.dither = 0

    config = sherpa.OfflineRecognizerConfig(
        nn_model=nn_model,
        tokens=tokens,
        use_gpu=False,
        feat_config=feat_config,
        decoding_method=decoding_method,
        num_active_paths=num_active_paths,
    )

    recognizer = sherpa.OfflineRecognizer(config)

    return recognizer


@lru_cache(maxsize=10)
def _get_chinese_english_mixed_model(
    repo_id: str,
    decoding_method: str,
    num_active_paths: int,
):
    assert repo_id in [
        "luomingshuang/icefall_asr_tal-csasr_pruned_transducer_stateless5",
        "ptrnull/icefall-asr-conv-emformer-transducer-stateless2-zh",
    ], repo_id

    if repo_id == "luomingshuang/icefall_asr_tal-csasr_pruned_transducer_stateless5":
        filename = "cpu_jit.pt"
        subfolder = "data/lang_char"
    elif repo_id == "ptrnull/icefall-asr-conv-emformer-transducer-stateless2-zh":
        filename = "cpu_jit-epoch-11-avg-1.pt"
        subfolder = "data/lang_char_bpe"

    nn_model = _get_nn_model_filename(
        repo_id=repo_id,
        filename=filename,
    )
    tokens = _get_token_filename(repo_id=repo_id, subfolder=subfolder)

    feat_config = sherpa.FeatureConfig()
    feat_config.fbank_opts.frame_opts.samp_freq = sample_rate
    feat_config.fbank_opts.mel_opts.num_bins = 80
    feat_config.fbank_opts.frame_opts.dither = 0

    config = sherpa.OfflineRecognizerConfig(
        nn_model=nn_model,
        tokens=tokens,
        use_gpu=False,
        feat_config=feat_config,
        decoding_method=decoding_method,
        num_active_paths=num_active_paths,
    )

    recognizer = sherpa.OfflineRecognizer(config)

    return recognizer


@lru_cache(maxsize=10)
def _get_alimeeting_pre_trained_model(
    repo_id: str,
    decoding_method: str,
    num_active_paths: int,
):
    assert repo_id in [
        "desh2608/icefall-asr-alimeeting-pruned-transducer-stateless7",
        "luomingshuang/icefall_asr_alimeeting_pruned_transducer_stateless2",
    ], repo_id

    if repo_id == "desh2608/icefall-asr-alimeeting-pruned-transducer-stateless7":
        filename = "cpu_jit.pt"
    elif repo_id == "luomingshuang/icefall_asr_alimeeting_pruned_transducer_stateless2":
        filename = "cpu_jit_torch_1.7.1.pt"

    nn_model = _get_nn_model_filename(
        repo_id=repo_id,
        filename=filename,
    )
    tokens = _get_token_filename(repo_id=repo_id)

    feat_config = sherpa.FeatureConfig()
    feat_config.fbank_opts.frame_opts.samp_freq = sample_rate
    feat_config.fbank_opts.mel_opts.num_bins = 80
    feat_config.fbank_opts.frame_opts.dither = 0

    config = sherpa.OfflineRecognizerConfig(
        nn_model=nn_model,
        tokens=tokens,
        use_gpu=False,
        feat_config=feat_config,
        decoding_method=decoding_method,
        num_active_paths=num_active_paths,
    )

    recognizer = sherpa.OfflineRecognizer(config)

    return recognizer


@lru_cache(maxsize=10)
def _get_wenet_model(
    repo_id: str,
    decoding_method: str,
    num_active_paths: int,
):
    assert repo_id in [
        "csukuangfj/wenet-chinese-model",
        "csukuangfj/wenet-english-model",
    ], repo_id

    nn_model = _get_nn_model_filename(
        repo_id=repo_id,
        filename="final.zip",
        subfolder=".",
    )
    tokens = _get_token_filename(
        repo_id=repo_id,
        filename="units.txt",
        subfolder=".",
    )

    feat_config = sherpa.FeatureConfig(normalize_samples=False)
    feat_config.fbank_opts.frame_opts.samp_freq = sample_rate
    feat_config.fbank_opts.mel_opts.num_bins = 80
    feat_config.fbank_opts.frame_opts.dither = 0

    config = sherpa.OfflineRecognizerConfig(
        nn_model=nn_model,
        tokens=tokens,
        use_gpu=False,
        feat_config=feat_config,
        decoding_method=decoding_method,
        num_active_paths=num_active_paths,
    )

    recognizer = sherpa.OfflineRecognizer(config)

    return recognizer


@lru_cache(maxsize=10)
def _get_aidatatang_200zh_pretrained_mode(
    repo_id: str,
    decoding_method: str,
    num_active_paths: int,
):
    assert repo_id in [
        "luomingshuang/icefall_asr_aidatatang-200zh_pruned_transducer_stateless2",
    ], repo_id

    nn_model = _get_nn_model_filename(
        repo_id=repo_id,
        filename="cpu_jit_torch.1.7.1.pt",
    )
    tokens = _get_token_filename(repo_id=repo_id)

    feat_config = sherpa.FeatureConfig()
    feat_config.fbank_opts.frame_opts.samp_freq = sample_rate
    feat_config.fbank_opts.mel_opts.num_bins = 80
    feat_config.fbank_opts.frame_opts.dither = 0

    config = sherpa.OfflineRecognizerConfig(
        nn_model=nn_model,
        tokens=tokens,
        use_gpu=False,
        feat_config=feat_config,
        decoding_method=decoding_method,
        num_active_paths=num_active_paths,
    )

    recognizer = sherpa.OfflineRecognizer(config)

    return recognizer


@lru_cache(maxsize=10)
def _get_tibetan_pre_trained_model(
    repo_id: str,
    decoding_method: str,
    num_active_paths: int,
):
    assert repo_id in [
        "syzym/icefall-asr-xbmu-amdo31-pruned-transducer-stateless7-2022-12-02",
        "syzym/icefall-asr-xbmu-amdo31-pruned-transducer-stateless5-2022-11-29",
    ], repo_id

    filename = "cpu_jit.pt"
    if (
        repo_id
        == "syzym/icefall-asr-xbmu-amdo31-pruned-transducer-stateless5-2022-11-29"
    ):
        filename = "cpu_jit-epoch-28-avg-23-torch-1.10.0.pt"

    nn_model = _get_nn_model_filename(
        repo_id=repo_id,
        filename=filename,
    )

    tokens = _get_token_filename(repo_id=repo_id, subfolder="data/lang_bpe_500")

    feat_config = sherpa.FeatureConfig()
    feat_config.fbank_opts.frame_opts.samp_freq = sample_rate
    feat_config.fbank_opts.mel_opts.num_bins = 80
    feat_config.fbank_opts.frame_opts.dither = 0

    config = sherpa.OfflineRecognizerConfig(
        nn_model=nn_model,
        tokens=tokens,
        use_gpu=False,
        feat_config=feat_config,
        decoding_method=decoding_method,
        num_active_paths=num_active_paths,
    )

    recognizer = sherpa.OfflineRecognizer(config)

    return recognizer


@lru_cache(maxsize=10)
def _get_arabic_pre_trained_model(
    repo_id: str,
    decoding_method: str,
    num_active_paths: int,
):
    assert repo_id in [
        "AmirHussein/icefall-asr-mgb2-conformer_ctc-2022-27-06",
    ], repo_id

    nn_model = _get_nn_model_filename(
        repo_id=repo_id,
        filename="cpu_jit.pt",
    )

    tokens = _get_token_filename(repo_id=repo_id, subfolder="data/lang_bpe_5000")

    feat_config = sherpa.FeatureConfig()
    feat_config.fbank_opts.frame_opts.samp_freq = sample_rate
    feat_config.fbank_opts.mel_opts.num_bins = 80
    feat_config.fbank_opts.frame_opts.dither = 0

    config = sherpa.OfflineRecognizerConfig(
        nn_model=nn_model,
        tokens=tokens,
        use_gpu=False,
        feat_config=feat_config,
        decoding_method=decoding_method,
        num_active_paths=num_active_paths,
    )

    recognizer = sherpa.OfflineRecognizer(config)

    return recognizer


@lru_cache(maxsize=10)
def _get_german_pre_trained_model(
    repo_id: str,
    decoding_method: str,
    num_active_paths: int,
):
    assert repo_id in [
        "csukuangfj/wav2vec2.0-torchaudio",
    ], repo_id

    nn_model = _get_nn_model_filename(
        repo_id=repo_id,
        filename="voxpopuli_asr_base_10k_de.pt",
        subfolder=".",
    )

    tokens = _get_token_filename(
        repo_id=repo_id,
        filename="tokens-de.txt",
        subfolder=".",
    )

    config = sherpa.OfflineRecognizerConfig(
        nn_model=nn_model,
        tokens=tokens,
        use_gpu=False,
        decoding_method=decoding_method,
        num_active_paths=num_active_paths,
    )

    recognizer = sherpa.OfflineRecognizer(config)

    return recognizer


@lru_cache(maxsize=10)
def _get_french_pre_trained_model(
    repo_id: str,
    decoding_method: str,
    num_active_paths: int,
):
    assert repo_id in [
        "shaojieli/sherpa-onnx-streaming-zipformer-fr-2023-04-14",
    ], repo_id

    encoder_model = _get_nn_model_filename(
        repo_id=repo_id,
        filename="encoder-epoch-29-avg-9-with-averaged-model.onnx",
        subfolder=".",
    )

    decoder_model = _get_nn_model_filename(
        repo_id=repo_id,
        filename="decoder-epoch-29-avg-9-with-averaged-model.onnx",
        subfolder=".",
    )

    joiner_model = _get_nn_model_filename(
        repo_id=repo_id,
        filename="joiner-epoch-29-avg-9-with-averaged-model.onnx",
        subfolder=".",
    )

    tokens = _get_token_filename(repo_id=repo_id, subfolder=".")

    recognizer = sherpa_onnx.OnlineRecognizer.from_transducer(
        tokens=tokens,
        encoder=encoder_model,
        decoder=decoder_model,
        joiner=joiner_model,
        num_threads=2,
        sample_rate=16000,
        feature_dim=80,
        decoding_method=decoding_method,
        max_active_paths=num_active_paths,
    )

    return recognizer


@lru_cache(maxsize=10)
def _get_japanese_pre_trained_model(
    repo_id: str,
    decoding_method: str,
    num_active_paths: int,
) -> sherpa.OnlineRecognizer:
    repo_id, kind = repo_id.rsplit("-", maxsplit=1)

    assert repo_id in [
        "TeoWenShen/icefall-asr-csj-pruned-transducer-stateless7-streaming-230208"
    ], repo_id
    assert kind in ("fluent", "disfluent"), kind

    encoder_model = _get_nn_model_filename(
        repo_id=repo_id, filename="encoder_jit_trace.pt", subfolder=f"exp_{kind}"
    )

    decoder_model = _get_nn_model_filename(
        repo_id=repo_id, filename="decoder_jit_trace.pt", subfolder=f"exp_{kind}"
    )

    joiner_model = _get_nn_model_filename(
        repo_id=repo_id, filename="joiner_jit_trace.pt", subfolder=f"exp_{kind}"
    )

    tokens = _get_token_filename(repo_id=repo_id)

    feat_config = sherpa.FeatureConfig()
    feat_config.fbank_opts.frame_opts.samp_freq = sample_rate
    feat_config.fbank_opts.mel_opts.num_bins = 80
    feat_config.fbank_opts.frame_opts.dither = 0

    config = sherpa.OnlineRecognizerConfig(
        nn_model="",
        encoder_model=encoder_model,
        decoder_model=decoder_model,
        joiner_model=joiner_model,
        tokens=tokens,
        use_gpu=False,
        feat_config=feat_config,
        decoding_method=decoding_method,
        num_active_paths=num_active_paths,
        chunk_size=32,
    )

    recognizer = sherpa.OnlineRecognizer(config)

    return recognizer


@lru_cache(maxsize=10)
def _get_paraformer_zh_pre_trained_model(
    repo_id: str,
    decoding_method: str,
    num_active_paths: int,
) -> sherpa_onnx.OfflineRecognizer:
    assert repo_id in [
        "csukuangfj/sherpa-onnx-paraformer-zh-2023-03-28",
    ], repo_id

    nn_model = _get_nn_model_filename(
        repo_id=repo_id,
        filename="model.onnx",
        subfolder=".",
    )

    tokens = _get_token_filename(repo_id=repo_id, subfolder=".")

    recognizer = sherpa_onnx.OfflineRecognizer.from_paraformer(
        paraformer=nn_model,
        tokens=tokens,
        num_threads=2,
        sample_rate=sample_rate,
        feature_dim=80,
        decoding_method="greedy_search",
        debug=False,
    )

    return recognizer


def _get_multi_zh_hans_pre_trained_model(
    repo_id: str,
    decoding_method: str,
    num_active_paths: int,
) -> sherpa_onnx.OfflineRecognizer:
    assert repo_id in ("zrjin/sherpa-onnx-zipformer-multi-zh-hans-2023-9-2",), repo_id

    encoder_model = _get_nn_model_filename(
        repo_id=repo_id,
        filename="encoder-epoch-20-avg-1.onnx",
        subfolder=".",
    )

    decoder_model = _get_nn_model_filename(
        repo_id=repo_id,
        filename="decoder-epoch-20-avg-1.onnx",
        subfolder=".",
    )

    joiner_model = _get_nn_model_filename(
        repo_id=repo_id,
        filename="joiner-epoch-20-avg-1.onnx",
        subfolder=".",
    )

    tokens = _get_token_filename(repo_id=repo_id, subfolder=".")

    recognizer = sherpa_onnx.OfflineRecognizer.from_transducer(
        tokens=tokens,
        encoder=encoder_model,
        decoder=decoder_model,
        joiner=joiner_model,
        num_threads=2,
        sample_rate=16000,
        feature_dim=80,
        decoding_method=decoding_method,
        max_active_paths=num_active_paths,
    )

    return recognizer


chinese_models = {
    "csukuangfj/sherpa-onnx-paraformer-zh-2023-03-28": _get_paraformer_zh_pre_trained_model,
    "luomingshuang/icefall_asr_wenetspeech_pruned_transducer_stateless2": _get_wenetspeech_pre_trained_model,  # noqa
    "zrjin/sherpa-onnx-zipformer-multi-zh-hans-2023-9-2": _get_multi_zh_hans_pre_trained_model,  # noqa
    "desh2608/icefall-asr-alimeeting-pruned-transducer-stateless7": _get_alimeeting_pre_trained_model,
    "yuekai/icefall-asr-aishell2-pruned-transducer-stateless5-A-2022-07-12": _get_aishell2_pretrained_model,  # noqa
    "yuekai/icefall-asr-aishell2-pruned-transducer-stateless5-B-2022-07-12": _get_aishell2_pretrained_model,  # noqa
    "luomingshuang/icefall_asr_aidatatang-200zh_pruned_transducer_stateless2": _get_aidatatang_200zh_pretrained_mode,  # noqa
    "luomingshuang/icefall_asr_alimeeting_pruned_transducer_stateless2": _get_alimeeting_pre_trained_model,  # noqa
    "csukuangfj/wenet-chinese-model": _get_wenet_model,
    #  "csukuangfj/icefall-asr-wenetspeech-lstm-transducer-stateless-2022-10-14": _get_lstm_transducer_model,
}

english_models = {
    "whisper-tiny.en": _get_whisper_model,
    "whisper-base.en": _get_whisper_model,
    "whisper-small.en": _get_whisper_model,
    #  "whisper-medium.en": _get_whisper_model,
    "wgb14/icefall-asr-gigaspeech-pruned-transducer-stateless2": _get_gigaspeech_pre_trained_model,  # noqa
    "yfyeung/icefall-asr-multidataset-pruned_transducer_stateless7-2023-05-04": _get_english_model,  # noqa
    "yfyeung/icefall-asr-finetune-mux-pruned_transducer_stateless7-2023-05-19": _get_english_model,  # noqa
    "WeijiZhuang/icefall-asr-librispeech-pruned-transducer-stateless8-2022-12-02": _get_english_model,  # noqa
    "csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless8-2022-11-14": _get_english_model,  # noqa
    "csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless7-2022-11-11": _get_english_model,  # noqa
    "csukuangfj/icefall-asr-librispeech-pruned-transducer-stateless3-2022-05-13": _get_english_model,  # noqa
    "Zengwei/icefall-asr-librispeech-zipformer-large-2023-05-16": _get_english_model,  # noqa
    "Zengwei/icefall-asr-librispeech-zipformer-2023-05-15": _get_english_model,  # noqa
    "Zengwei/icefall-asr-librispeech-zipformer-small-2023-05-16": _get_english_model,  # noqa
    "videodanchik/icefall-asr-tedlium3-conformer-ctc2": _get_english_model,
    "pkufool/icefall_asr_librispeech_conformer_ctc": _get_english_model,
    "WayneWiser/icefall-asr-librispeech-conformer-ctc2-jit-bpe-500-2022-07-21": _get_english_model,
    "csukuangfj/wenet-english-model": _get_wenet_model,
}

chinese_english_mixed_models = {
    "ptrnull/icefall-asr-conv-emformer-transducer-stateless2-zh": _get_chinese_english_mixed_model,
    "luomingshuang/icefall_asr_tal-csasr_pruned_transducer_stateless5": _get_chinese_english_mixed_model,  # noqa
}

tibetan_models = {
    "syzym/icefall-asr-xbmu-amdo31-pruned-transducer-stateless7-2022-12-02": _get_tibetan_pre_trained_model,  # noqa
    "syzym/icefall-asr-xbmu-amdo31-pruned-transducer-stateless5-2022-11-29": _get_tibetan_pre_trained_model,  # noqa
}

arabic_models = {
    "AmirHussein/icefall-asr-mgb2-conformer_ctc-2022-27-06": _get_arabic_pre_trained_model,  # noqa
}

german_models = {
    "csukuangfj/wav2vec2.0-torchaudio": _get_german_pre_trained_model,
}

french_models = {
    "shaojieli/sherpa-onnx-streaming-zipformer-fr-2023-04-14": _get_french_pre_trained_model,
}

japanese_models = {
    "TeoWenShen/icefall-asr-csj-pruned-transducer-stateless7-streaming-230208-fluent": _get_japanese_pre_trained_model,
    "TeoWenShen/icefall-asr-csj-pruned-transducer-stateless7-streaming-230208-disfluent": _get_japanese_pre_trained_model,
}

russian_models = {
    "alphacep/vosk-model-ru": _get_russian_pre_trained_model,
    "alphacep/vosk-model-small-ru": _get_russian_pre_trained_model,
}

all_models = {
    **chinese_models,
    **english_models,
    **chinese_english_mixed_models,
    #  **japanese_models,
    **tibetan_models,
    **arabic_models,
    **german_models,
    **french_models,
    **russian_models,
}

language_to_models = {
    "Chinese": list(chinese_models.keys()),
    "English": list(english_models.keys()),
    "Chinese+English": list(chinese_english_mixed_models.keys()),
    #  "Japanese": list(japanese_models.keys()),
    "Tibetan": list(tibetan_models.keys()),
    "Arabic": list(arabic_models.keys()),
    "German": list(german_models.keys()),
    "French": list(french_models.keys()),
    "Russian": list(russian_models.keys()),
}