Spaces:
Edmond98
/
Running on A100

vineelpratap commited on
Commit
5442f52
1 Parent(s): a45002a

Update asr.py

Browse files
Files changed (1) hide show
  1. asr.py +45 -43
asr.py CHANGED
@@ -20,50 +20,52 @@ processor = AutoProcessor.from_pretrained(MODEL_ID)
20
  model = Wav2Vec2ForCTC.from_pretrained(MODEL_ID)
21
 
22
 
23
- lm_decoding_config = {}
24
- lm_decoding_configfile = hf_hub_download(
25
- repo_id="facebook/mms-cclms",
26
- filename="decoding_config.json",
27
- subfolder="mms-1b-all",
28
- )
29
-
30
- with open(lm_decoding_configfile) as f:
31
- lm_decoding_config = json.loads(f.read())
32
-
33
- # allow language model decoding for "eng"
34
-
35
- decoding_config = lm_decoding_config["eng"]
36
-
37
- lm_file = hf_hub_download(
38
- repo_id="facebook/mms-cclms",
39
- filename=decoding_config["lmfile"].rsplit("/", 1)[1],
40
- subfolder=decoding_config["lmfile"].rsplit("/", 1)[0],
41
- )
42
- token_file = hf_hub_download(
43
- repo_id="facebook/mms-cclms",
44
- filename=decoding_config["tokensfile"].rsplit("/", 1)[1],
45
- subfolder=decoding_config["tokensfile"].rsplit("/", 1)[0],
46
- )
47
- lexicon_file = None
48
- if decoding_config["lexiconfile"] is not None:
49
- lexicon_file = hf_hub_download(
50
- repo_id="facebook/mms-cclms",
51
- filename=decoding_config["lexiconfile"].rsplit("/", 1)[1],
52
- subfolder=decoding_config["lexiconfile"].rsplit("/", 1)[0],
53
- )
 
 
 
 
 
 
 
 
 
 
 
 
 
 
54
 
55
- beam_search_decoder = ctc_decoder(
56
- lexicon=lexicon_file,
57
- tokens=token_file,
58
- lm=lm_file,
59
- nbest=1,
60
- beam_size=500,
61
- beam_size_token=50,
62
- lm_weight=float(decoding_config["lmweight"]),
63
- word_score=float(decoding_config["wordscore"]),
64
- sil_score=float(decoding_config["silweight"]),
65
- blank_token="<s>",
66
- )
67
 
68
  def transcribe(
69
  audio_source=None, microphone=None, file_upload=None, lang="eng (English)"
 
20
  model = Wav2Vec2ForCTC.from_pretrained(MODEL_ID)
21
 
22
 
23
+ # lm_decoding_config = {}
24
+ # lm_decoding_configfile = hf_hub_download(
25
+ # repo_id="facebook/mms-cclms",
26
+ # filename="decoding_config.json",
27
+ # subfolder="mms-1b-all",
28
+ # )
29
+
30
+ # with open(lm_decoding_configfile) as f:
31
+ # lm_decoding_config = json.loads(f.read())
32
+
33
+ # # allow language model decoding for "eng"
34
+
35
+ # decoding_config = lm_decoding_config["eng"]
36
+
37
+ # lm_file = hf_hub_download(
38
+ # repo_id="facebook/mms-cclms",
39
+ # filename=decoding_config["lmfile"].rsplit("/", 1)[1],
40
+ # subfolder=decoding_config["lmfile"].rsplit("/", 1)[0],
41
+ # )
42
+ # token_file = hf_hub_download(
43
+ # repo_id="facebook/mms-cclms",
44
+ # filename=decoding_config["tokensfile"].rsplit("/", 1)[1],
45
+ # subfolder=decoding_config["tokensfile"].rsplit("/", 1)[0],
46
+ # )
47
+ # lexicon_file = None
48
+ # if decoding_config["lexiconfile"] is not None:
49
+ # lexicon_file = hf_hub_download(
50
+ # repo_id="facebook/mms-cclms",
51
+ # filename=decoding_config["lexiconfile"].rsplit("/", 1)[1],
52
+ # subfolder=decoding_config["lexiconfile"].rsplit("/", 1)[0],
53
+ # )
54
+
55
+ beam_search_decoder = None
56
+ # beam_search_decoder = ctc_decoder(
57
+ # lexicon=lexicon_file,
58
+ # tokens=token_file,
59
+ # lm=lm_file,
60
+ # nbest=1,
61
+ # beam_size=500,
62
+ # beam_size_token=50,
63
+ # lm_weight=float(decoding_config["lmweight"]),
64
+ # word_score=float(decoding_config["wordscore"]),
65
+ # sil_score=float(decoding_config["silweight"]),
66
+ # blank_token="<s>",
67
+ # )
68
 
 
 
 
 
 
 
 
 
 
 
 
 
69
 
70
  def transcribe(
71
  audio_source=None, microphone=None, file_upload=None, lang="eng (English)"