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
|
from __future__ import absolute_import
import logging
from requests import Session
from subliminal_patch.subtitle import Subtitle
from subliminal_patch.providers import Provider
from subliminal import __short_version__
from subliminal.exceptions import ConfigurationError
from subzero.language import Language
from subliminal.video import Episode, Movie
from babelfish.exceptions import LanguageReverseError
import ffmpeg
import functools
# These are all the languages Whisper supports.
# from whisper.tokenizer import LANGUAGES
whisper_languages = {
"en": "english",
"zh": "chinese",
"de": "german",
"es": "spanish",
"ru": "russian",
"ko": "korean",
"fr": "french",
"ja": "japanese",
"pt": "portuguese",
"tr": "turkish",
"pl": "polish",
"ca": "catalan",
"nl": "dutch",
"ar": "arabic",
"sv": "swedish",
"it": "italian",
"id": "indonesian",
"hi": "hindi",
"fi": "finnish",
"vi": "vietnamese",
"he": "hebrew",
"uk": "ukrainian",
"el": "greek",
"ms": "malay",
"cs": "czech",
"ro": "romanian",
"da": "danish",
"hu": "hungarian",
"ta": "tamil",
"no": "norwegian",
"th": "thai",
"ur": "urdu",
"hr": "croatian",
"bg": "bulgarian",
"lt": "lithuanian",
"la": "latin",
"mi": "maori",
"ml": "malayalam",
"cy": "welsh",
"sk": "slovak",
"te": "telugu",
"fa": "persian",
"lv": "latvian",
"bn": "bengali",
"sr": "serbian",
"az": "azerbaijani",
"sl": "slovenian",
"kn": "kannada",
"et": "estonian",
"mk": "macedonian",
"br": "breton",
"eu": "basque",
"is": "icelandic",
"hy": "armenian",
"ne": "nepali",
"mn": "mongolian",
"bs": "bosnian",
"kk": "kazakh",
"sq": "albanian",
"sw": "swahili",
"gl": "galician",
"mr": "marathi",
"pa": "punjabi",
"si": "sinhala",
"km": "khmer",
"sn": "shona",
"yo": "yoruba",
"so": "somali",
"af": "afrikaans",
"oc": "occitan",
"ka": "georgian",
"be": "belarusian",
"tg": "tajik",
"sd": "sindhi",
"gu": "gujarati",
"am": "amharic",
"yi": "yiddish",
"lo": "lao",
"uz": "uzbek",
"fo": "faroese",
"ht": "haitian creole",
"ps": "pashto",
"tk": "turkmen",
"nn": "nynorsk",
"mt": "maltese",
"sa": "sanskrit",
"lb": "luxembourgish",
"my": "myanmar",
"bo": "tibetan",
"tl": "tagalog",
"mg": "malagasy",
"as": "assamese",
"tt": "tatar",
"haw": "hawaiian",
"ln": "lingala",
"ha": "hausa",
"ba": "bashkir",
"jw": "javanese",
"su": "sundanese",
}
logger = logging.getLogger(__name__)
@functools.lru_cache(2)
def encode_audio_stream(path, audio_stream_language=None):
logger.debug("Encoding audio stream to WAV with ffmpeg")
try:
# This launches a subprocess to decode audio while down-mixing and resampling as necessary.
inp = ffmpeg.input(path, threads=0)
if audio_stream_language:
logger.debug(f"Whisper will only use the {audio_stream_language} audio stream for {path}")
inp = inp[f'a:m:language:{audio_stream_language}']
out, _ = inp.output("-", format="s16le", acodec="pcm_s16le", ac=1, ar=16000) \
.run(cmd=["ffmpeg", "-nostdin"], capture_stdout=True, capture_stderr=True)
except ffmpeg.Error as e:
raise RuntimeError(f"Failed to load audio: {e.stderr.decode()}") from e
logger.debug(f"Finished encoding audio stream in {path} with no errors")
return out
def whisper_get_language(code, name):
# Whisper uses an inconsistent mix of alpha2 and alpha3 language codes
try:
return Language.fromalpha2(code)
except LanguageReverseError:
return Language.fromname(name)
def whisper_get_language_reverse(alpha3):
# Returns the whisper language code given an alpha3b language
for wl in whisper_languages:
lan = whisper_get_language(wl, whisper_languages[wl])
if lan.alpha3 == alpha3:
return wl
raise ValueError
class WhisperAISubtitle(Subtitle):
'''Whisper AI Subtitle.'''
provider_name = 'whisperai'
hash_verifiable = False
def __init__(self, language, video):
super(WhisperAISubtitle, self).__init__(language)
self.video = video
self.task = None
self.audio_language = None
self.force_audio_stream = None
@property
def id(self):
return self.video.original_name
def get_matches(self, video):
matches = set()
if isinstance(video, Episode):
matches.update(["series", "season", "episode"])
elif isinstance(video, Movie):
matches.update(["title"])
return matches
class WhisperAIProvider(Provider):
'''Whisper AI Provider.'''
languages = set()
for lan in whisper_languages:
languages.update({whisper_get_language(lan, whisper_languages[lan])})
languages.update(set(Language.rebuild(lang, hi=True) for lang in languages))
languages.update(set(Language.rebuild(lang, forced=True) for lang in languages))
video_types = (Episode, Movie)
def __init__(self, endpoint=None, timeout=None):
if not endpoint:
raise ConfigurationError('Whisper Web Service Endpoint must be provided')
if not timeout:
raise ConfigurationError('Whisper Web Service Timeout must be provided')
self.endpoint = endpoint
self.timeout = int(timeout)
self.session = None
def initialize(self):
self.session = Session()
self.session.headers['User-Agent'] = 'Subliminal/%s' % __short_version__
def terminate(self):
self.session.close()
@functools.lru_cache(2048)
def detect_language(self, path) -> Language:
out = encode_audio_stream(path)
r = self.session.post(f"{self.endpoint}/detect-language",
params={'encode': 'false'},
files={'audio_file': out},
timeout=self.timeout)
logger.info(f"Whisper detected language of {path} as {r.json()['detected_language']}")
return whisper_get_language(r.json()["language_code"], r.json()["detected_language"])
def query(self, language, video):
if language not in self.languages:
return None
sub = WhisperAISubtitle(language, video)
sub.task = "transcribe"
if video.audio_languages:
if language.alpha3 in video.audio_languages:
sub.audio_language = language.alpha3
if len(list(video.audio_languages)) > 1:
sub.force_audio_stream = language.alpha3
else:
sub.task = "translate"
sub.audio_language = list(video.audio_languages)[0]
else:
# We must detect the language manually
detected_lang = self.detect_language(video.original_path)
if detected_lang != language:
sub.task = "translate"
sub.audio_language = detected_lang.alpha3
if sub.task == "translate":
if language.alpha3 != "eng":
logger.info(f"Translation only possible from {language} to English")
return None
logger.debug(f"Whisper ({video.original_path}): {sub.audio_language} -> {language.alpha3} [TASK: {sub.task}]")
return sub
def list_subtitles(self, video, languages):
subtitles = [self.query(l, video) for l in languages]
return [s for s in subtitles if s is not None]
def download_subtitle(self, subtitle: WhisperAISubtitle):
# Invoke Whisper through the API. This may take a long time depending on the file.
# TODO: This loads the entire file into memory, find a good way to stream the file in chunks
out = encode_audio_stream(subtitle.video.original_path, subtitle.force_audio_stream)
r = self.session.post(f"{self.endpoint}/asr",
params={'task': subtitle.task, 'language': whisper_get_language_reverse(subtitle.audio_language), 'output': 'srt', 'encode': 'false'},
files={'audio_file': out},
timeout=self.timeout)
subtitle.content = r.content
|
