-
Notifications
You must be signed in to change notification settings - Fork 4
/
engine.py
419 lines (336 loc) · 14.7 KB
/
engine.py
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
import json
import os
import tempfile
import time
import uuid
from datetime import datetime, timedelta
from enum import Enum
from typing import *
import boto3
import pvfalcon
import requests
import swagger_client
import torch
from azure.storage.blob import BlobServiceClient, ResourceTypes, AccountSasPermissions, generate_account_sas
from google.cloud import speech
from google.cloud import storage
from google.protobuf.json_format import MessageToDict
from pyannote.audio import Pipeline
from pyannote.core import Annotation, Segment
NUM_THREADS = 1
os.environ["OMP_NUM_THREADS"] = str(NUM_THREADS)
os.environ["MKL_NUM_THREADS"] = str(NUM_THREADS)
torch.set_num_threads(NUM_THREADS)
torch.set_num_interop_threads(NUM_THREADS)
class Engines(Enum):
AWS_TRANSCRIBE = "AWS_TRANSCRIBE"
AZURE_SPEECH_TO_TEXT = "AZURE_SPEECH_TO_TEXT"
GOOGLE_SPEECH_TO_TEXT = "GOOGLE_SPEECH_TO_TEXT"
GOOGLE_SPEECH_TO_TEXT_ENHANCED = "GOOGLE_SPEECH_TO_TEXT_ENHANCED"
PICOVOICE_FALCON = "PICOVOICE_FALCON"
PYANNOTE = "PYANNOTE"
class Engine:
def diarization(self, path: str) -> "Annotation":
raise NotImplementedError()
def cleanup(self) -> None:
raise NotImplementedError()
def is_offline(self) -> bool:
raise NotImplementedError()
def __str__(self) -> str:
raise NotImplementedError()
@classmethod
def create(cls, x: Engines, **kwargs: Any) -> "Engine":
try:
subclass = {
Engines.AWS_TRANSCRIBE: AWSTranscribeEngine,
Engines.AZURE_SPEECH_TO_TEXT: AzureSpeechToTextEngine,
Engines.GOOGLE_SPEECH_TO_TEXT: GoogleSpeechToTextEngine,
Engines.GOOGLE_SPEECH_TO_TEXT_ENHANCED: GoogleSpeechToTextEnhancedEngine,
Engines.PICOVOICE_FALCON: PicovoiceFalconEngine,
Engines.PYANNOTE: PyAnnoteEngine,
}[x]
except KeyError:
raise ValueError(f"cannot create `{cls.__name__}` of type `{x.value}`")
return subclass(**kwargs)
class PicovoiceFalconEngine(Engine):
def __init__(self, access_key: str) -> None:
self._falcon = pvfalcon.create(access_key=access_key)
super().__init__()
def diarization(self, path: str) -> "Annotation":
segments = self._falcon.process_file(path)
return self._segments_to_annotation(segments)
@staticmethod
def _segments_to_annotation(segments):
annotation = Annotation()
for segment in segments:
start = segment.start_sec
end = segment.end_sec
annotation[Segment(start, end)] = segment.speaker_tag
return annotation.support()
def cleanup(self) -> None:
self._falcon.delete()
def is_offline(self) -> bool:
return True
def __str__(self):
return Engines.PICOVOICE_FALCON.value
class PyAnnoteEngine(Engine):
def __init__(self, auth_token: str, use_gpu: bool = False) -> None:
if use_gpu and torch.cuda.is_available():
torch_device = torch.device("cuda")
else:
torch_device = torch.device("cpu")
self._pretrained_pipeline = Pipeline.from_pretrained(
checkpoint_path="pyannote/speaker-diarization-3.1",
use_auth_token=auth_token,
)
self._pretrained_pipeline.to(torch_device)
super().__init__()
def diarization(self, path: str) -> "Annotation":
return self._pretrained_pipeline(path)
def cleanup(self) -> None:
self._pretrained_pipeline = None
def is_offline(self) -> bool:
return True
def __str__(self) -> str:
return Engines.PYANNOTE.value
class AWSTranscribeEngine(Engine):
def __init__(self, bucket_name: str) -> None:
self._bucket_name = bucket_name
self._storage = boto3.client("s3")
self._transcribe = boto3.client("transcribe")
super().__init__()
def diarization(self, path: str) -> "Annotation":
blob_name = os.path.basename(path)
temp_path = os.path.join(tempfile.mkdtemp(), blob_name)
self._storage.upload_file(Filename=path, Bucket=self._bucket_name, Key=blob_name)
self._transcribe_blob(blob_name=blob_name, results_path=temp_path)
with open(temp_path) as f:
transcript = json.load(f)["results"]
print(transcript)
return self._transcript_to_annotation(transcript)
@staticmethod
def _transcript_to_annotation(transcript: Dict) -> "Annotation":
segments = transcript["speaker_labels"]["segments"]
annotation = Annotation()
for segment in segments:
start = float(segment["start_time"])
end = float(segment["end_time"])
annotation[Segment(start, end)] = segment["speaker_label"]
return annotation.support()
def _transcribe_blob(self, blob_name: str, results_path: str) -> None:
completed = False
job_name = uuid.uuid4().hex
response = self._transcribe.start_transcription_job(
TranscriptionJobName=job_name,
LanguageCode="en-US",
OutputBucketName=self._bucket_name,
OutputKey=blob_name + ".json",
Media={"MediaFileUri": "s3://" + self._bucket_name + "/" + blob_name},
MediaFormat="wav",
Settings={
"ShowSpeakerLabels": True,
"MaxSpeakerLabels": 9,
},
)
if response["TranscriptionJob"]["TranscriptionJobStatus"] != "IN_PROGRESS":
completed = True
while not completed:
time.sleep(2)
response = self._transcribe.get_transcription_job(
TranscriptionJobName=job_name,
)
if response["TranscriptionJob"]["TranscriptionJobStatus"] != "IN_PROGRESS":
completed = True
self._storage.download_file(Filename=results_path, Bucket=self._bucket_name, Key=blob_name + ".json")
def is_offline(self) -> bool:
return False
def cleanup(self) -> None:
pass
def __str__(self):
return Engines.AWS_TRANSCRIBE.value
class GoogleSpeechToTextEngine(Engine):
_diarization_config = speech.SpeakerDiarizationConfig(
enable_speaker_diarization=True,
min_speaker_count=1,
max_speaker_count=20,
)
_config = speech.RecognitionConfig(
encoding=speech.RecognitionConfig.AudioEncoding.LINEAR16,
sample_rate_hertz=16000,
language_code="en-US",
enable_word_time_offsets=True,
diarization_config=_diarization_config,
)
def __init__(self, bucket_name: str) -> None:
self._speech_client = speech.SpeechClient()
self._storage_client = storage.Client()
self._bucket_name = bucket_name
self._bucket = self._storage_client.bucket(bucket_name)
def diarization(self, path: str) -> "Annotation":
blob_name = os.path.basename(path)
self._upload_audio_to_storage(path, blob_name)
response = self._transcribe_from_storage(path)
transcript = response["results"]
return self._transcript_to_annotation(transcript)
@staticmethod
def _transcript_to_annotation(transcript: List[Dict]) -> "Annotation":
words = transcript[-1]["alternatives"][0]["words"]
annotation = Annotation()
for word in words:
start = float(word["startTime"][:-1])
end = float(word["endTime"][:-1])
annotation[Segment(start, end)] = word["speakerTag"]
return annotation.support()
def _transcribe_from_storage(self, path: str) -> Dict:
audio = speech.RecognitionAudio(uri=f"gs://{self._bucket_name}/{os.path.basename(path)}")
operation = self._speech_client.long_running_recognize(config=self._config, audio=audio)
response = operation.result(timeout=600)
response_dict = MessageToDict(response._pb)
return response_dict
def _upload_audio_to_storage(self, source_file_name: str, blob_name: str) -> None:
blob = self._bucket.blob(blob_name)
stats = storage.Blob(bucket=self._bucket, name=blob_name).exists(self._storage_client)
if not stats:
blob.upload_from_filename(source_file_name)
def is_offline(self) -> bool:
return False
def cleanup(self) -> None:
pass
def __str__(self):
return Engines.GOOGLE_SPEECH_TO_TEXT.value
class GoogleSpeechToTextEnhancedEngine(GoogleSpeechToTextEngine):
_diarization_config = speech.SpeakerDiarizationConfig(
enable_speaker_diarization=True,
min_speaker_count=1,
max_speaker_count=20,
)
_config = speech.RecognitionConfig(
encoding=speech.RecognitionConfig.AudioEncoding.LINEAR16,
sample_rate_hertz=16000,
language_code="en-US",
enable_word_time_offsets=True,
diarization_config=_diarization_config,
model="latest_long",
use_enhanced=True,
)
def __str__(self):
return Engines.GOOGLE_SPEECH_TO_TEXT_ENHANCED.value
class AzureSpeechToTextEngine(Engine):
def __init__(
self,
storage_account_key: str,
storage_account_name: str,
storage_container_name: str,
subscription_key: str,
region: str,
) -> None:
self._storage_account_key = storage_account_key
self._storage_account_name = storage_account_name
self._connection_string = \
(f"DefaultEndpointsProtocol=https;"
f"AccountName={storage_account_name};"
f"AccountKey={storage_account_key};"
f"EndpointSuffix=core.windows.net")
self._container_name = storage_container_name
self._subscription_key = subscription_key
self._service_region = region
def _upload_to_blob_storage(self, file_path: str, blob_name: str) -> None:
blob_service_client = BlobServiceClient.from_connection_string(self._connection_string)
blob_client = blob_service_client.get_blob_client(container=self._container_name, blob=blob_name)
if blob_client.exists():
return
with open(file_path, "rb") as data:
blob_client.upload_blob(data)
@staticmethod
def _transcribe_from_single_blob(uri: str, properties: swagger_client.TranscriptionProperties):
transcription_definition = swagger_client.Transcription(
display_name="diariazation",
description="no-description",
locale="en-US",
content_urls=[uri],
properties=properties,
)
return transcription_definition
@staticmethod
def _paginate(api, paginated_object):
yield from paginated_object.values
typename = type(paginated_object).__name__
auth_settings = ["api_key"]
while paginated_object.next_link:
link = paginated_object.next_link[len(api.api_client.configuration.host):]
paginated_object, status, headers = api.api_client.call_api(
link, "GET", response_type=typename, auth_settings=auth_settings
)
if status == 200:
yield from paginated_object.values
else:
raise Exception(f"could not receive paginated data: status {status}")
def diarization(self, path: str) -> "Annotation":
blob_name = os.path.basename(path)
self._upload_to_blob_storage(path, blob_name)
transcripts = self._transcribe(blob_name)
return self._transcript_to_annotation(transcripts)
def _transcribe(self, blob_name: str) -> Dict:
transcripts = {"result": []}
sas_token = generate_account_sas(
account_name=self._storage_account_name,
account_key=self._storage_account_key,
resource_types=ResourceTypes(service=True, container=True, object=True),
permission=AccountSasPermissions(read=True, write=True, list=True, delete=True),
expiry=datetime.utcnow() + timedelta(hours=1),
)
blob_service_client = BlobServiceClient.from_connection_string(self._connection_string)
blob_client = blob_service_client.get_blob_client(container=self._container_name, blob=blob_name)
blob_url = f"{blob_client.url}?{sas_token}"
configuration = swagger_client.Configuration()
configuration.api_key["Ocp-Apim-Subscription-Key"] = self._subscription_key
configuration.host = f"https://{self._service_region}.api.cognitive.microsoft.com/speechtotext/v3.1"
client = swagger_client.ApiClient(configuration)
api = swagger_client.CustomSpeechTranscriptionsApi(api_client=client)
properties = swagger_client.TranscriptionProperties()
properties.word_level_timestamps_enabled = True
properties.display_form_word_level_timestamps_enabled = True
properties.diarization_enabled = True
properties.diarization = swagger_client.DiarizationProperties(
swagger_client.DiarizationSpeakersProperties(min_count=1, max_count=20)
)
transcription_definition = self._transcribe_from_single_blob(blob_url, properties)
(
created_transcription,
status,
headers,
) = api.transcriptions_create_with_http_info(transcription=transcription_definition)
transcription_id = headers["location"].split("/")[-1]
completed = False
while not completed:
time.sleep(5)
transcription = api.transcriptions_get(transcription_id)
if transcription.status in ("Failed", "Succeeded"):
completed = True
if transcription.status == "Succeeded":
pag_files = api.transcriptions_list_files(transcription_id)
for file_data in self._paginate(api, pag_files):
if file_data.kind != "Transcription":
continue
results_url = file_data.links.content_url
results = requests.get(results_url)
transcripts["result"].append(json.loads(results.content.decode("utf-8")))
elif transcription.status == "Failed":
raise Exception(f"Transcription failed: {transcription.properties.error.message}")
return transcripts
@staticmethod
def _transcript_to_annotation(transcript: Dict) -> "Annotation":
annotation = Annotation()
pages = transcript["result"]
for page in pages:
for segment in page["recognizedPhrases"]:
start = float(segment["offsetInTicks"] / 10000000)
end = start + float(segment["durationInTicks"] / 10000000)
annotation[Segment(start, end)] = segment["speaker"]
return annotation.support()
def cleanup(self) -> None:
pass
def is_offline(self) -> bool:
return False
def __str__(self):
return Engines.AZURE_SPEECH_TO_TEXT.value