aadnk commited on
Commit
60d420f
1 Parent(s): a1b1422

Make a separate process timeout for diarization

Browse files
app.py CHANGED
@@ -33,7 +33,7 @@ import ffmpeg
33
  import gradio as gr
34
 
35
  from src.download import ExceededMaximumDuration, download_url
36
- from src.utils import optional_int, slugify, write_srt, write_vtt
37
  from src.vad import AbstractTranscription, NonSpeechStrategy, PeriodicTranscriptionConfig, TranscriptionConfig, VadPeriodicTranscription, VadSileroTranscription
38
  from src.whisper.abstractWhisperContainer import AbstractWhisperContainer
39
  from src.whisper.whisperFactory import create_whisper_container
@@ -95,7 +95,8 @@ class WhisperTranscriber:
95
  def set_diarization(self, auth_token: str, enable_daemon_process: bool = True, **kwargs):
96
  if self.diarization is None:
97
  self.diarization = DiarizationContainer(auth_token=auth_token, enable_daemon_process=enable_daemon_process,
98
- auto_cleanup_timeout_seconds=self.vad_process_timeout, cache=self.model_cache)
 
99
  # Set parameters
100
  self.diarization_kwargs = kwargs
101
 
@@ -688,6 +689,15 @@ if __name__ == '__main__':
688
  help="the compute type to use for inference")
689
  parser.add_argument("--threads", type=optional_int, default=0,
690
  help="number of threads used by torch for CPU inference; supercedes MKL_NUM_THREADS/OMP_NUM_THREADS")
 
 
 
 
 
 
 
 
 
691
 
692
  args = parser.parse_args().__dict__
693
 
 
33
  import gradio as gr
34
 
35
  from src.download import ExceededMaximumDuration, download_url
36
+ from src.utils import optional_int, slugify, str2bool, write_srt, write_vtt
37
  from src.vad import AbstractTranscription, NonSpeechStrategy, PeriodicTranscriptionConfig, TranscriptionConfig, VadPeriodicTranscription, VadSileroTranscription
38
  from src.whisper.abstractWhisperContainer import AbstractWhisperContainer
39
  from src.whisper.whisperFactory import create_whisper_container
 
95
  def set_diarization(self, auth_token: str, enable_daemon_process: bool = True, **kwargs):
96
  if self.diarization is None:
97
  self.diarization = DiarizationContainer(auth_token=auth_token, enable_daemon_process=enable_daemon_process,
98
+ auto_cleanup_timeout_seconds=self.app_config.diarization_process_timeout,
99
+ cache=self.model_cache)
100
  # Set parameters
101
  self.diarization_kwargs = kwargs
102
 
 
689
  help="the compute type to use for inference")
690
  parser.add_argument("--threads", type=optional_int, default=0,
691
  help="number of threads used by torch for CPU inference; supercedes MKL_NUM_THREADS/OMP_NUM_THREADS")
692
+
693
+ parser.add_argument('--auth_token', type=str, default=default_app_config.auth_token, help='HuggingFace API Token (optional)')
694
+ parser.add_argument("--diarization", type=str2bool, default=default_app_config.diarization, \
695
+ help="whether to perform speaker diarization")
696
+ parser.add_argument("--diarization_num_speakers", type=int, default=default_app_config.diarization_speakers, help="Number of speakers")
697
+ parser.add_argument("--diarization_min_speakers", type=int, default=default_app_config.diarization_min_speakers, help="Minimum number of speakers")
698
+ parser.add_argument("--diarization_max_speakers", type=int, default=default_app_config.diarization_max_speakers, help="Maximum number of speakers")
699
+ parser.add_argument("--diarization_process_timeout", type=int, default=default_app_config.diarization_process_timeout, \
700
+ help="Number of seconds before inactivate diarization processes are terminated. Use 0 to close processes immediately, or None for no timeout.")
701
 
702
  args = parser.parse_args().__dict__
703
 
config.json5 CHANGED
@@ -150,5 +150,7 @@
150
  // The minimum number of speakers to detect
151
  "diarization_min_speakers": 1,
152
  // The maximum number of speakers to detect
153
- "diarization_max_speakers": 5,
 
 
154
  }
 
150
  // The minimum number of speakers to detect
151
  "diarization_min_speakers": 1,
152
  // The maximum number of speakers to detect
153
+ "diarization_max_speakers": 8,
154
+ // The number of seconds before inactivate processes are terminated. Use 0 to close processes immediately, or None for no timeout.
155
+ "diarization_process_timeout": 60,
156
  }
src/config.py CHANGED
@@ -72,7 +72,8 @@ class ApplicationConfig:
72
  highlight_words: bool = False,
73
  # Diarization
74
  auth_token: str = None, diarization: bool = False, diarization_speakers: int = 2,
75
- diarization_min_speakers: int = 1, diarization_max_speakers: int = 5):
 
76
 
77
  self.models = models
78
 
@@ -130,6 +131,7 @@ class ApplicationConfig:
130
  self.diarization_speakers = diarization_speakers
131
  self.diarization_min_speakers = diarization_min_speakers
132
  self.diarization_max_speakers = diarization_max_speakers
 
133
 
134
  def get_model_names(self):
135
  return [ x.name for x in self.models ]
 
72
  highlight_words: bool = False,
73
  # Diarization
74
  auth_token: str = None, diarization: bool = False, diarization_speakers: int = 2,
75
+ diarization_min_speakers: int = 1, diarization_max_speakers: int = 5,
76
+ diarization_process_timeout: int = 60):
77
 
78
  self.models = models
79
 
 
131
  self.diarization_speakers = diarization_speakers
132
  self.diarization_min_speakers = diarization_min_speakers
133
  self.diarization_max_speakers = diarization_max_speakers
134
+ self.diarization_process_timeout = diarization_process_timeout
135
 
136
  def get_model_names(self):
137
  return [ x.name for x in self.models ]
src/diarization/diarizationContainer.py CHANGED
@@ -16,7 +16,8 @@ class DiarizationContainer:
16
  # Create parallel context if needed
17
  if self.diarization_context is None and self.enable_daemon_process:
18
  # Number of processes is set to 1 as we mainly use this in order to clean up GPU memory
19
- self.diarization_context = ParallelContext(num_processes=1)
 
20
 
21
  # Run directly
22
  if self.diarization_context is None:
 
16
  # Create parallel context if needed
17
  if self.diarization_context is None and self.enable_daemon_process:
18
  # Number of processes is set to 1 as we mainly use this in order to clean up GPU memory
19
+ self.diarization_context = ParallelContext(num_processes=1, auto_cleanup_timeout_seconds=self.auto_cleanup_timeout_seconds)
20
+ print("Created diarization context with auto cleanup timeout of %d seconds" % self.auto_cleanup_timeout_seconds)
21
 
22
  # Run directly
23
  if self.diarization_context is None: