Helw150 commited on
Commit
7b1b100
·
1 Parent(s): 809dcd8

Add Duration

Browse files
Files changed (1) hide show
  1. app.py +7 -5
app.py CHANGED
@@ -24,7 +24,7 @@ diva_model = AutoModel.from_pretrained(
24
  resampler = Audio(sampling_rate=16_000)
25
 
26
 
27
- @spaces.GPU
28
  @torch.no_grad
29
  def diva_audio(audio_input, do_sample=False, temperature=0.001, prev_outs=None):
30
  sr, y = audio_input
@@ -116,8 +116,10 @@ def process_audio(audio: tuple, state: AppState):
116
  if state.stream is None:
117
  state.stream = audio[1]
118
  state.sampling_rate = audio[0]
119
- else:
120
  state.stream = np.concatenate((state.stream, audio[1]))
 
 
121
 
122
  pause_detected = determine_pause(state.stream, state.sampling_rate, state)
123
  state.pause_detected = pause_detected
@@ -190,8 +192,8 @@ with gr.Blocks(theme=theme) as demo:
190
  process_audio,
191
  [input_audio, state],
192
  [input_audio, state],
193
- stream_every=0.50,
194
- time_limit=30,
195
  )
196
  respond = input_audio.stop_recording(response, [state], [state, chatbot])
197
  respond.then(start_recording_user, [state], [input_audio])
@@ -205,4 +207,4 @@ with gr.Blocks(theme=theme) as demo:
205
  )
206
 
207
 
208
- demo.launch(share=True)
 
24
  resampler = Audio(sampling_rate=16_000)
25
 
26
 
27
+ @spaces.GPU(duration=20)
28
  @torch.no_grad
29
  def diva_audio(audio_input, do_sample=False, temperature=0.001, prev_outs=None):
30
  sr, y = audio_input
 
116
  if state.stream is None:
117
  state.stream = audio[1]
118
  state.sampling_rate = audio[0]
119
+ elif audio is not None and audio[1] is not None:
120
  state.stream = np.concatenate((state.stream, audio[1]))
121
+ else:
122
+ return None, state
123
 
124
  pause_detected = determine_pause(state.stream, state.sampling_rate, state)
125
  state.pause_detected = pause_detected
 
192
  process_audio,
193
  [input_audio, state],
194
  [input_audio, state],
195
+ stream_every=0.25,
196
+ time_limit=10,
197
  )
198
  respond = input_audio.stop_recording(response, [state], [state, chatbot])
199
  respond.then(start_recording_user, [state], [input_audio])
 
207
  )
208
 
209
 
210
+ demo.launch()