CrazyMonkey0 commited on
Commit
c56327c
·
1 Parent(s): 431435a

fix(tts): improve performance of send_audio function by avoiding repeated np.concatenate

Browse files
Files changed (2) hide show
  1. app/main.py +1 -1
  2. app/routes/tts.py +4 -3
app/main.py CHANGED
@@ -14,7 +14,7 @@ async def startup_event():
14
  try:
15
  app.state.model_nlp, app.state.tokenizer_nlp = load_model_nlp()
16
  app.state.model_trans, app.state.tokenizer_trans = load_model_translation()
17
- app.state.model_tts= load_model_tts()
18
  app.state.model_asr = load_model_asr()
19
  except Exception as e:
20
  print("[ERROR] Failed to load models:", e)
 
14
  try:
15
  app.state.model_nlp, app.state.tokenizer_nlp = load_model_nlp()
16
  app.state.model_trans, app.state.tokenizer_trans = load_model_translation()
17
+ app.state.model_tts = load_model_tts()
18
  app.state.model_asr = load_model_asr()
19
  except Exception as e:
20
  print("[ERROR] Failed to load models:", e)
app/routes/tts.py CHANGED
@@ -19,13 +19,14 @@ def load_model_tts():
19
  def send_audio(request: Request, text: str, voice: str = "af_heart") -> str:
20
  pipeline = request.app.state.model_tts
21
 
22
- audio_total = np.array([], dtype=np.float32)
23
 
24
  generator = pipeline(text, voice=voice)
25
  for _, _, audio in generator:
26
- audio_total = np.concatenate([audio_total, audio])
 
 
27
 
28
- # zapis WAV do RAM (tak jak wcześniej)
29
  buffer = io.BytesIO()
30
  sf.write(buffer, audio_total, samplerate=24000, format="WAV")
31
  buffer.seek(0)
 
19
  def send_audio(request: Request, text: str, voice: str = "af_heart") -> str:
20
  pipeline = request.app.state.model_tts
21
 
22
+ audio_chunks = []
23
 
24
  generator = pipeline(text, voice=voice)
25
  for _, _, audio in generator:
26
+ audio_chunks.append(audio)
27
+
28
+ audio_total = np.concatenate(audio_chunks)
29
 
 
30
  buffer = io.BytesIO()
31
  sf.write(buffer, audio_total, samplerate=24000, format="WAV")
32
  buffer.seek(0)