fix: adjust Whisper parameters to disable temperature fallback and prevent hallucinations on silence/noise

local_worker/vod_backfiller.py

@@ -256,6 +256,9 @@ def transcribe_vod_audio(vod_id, start_time_iso, model, start_offset=0, end_offs
                 audio_np, 
                 beam_size=5, 
                 language="ru",
+                temperature=0.0,
+                log_prob_threshold=-0.8,
+                no_speech_threshold=0.6,
                 vad_filter=True,
                 vad_parameters=dict(min_silence_duration_ms=500)
             )

local_worker/worker.py

@@ -239,6 +239,9 @@ def transcribe_audio_segment(model, pcm_bytes):
         audio_np, 
         beam_size=5, 
         language="ru",
+        temperature=0.0,
+        log_prob_threshold=-0.8,
+        no_speech_threshold=0.6,
         vad_filter=True, # Voice Activity Detection filters out silence
         vad_parameters=dict(min_silence_duration_ms=500)
     )
@@ -507,9 +510,17 @@ def run_microphone_capture(model):
             # sounddevice records float32 directly, no need for raw conversion
             audio_np = recording.flatten()
             
-            # Transcribe and send words
             def process_mic(audio=audio_np):
-                segments, info = model.transcribe(audio, beam_size=5, language="ru")
+                segments, info = model.transcribe(
+                    audio, 
+                    beam_size=5, 
+                    language="ru",
+                    temperature=0.0,
+                    log_prob_threshold=-0.8,
+                    no_speech_threshold=0.6,
+                    vad_filter=True,
+                    vad_parameters=dict(min_silence_duration_ms=500)
+                )
                 words_list = []
                 for segment in segments:
                     text = segment.text.strip()