SebKr commited on
Commit
d7feaf9
·
1 Parent(s): d9be7f8

fix: increase VAD silence to 800ms, slow TTS to 0.85x, lower threshold to 0.4

Browse files

--min_silence_ms 800 (was 500ms — still cutting mid-word)
--min_speech_ms 300 (was 250ms)
--thresh 0.4 (was 0.5 — even more sensitive)
--kokoro_speed 0.85 (was 1.0 — TTS was racing through text)

Files changed (1) hide show
  1. Dockerfile +4 -4
Dockerfile CHANGED
@@ -19,7 +19,7 @@ CMD speech-to-speech \
19
  --llm_backend responses-api \
20
  --tts kokoro \
21
  --kokoro_voice af_heart \
22
- --kokoro_speed 1.0 \
23
  --model_name "google/gemma-4-31B-it:cerebras" \
24
  --responses_api_base_url "https://router.huggingface.co/v1" \
25
  --responses_api_api_key "$HF_TOKEN" \
@@ -27,8 +27,8 @@ CMD speech-to-speech \
27
  --enable_live_transcription \
28
  --mode realtime \
29
  --num_pipelines 3 \
30
- --min_silence_ms 500 \
31
- --min_speech_ms 250 \
32
- --thresh 0.5 \
33
  --ws_host 0.0.0.0 \
34
  --ws_port 7860
 
19
  --llm_backend responses-api \
20
  --tts kokoro \
21
  --kokoro_voice af_heart \
22
+ --kokoro_speed 0.85 \
23
  --model_name "google/gemma-4-31B-it:cerebras" \
24
  --responses_api_base_url "https://router.huggingface.co/v1" \
25
  --responses_api_api_key "$HF_TOKEN" \
 
27
  --enable_live_transcription \
28
  --mode realtime \
29
  --num_pipelines 3 \
30
+ --min_silence_ms 800 \
31
+ --min_speech_ms 300 \
32
+ --thresh 0.4 \
33
  --ws_host 0.0.0.0 \
34
  --ws_port 7860