Spaces:
Paused
Paused
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)
- 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
|
| 23 |
--model_name "google/gemma-4-31B-it:cerebras" \
|
| 24 |
--responses_api_base_url "https://router.huggingface.co/v1" \
|
| 25 |
--responses_api_api_key "$HF_TOKEN" \
|
|
@@ -27,8 +27,8 @@ CMD speech-to-speech \
|
|
| 27 |
--enable_live_transcription \
|
| 28 |
--mode realtime \
|
| 29 |
--num_pipelines 3 \
|
| 30 |
-
--min_silence_ms
|
| 31 |
-
--min_speech_ms
|
| 32 |
-
--thresh 0.
|
| 33 |
--ws_host 0.0.0.0 \
|
| 34 |
--ws_port 7860
|
|
|
|
| 19 |
--llm_backend responses-api \
|
| 20 |
--tts kokoro \
|
| 21 |
--kokoro_voice af_heart \
|
| 22 |
+
--kokoro_speed 0.85 \
|
| 23 |
--model_name "google/gemma-4-31B-it:cerebras" \
|
| 24 |
--responses_api_base_url "https://router.huggingface.co/v1" \
|
| 25 |
--responses_api_api_key "$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
|