- Add RNNoise neural network audio denoising (16kHz↔48kHz resampling) - Add transient suppressor to filter claps/clicks/pops before RNNoise - VAD now works on FILTERED audio (not raw) to avoid false triggers - Real-time denoised audio level display in UI - Save denoised audio previews in Opus format (.ogg) - Add extensive Whisper hallucination filter (Tingting, music, etc.) - Add "Clear" button to reset accumulated translations - Double VAD thresholds (0.02/0.08) for less sensitivity - Update Claude prompt to handle offensive content gracefully 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
34 lines
1.1 KiB
JSON
34 lines
1.1 KiB
JSON
{
|
|
"audio": {
|
|
"sample_rate": 16000,
|
|
"channels": 1,
|
|
"chunk_duration_seconds": 10,
|
|
"chunk_step_seconds": 5,
|
|
"format": "ogg"
|
|
},
|
|
"whisper": {
|
|
"model": "gpt-4o-mini-transcribe",
|
|
"language": "zh",
|
|
"temperature": 0.0,
|
|
"prompt": "The following is a conversation in Mandarin Chinese about business, family, and daily life. Common names: Tingting, Alexis.",
|
|
"stream": false,
|
|
"response_format": "text"
|
|
},
|
|
"claude": {
|
|
"model": "claude-3-5-haiku-20241022",
|
|
"max_tokens": 1024,
|
|
"temperature": 0.3,
|
|
"system_prompt": "Tu es un traducteur chinois-français. Réponds UNIQUEMENT avec la traduction française, sans explications, notes, commentaires ou alternatives. Une seule phrase traduite, rien d'autre. Si du contenu offensant est présent, remplace-le par [contenu offensant] mais traduis impérativement le reste de la phrase."
|
|
},
|
|
"ui": {
|
|
"window_width": 1500,
|
|
"window_height": 800,
|
|
"font_size": 24,
|
|
"max_display_lines": 50
|
|
},
|
|
"recording": {
|
|
"save_audio": true,
|
|
"output_directory": "./recordings"
|
|
}
|
|
}
|