-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy path.env.example
More file actions
62 lines (53 loc) · 2.22 KB
/
Copy path.env.example
File metadata and controls
62 lines (53 loc) · 2.22 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
# Optional API keys (leave empty to use local models only)
GROQ_API_KEY=
OPENAI_API_KEY=
HF_TOKEN=
# Model preferences
LLM_BACKEND=ollama # Options: ollama, openai, groq
WHISPER_MODEL=large-v3
WHISPER_BACKEND=local # Options: local, groq, openai
# Note: local uses faster-whisper, groq and openai use cloud APIs (requires API key)
# Whisper model settings
WHISPER_MODEL_SIZE=medium
WHISPER_LANGUAGE=en
INFERENCE_DEVICE= # Options: cuda, cpu. Leave empty to auto-detect (prefers CUDA when available).
PYANNOTE_DIARIZATION_MODEL=pyannote/speaker-diarization-3.1
PYANNOTE_EMBEDDING_MODEL=pyannote/embedding
# Processing settings
CHUNK_LENGTH_SECONDS=600
CHUNK_OVERLAP_SECONDS=10
AUDIO_SAMPLE_RATE=16000
SAVE_INTERMEDIATE_OUTPUTS=true # Save intermediate stage outputs (transcript, diarization, classification) to JSON
# Audio snippet export
CLEAN_STALE_CLIPS=true # Remove old snippet WAV clips before reprocessing
USE_STREAMING_SNIPPET_EXPORT=true # Use FFmpeg streaming (90% memory reduction, recommended)
# Logging
LOG_LEVEL_CONSOLE=INFO # Options: DEBUG, INFO, WARNING, ERROR, CRITICAL
LOG_LEVEL_FILE=DEBUG # Options: DEBUG, INFO, WARNING, ERROR, CRITICAL
AUDIT_LOG_ENABLED=true # Disable (false) only for local experimentation
AUDIT_LOG_ACTOR=local
AUDIT_LOG_PATH=logs/audit.log
# Classifier tuning
CLASSIFIER_CONTEXT_MAX_SEGMENTS=11
CLASSIFIER_CONTEXT_PAST_SECONDS=45
CLASSIFIER_CONTEXT_FUTURE_SECONDS=30
CLASSIFIER_AUDIT_MODE=0
CLASSIFIER_PROMPT_PREVIEW_CHARS=256
# Ollama settings (if using local LLM)
# Recommended models:
# qwen2.5:7b - RECOMMENDED: Best for Dutch, 4.7GB, 8GB VRAM (Top-ranked 2025)
# llama3.2:8b - Good balance, 4.7GB, 8GB VRAM
# llama3.2:3b - Fastest, 2GB, 4GB VRAM
# gpt-oss:20b - OpenAI open-weight, 12.8GB, 16GB+ VRAM (requires powerful GPU)
OLLAMA_MODEL=qwen2.5:7b
# Optional fallback when the primary model exceeds available memory (disabled by default).
OLLAMA_FALLBACK_MODEL=
OLLAMA_BASE_URL=http://localhost:11434
# Groq rate limit tuning (applies when LLM_BACKEND=groq)
GROQ_MAX_CALLS_PER_SECOND=2
GROQ_RATE_LIMIT_BURST=2
GROQ_RATE_LIMIT_PERIOD_SECONDS=1.0
# Interactive Clarification
INTERACTIVE_CLARIFICATION_ENABLED=false
INTERACTIVE_CLARIFICATION_TIMEOUT=30
INTERACTIVE_CLARIFICATION_MAX_QUESTIONS=10