.env.example

# Server Configuration
HOST=0.0.0.0
PORT=8000
WORKERS=1
DEBUG=false

# Security Configuration
AUTH_ENABLED=false
API_KEY=

# Model Configuration
DEFAULT_MODEL=large-v3
DETECTOR_MODEL=small
DEVICE=cuda
BATCH_SIZE=8

# Language Detection
MIN_PROB=0.6
# DETECTOR_BATCH_SIZE overrides BATCH_SIZE for language detection when set
DETECTOR_BATCH_SIZE=4
DETECTOR_COMPUTE_TYPE=int8

# Language Detection Weighting
CONFIDENCE_WEIGHT=2.0

# Enhanced Language Detection
NUM_LANGUAGE_CHUNKS=6
MIN_LANGUAGE_CONFIDENCE=0.3
DYNAMIC_THRESHOLD_ENABLED=true
MIN_CONSENSUS_RATIO=0.5

# Audio Quality Filtering
ENABLE_QUALITY_FILTERING=true
MIN_CHUNK_RMS=100
MIN_CHUNK_DURATION_MS=1000
MIN_CHUNK_AMPLITUDE=1000
QUALITY_FALLBACK_THRESHOLD=0.5

# File Upload
MAX_FILE_SIZE=209715200

# Audio Processing
SILENCE_MIN_DURATION=300
SILENCE_THRESHOLD=-35
CHUNK_DURATION=10000
CHUNK_OFFSET=5000

# ASR Options
ASR_BEAM_SIZE=2
ASR_CONDITION_ON_PREVIOUS_TEXT=false
ASR_TEMPERATURES=[0.0]
ASR_NO_SPEECH_THRESHOLD=0.6

# VAD Options
VAD_MIN_SILENCE_DURATION_MS=500
VAD_SPEECH_PAD_MS=200

# Suppression Tokens
# Comma-separated list of phrases to suppress
SUPPRESS_PHRASES="дима торжок,dima torzok,dima torzhok,субтитры подогнал"

# Speaker Diarization
# Enable speaker diarization by default for all requests
ENABLE_DIARIZATION=false
# HuggingFace token for accessing pyannote models (required for diarization)
# Get token at: https://huggingface.co/settings/tokens
# Accept agreements: https://huggingface.co/pyannote/speaker-diarization-3.1
HF_TOKEN=
# Diarization model to use
DIARIZATION_MODEL=pyannote/speaker-diarization-3.1
# Default speaker count constraints
DEFAULT_MIN_SPEAKERS=1
DEFAULT_MAX_SPEAKERS=10

# Logging
LOG_LEVEL=INFO
LOG_FORMAT="%(asctime)s - %(name)s - %(levelname)s - %(message)s"