# =============================================================================

# DTD Pipeline Configuration

# =============================================================================

# Copy this file to .env and fill in your credentials

# Required variables are marked with (REQUIRED)

# =============================================================================

# TADDY API (REQUIRED)

# =============================================================================

# Get credentials from: https://taddy.org/developers

TADDY_API_KEY=your_taddy_api_key_here

TADDY_USER_ID=your_taddy_user_id_here

# Days to look back for new episodes

LOOKBACK_DAYS=7

# =============================================================================

# HUGGINGFACE (REQUIRED)

# =============================================================================

# Get token from: https://huggingface.co/settings/tokens

# IMPORTANT: You must accept the pyannote model license first:

# https://huggingface.co/pyannote/speaker-diarization-3.1

HF_TOKEN=hf_xxxxxxxxxxxxxxxxxxxxxxxxxxxx

# =============================================================================

# SUPABASE (REQUIRED)

# =============================================================================

# Get credentials from: Supabase Dashboard → Settings → API

SUPABASE_URL=https://xxxxxxxxxxxxx.supabase.co

SUPABASE_KEY=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.xxxxx

# Storage bucket name (create in Supabase Dashboard → Storage)

SUPABASE_BUCKET=podcast-transcripts

# =============================================================================

# WHISPER CONFIGURATION

# =============================================================================

# Model size: tiny, base, small, medium, large-v2, large-v3

# Larger = better quality, more VRAM required

# large-v3: ~6GB VRAM, medium: ~4GB, small: ~2GB

WHISPER_MODEL=large-v3

# Device: cuda (GPU) or cpu

WHISPER_DEVICE=cuda

# Compute type: float16 (fastest), int8 (less VRAM), float32 (most accurate)

WHISPER_COMPUTE_TYPE=float16

# =============================================================================

# PYANNOTE CONFIGURATION

# =============================================================================

# Speaker diarization model (requires ~3GB VRAM)

PYANNOTE_MODEL=pyannote/speaker-diarization-3.1

# =============================================================================

# SPEAKER IDENTIFICATION

# =============================================================================

# Cosine similarity threshold for matching known speakers (0.0-1.0)

# Higher = stricter matching (fewer false positives, more unknowns)

# Lower = looser matching (more matches, risk of false positives)

SPEAKER_MATCH_THRESHOLD=0.70

# =============================================================================

# PATHS

# =============================================================================

# Base data directory

DATA_DIR=data

# Audio download directory (files are auto-deleted after processing)

AUDIO_DIR=data/audio

# Speaker embeddings directory

SPEAKERS_DIR=data/speakers

# Log files directory

LOGS_DIR=data/logs

# SQLite state database path

STATE_DB_PATH=data/state.db

# Podcast configuration file

PODCASTS_CONFIG_PATH=podcasts.json