# Chatterbox TTS API Configuration
# Copy this file to .env and modify the values as needed

# =============================================================================
# Server Configuration
# =============================================================================

# Port to run the API server on
PORT=4123

# Port for frontend/proxy when using fullstack mode (when API runs behind proxy)
# Only used when running with --profile frontend
FRONTEND_PORT=4321

# Host to bind the server to (0.0.0.0 for all interfaces)
HOST=0.0.0.0

# CORS origins (comma-separated list)
CORS_ORIGINS=*

# =============================================================================
# Voice and Model Configuration
# =============================================================================

# Device to use for inference (auto/cuda/mps/cpu)
# auto = automatically detect best available device
DEVICE=auto

# Path to the voice sample file for cloning
VOICE_SAMPLE_PATH=./voice-sample.mp3

# Alternative: Use a voice sample from a mounted directory
# VOICE_SAMPLE_PATH=/app/voice-samples/my-voice.mp3

# Directory to cache downloaded models (for persistence)
# Use ./models for local development, /cache for Docker
MODEL_CACHE_DIR=./models

# Directory to store uploaded voice library (for persistence)
# Use ./voices for local development, /voices for Docker
VOICE_LIBRARY_DIR=./voices

# =============================================================================
# TTS Model Settings
# =============================================================================

# Enable multilingual TTS model (true/false)
# true = Support 23 languages with multilingual model (default)
# false = English-only with standard model
USE_MULTILINGUAL_MODEL=true

# Emotion intensity/exaggeration level (0.25 - 2.0)
# 0.5 = neutral, higher values = more expressive
EXAGGERATION=0.5

# CFG weight for pace control (0.0 - 1.0)
# Lower values = faster speech, higher values = slower/more deliberate
CFG_WEIGHT=0.5

# Sampling temperature (0.05 - 5.0)
# Lower values = more deterministic, higher values = more random/creative
TEMPERATURE=0.8

# =============================================================================
# Text Processing
# =============================================================================

# Maximum characters per text chunk (recommended: 200-300)
MAX_CHUNK_LENGTH=280

# Maximum total characters for entire input (hard limit)
MAX_TOTAL_LENGTH=3000

# =============================================================================
# Docker-specific Configuration
# =============================================================================

# Host path to voice sample file (for Docker volume mounting)
VOICE_SAMPLE_HOST_PATH=./voice-sample.mp3

# Host directory containing voice samples (optional)
# VOICE_SAMPLES_DIR=./voice-samples

# =============================================================================
# Advanced Settings (usually don't need to change)
# =============================================================================

# Memory Management
# Cleanup memory every N requests (default: 5)
MEMORY_CLEANUP_INTERVAL=5

# Clear CUDA cache every N requests (default: 3)
CUDA_CACHE_CLEAR_INTERVAL=3

# Enable detailed memory monitoring and logging (true/false)
ENABLE_MEMORY_MONITORING=true

# HuggingFace cache directory (optional)
# HF_HOME=/cache/huggingface

# Disable HuggingFace telemetry (true/false)
# HF_HUB_DISABLE_TELEMETRY=true

# PyTorch cache directory (optional)
# TORCH_HOME=/cache/torch

# =============================================================================
# Examples for different use cases:
# =============================================================================

# For more expressive/dramatic speech:
# EXAGGERATION=0.8
# CFG_WEIGHT=0.3
# TEMPERATURE=1.0

# For neutral/professional speech:
# EXAGGERATION=0.4
# CFG_WEIGHT=0.6
# TEMPERATURE=0.6

# For faster processing (less stable):
# MAX_CHUNK_LENGTH=400
# TEMPERATURE=0.5

# For slower, more careful speech:
# CFG_WEIGHT=0.8
# TEMPERATURE=0.4