# Copy this file to scripts/quickstart/env and edit it for your machine.
#
# The scripts infer DIGITAL_HUMAN_HOME from the repository location when this is
# unset. If your layout is:
#
#   /path/to/digital_human/opentalking
#   /path/to/digital_human/models
#   /path/to/digital_human/model-repos/omnirt
#   /path/to/digital_human/runtimes
#
# then DIGITAL_HUMAN_HOME should be /path/to/digital_human.

# Change this to your deployment root
# DIGITAL_HUMAN_HOME=/path/to/digital_human
# OPENTALKING_HOME=$DIGITAL_HUMAN_HOME/opentalking
# OPENTALKING_MODEL_ROOT=$DIGITAL_HUMAN_HOME/models
# OPENTALKING_MODEL_REPO_ROOT=$DIGITAL_HUMAN_HOME/model-repos
# OPENTALKING_RUNTIME_ROOT=$DIGITAL_HUMAN_HOME/runtimes
# OMNIRT_MODEL_ROOT=$OPENTALKING_MODEL_ROOT

# Optional PyPI mirror for uv. The QuickTalk CUDA wheels for torch and
# torchvision are still selected by OmniRT's pyproject source pin.
# UV_DEFAULT_INDEX=https://pypi.tuna.tsinghua.edu.cn/simple
# UV_INDEX_URL=https://mirrors.aliyun.com/pypi/simple

# OpenTalking API and frontend ports.
# OPENTALKING_API_PORT is mapped to OPENTALKING_UNIFIED_PORT for the backend
# and VITE_BACKEND_PORT for the frontend dev proxy.
# OPENTALKING_API_PORT=8000
# OPENTALKING_API_HOST=0.0.0.0
# OPENTALKING_WEB_PORT=5173
# OPENTALKING_WEB_HOST=0.0.0.0

# Character memory / Mem0. Sessions/personas must also turn on memory_enabled.
# This adapter uses mem0.Memory, so MEM0_API_KEY is not required.
# OPENTALKING_MEMORY_SUMMARY_ENABLED=true
# OPENTALKING_MEMORY_SUMMARY_TURN_WINDOW=8
# OPENTALKING_MEMORY_SUMMARY_MAX_ITEMS=3
# OPENTALKING_MEMORY_MEM0_LLM_PROVIDER=openai
# OPENTALKING_MEMORY_MEM0_LLM_BASE_URL=https://dashscope.aliyuncs.com/compatible-mode/v1
# OPENTALKING_MEMORY_MEM0_LLM_API_KEY=<llm-api-key>
# OPENTALKING_MEMORY_MEM0_LLM_MODEL=qwen-flash
# OPENTALKING_MEMORY_MEM0_EMBEDDER_PROVIDER=openai
# OPENTALKING_MEMORY_MEM0_EMBEDDER_BASE_URL=https://dashscope.aliyuncs.com/compatible-mode/v1
# OPENTALKING_MEMORY_MEM0_EMBEDDER_API_KEY=<embedding-api-key>
# OPENTALKING_MEMORY_MEM0_EMBEDDER_MODEL=text-embedding-v4

# Real model serving endpoint. Leave empty for mock-only startup.
# OMNIRT_ENDPOINT=http://127.0.0.1:9000
# OMNIRT_AUDIO2VIDEO_PATH_TEMPLATE=/v1/audio2video/{model}

# Local CosyVoice3 sidecar. Keep TensorRT off until the CosyVoice runtime has
# built/loaded compatible TRT engines for this GPU and model directory.
# Prepare the sidecar venv with OPENTALKING_COSYVOICE_INSTALL_TENSORRT=1 before
# setting OPENTALKING_TTS_LOCAL_COSYVOICE_LOAD_TRT=1. The sidecar starter adds
# the CosyVoice venv's site-packages/tensorrt_libs directory to LD_LIBRARY_PATH.
# For CosyVoice3 fp16 TRT, place the official flow.decoder.estimator.autocast_fp16.onnx
# in the model directory; OpenTalking will build flow.decoder.estimator.autocast_fp16.mygpu.plan.
# OPENTALKING_TTS_DEFAULT_PROVIDER=local_cosyvoice
# OPENTALKING_TTS_ENABLED_PROVIDERS=local_cosyvoice,dashscope,edge
# OPENTALKING_LOCAL_AUDIO_MODEL_ROOT=$OPENTALKING_MODEL_ROOT/local-audio
# OPENTALKING_TTS_LOCAL_COSYVOICE_MODEL=FunAudioLLM/Fun-CosyVoice3-0.5B-2512
# OPENTALKING_TTS_LOCAL_COSYVOICE_MODEL_DIR=$OPENTALKING_LOCAL_AUDIO_MODEL_ROOT/FunAudioLLM__Fun-CosyVoice3-0.5B-2512
# OPENTALKING_TTS_LOCAL_COSYVOICE_RUNTIME_DIR=$OPENTALKING_MODEL_REPO_ROOT/CosyVoice
# OPENTALKING_COSYVOICE_VENV_DIR=$OPENTALKING_RUNTIME_ROOT/cosyvoice/venv
# OPENTALKING_TTS_LOCAL_COSYVOICE_SERVICE_URL=http://127.0.0.1:19090/synthesize
# OPENTALKING_TTS_LOCAL_COSYVOICE_DEVICE=cuda:0
# OPENTALKING_TTS_LOCAL_COSYVOICE_FP16=auto
# OPENTALKING_TTS_LOCAL_COSYVOICE_LOAD_TRT=0
# OPENTALKING_TTS_LOCAL_COSYVOICE_TRT_CONCURRENT=1
# OPENTALKING_TTS_LOCAL_COSYVOICE_TOKEN_HOP_LEN=8
# OPENTALKING_TTS_LOCAL_COSYVOICE_TOKEN_MAX_HOP_LEN=16
# OPENTALKING_TTS_LOCAL_COSYVOICE_STREAM_SCALE_FACTOR=1
# OPENTALKING_TTS_LOCAL_COSYVOICE_MAX_TOKEN_TEXT_RATIO=6
# OPENTALKING_TTS_LOCAL_COSYVOICE_MASK_STOP_TOKENS=1

# OmniRT helper defaults.
# OMNIRT_PORT=9000
# OMNIRT_HOST=0.0.0.0
# OMNIRT_FLASHTALK_PORT=18766
# ASCEND_SET_ENV=/usr/local/Ascend/ascend-toolkit/set_env.sh

# QuickTalk OmniRT defaults. QuickTalk currently targets CUDA.
# OPENTALKING_QUICKTALK_ASSET_ROOT=$OPENTALKING_MODEL_ROOT/quicktalk
# OMNIRT_QUICKTALK_MODEL_ROOT=$OPENTALKING_QUICKTALK_ASSET_ROOT
# OMNIRT_QUICKTALK_DEVICE=cuda:0
# OMNIRT_QUICKTALK_HUBERT_DEVICE=cuda:1
# OMNIRT_QUICKTALK_MAX_LONG_EDGE=900
# OMNIRT_QUICKTALK_MAX_TEMPLATE_SECONDS=1
# OMNIRT_QUICKTALK_SCALE_H=1.6
# OMNIRT_QUICKTALK_SCALE_W=3.6
# OMNIRT_QUICKTALK_RESOLUTION=256
# OMNIRT_QUICKTALK_NECK_FADE_START=0.72
# OMNIRT_QUICKTALK_NECK_FADE_END=0.88