IndexTTS Local Deployment¶

IndexTTS is integrated through OpenTalking's indextts provider. Use it for controllable dubbing, emotion control, and cloned voices. This page covers the same-machine HTTP sidecar shape.

Use Cases¶

More voice control than the default Edge TTS path.
IndexTTS runtime should be isolated from the main OpenTalking process.
TTS must run locally instead of through a hosted API.

Deployment Root¶

Run the following commands in the same terminal. Enter the OpenTalking repository directory and define the deployment root once:

Terminal

cd /path/to/digital_human/opentalking
export OPENTALKING_HOME="$(pwd)"
case "${DIGITAL_HUMAN_HOME:-}" in
  ""|/path/to/digital_human) export DIGITAL_HUMAN_HOME="$(cd "$OPENTALKING_HOME/.." && pwd)" ;;
esac
export OPENTALKING_MODEL_ROOT="$DIGITAL_HUMAN_HOME/models"
export OPENTALKING_MODEL_REPO_ROOT="$DIGITAL_HUMAN_HOME/model-repos"
export OPENTALKING_RUNTIME_ROOT="$DIGITAL_HUMAN_HOME/runtimes"
export OPENTALKING_LOCAL_AUDIO_MODEL_ROOT="$OPENTALKING_MODEL_ROOT/local-audio"
export OPENTALKING_TTS_LOCAL_INDEXTTS_RUNTIME_DIR="$OPENTALKING_MODEL_REPO_ROOT/index-tts"
export OPENTALKING_INDEXTTS_VENV_DIR="$OPENTALKING_RUNTIME_ROOT/index-tts/venv"

Weight Preparation¶

Use consistent deployment directories: weights under $OPENTALKING_LOCAL_AUDIO_MODEL_ROOT, the IndexTTS source checkout under $OPENTALKING_MODEL_REPO_ROOT, and the sidecar venv under $OPENTALKING_RUNTIME_ROOT.

Terminal

cd "$OPENTALKING_HOME"
mkdir -p "$OPENTALKING_LOCAL_AUDIO_MODEL_ROOT"

export HF_ENDPOINT="${HF_ENDPOINT:-https://hf-mirror.com}"
# Some networks repeatedly fail on the HF Xet/CAS path; prefer plain HTTP downloads.
export HF_HUB_DISABLE_XET="${HF_HUB_DISABLE_XET:-1}"
uv sync --extra dev --extra models --extra local-audio --extra quicktalk-cuda --python 3.11

python scripts/download_local_audio_models.py \
  --root "$OPENTALKING_LOCAL_AUDIO_MODEL_ROOT" \
  --model indextts2 \
  --model indextts2-w2v-bert \
  --model indextts2-maskgct \
  --model indextts2-campplus \
  --model indextts2-bigvgan

Prepare the runtime:

Terminal

cd "$DIGITAL_HUMAN_HOME"
mkdir -p "$OPENTALKING_MODEL_REPO_ROOT" "$OPENTALKING_RUNTIME_ROOT/index-tts"
# Optional: use a GitHub proxy prefix when GitHub access is slow.
# export GITHUB_PROXY_PREFIX=https://gh-proxy.com/
# Optional: use the Tsinghua PyPI mirror when PyPI access is slow.
export UV_DEFAULT_INDEX="${UV_DEFAULT_INDEX:-https://pypi.tuna.tsinghua.edu.cn/simple}"
export PIP_INDEX_URL="${PIP_INDEX_URL:-$UV_DEFAULT_INDEX}"
# Avoid failures when ~/.cache/uv or ~/.cache/pip was previously written by root.
export UV_CACHE_DIR="${UV_CACHE_DIR:-$OPENTALKING_RUNTIME_ROOT/.uv-cache}"
export PIP_CACHE_DIR="${PIP_CACHE_DIR:-$OPENTALKING_RUNTIME_ROOT/.pip-cache}"
if [ ! -d "$OPENTALKING_MODEL_REPO_ROOT/index-tts/.git" ]; then
  GIT_LFS_SKIP_SMUDGE=1 git clone "${GITHUB_PROXY_PREFIX:-}https://github.com/index-tts/index-tts.git" "$OPENTALKING_MODEL_REPO_ROOT/index-tts"
fi

mkdir -p "$UV_CACHE_DIR" "$PIP_CACHE_DIR"
uv venv --seed --python 3.11 "$OPENTALKING_RUNTIME_ROOT/index-tts/venv"
"$OPENTALKING_RUNTIME_ROOT/index-tts/venv/bin/python" -m pip install -U pip wheel setuptools
cd "$OPENTALKING_MODEL_REPO_ROOT/index-tts"
"$OPENTALKING_RUNTIME_ROOT/index-tts/venv/bin/python" -m pip install -e .
cd "$DIGITAL_HUMAN_HOME"
"$OPENTALKING_RUNTIME_ROOT/index-tts/venv/bin/python" -m pip install fastapi "uvicorn[standard]" soundfile

Do not install index-tts into OpenTalking's main .venv; OpenTalking calls the sidecar over HTTP.

Configuration¶

.env

DIGITAL_HUMAN_HOME=/path/to/digital_human
OPENTALKING_TTS_DEFAULT_PROVIDER=indextts
OPENTALKING_TTS_INDEXTTS_BACKEND=local
OPENTALKING_TTS_LOCAL_INDEXTTS_SERVICE_URL=http://127.0.0.1:19092/synthesize
OPENTALKING_TTS_LOCAL_INDEXTTS_DEVICE=cuda:0
OPENTALKING_TTS_VOICE=indextts-xiaoxiao-cn

Only replace DIGITAL_HUMAN_HOME with the real deployment root. The model, runtime, and venv paths below are derived from it in the start command, which avoids stale absolute paths in .env.

When OmniRT hosts the IndexTTS runtime, OpenTalking still uses provider=indextts; set the backend to omnirt and point it at the resident OmniRT service. OmniRT owns model loading, segmented streaming, and token-window streaming:

.env

OPENTALKING_TTS_DEFAULT_PROVIDER=indextts
OPENTALKING_TTS_INDEXTTS_BACKEND=omnirt
OPENTALKING_TTS_OMNIRT_INDEXTTS_SERVICE_URL=http://127.0.0.1:9012/v1/text2audio/indextts
OPENTALKING_TTS_OMNIRT_INDEXTTS_STREAMING_MODE=token_window
OPENTALKING_TTS_OMNIRT_INDEXTTS_TOKEN_WINDOW_SIZE=40

Start Command¶

Start the IndexTTS sidecar first, then start OpenTalking + QuickTalk. The sidecar HTTP path and port must match OPENTALKING_TTS_LOCAL_INDEXTTS_SERVICE_URL.

Terminal

cd "$OPENTALKING_HOME"

bash scripts/quickstart/start_local_indextts.sh --port 19092 --device cuda:0

export OPENTALKING_TTS_DEFAULT_PROVIDER=indextts
export OPENTALKING_TTS_INDEXTTS_BACKEND=local
export OPENTALKING_TTS_LOCAL_INDEXTTS_SERVICE_URL=http://127.0.0.1:19092/synthesize
export OPENTALKING_TTS_VOICE=indextts-xiaoxiao-cn

export OPENTALKING_TORCH_DEVICE=cuda:0
export OPENTALKING_QUICKTALK_DEVICE=cuda:0
export OPENTALKING_QUICKTALK_ASSET_ROOT="$OPENTALKING_MODEL_ROOT/quicktalk"
export OPENTALKING_QUICKTALK_WORKER_CACHE=1

bash scripts/start_unified.sh --backend local --model quicktalk --api-port 8210 --web-port 5283

Verification¶

Terminal

curl -fsS http://127.0.0.1:19092/health
curl -fsS http://127.0.0.1:8210/health
curl -fsS http://127.0.0.1:8210/runtime/status | jq '.tts_providers.indextts.service_url_set'

The status value should be true. After creating a quicktalk session, call /speak to verify that the TTS provider returns audio and drives QuickTalk.

Stop Services¶

Terminal

cd "$OPENTALKING_HOME"
bash scripts/quickstart/stop_all.sh --api-port 8210 --web-port 5283

This stops the OpenTalking API, WebUI, and the IndexTTS sidecar started above with a pid file.

Common Errors¶

Symptom	Action
`service_url_set=false`	Use `OPENTALKING_TTS_LOCAL_INDEXTTS_SERVICE_URL`, not `OPENTALKING_TTS_INDEXTTS_SERVICE_URL`.
Sidecar API path mismatch	Check that the sidecar port and path match `OPENTALKING_TTS_LOCAL_INDEXTTS_SERVICE_URL`; the default port is `19092`.
Downloads repeatedly mention `cas-bridge.xethub.hf.co` or SSL resume	Keep `HF_HUB_DISABLE_XET=1` and rerun the same download command; the script reuses fully downloaded model directories. If needed, temporarily use `HF_ENDPOINT=https://huggingface.co`.
Missing downloaded files	Re-run the download script and confirm all five `indextts2*` model directories exist.
QuickTalk reports `onnxruntime` has no `InferenceSession`	Re-run `uv sync --extra dev --extra models --extra local-audio --extra quicktalk-cuda --python 3.11`; on CPU replace `quicktalk-cuda` with `quicktalk-cpu`.
Dependency conflicts	Keep the IndexTTS runtime in its own venv and do not install it into OpenTalking's main `.venv`.