Skip to content

webui.md

Latest commit

 

History

History
134 lines (98 loc) · 4.85 KB

webui.md

File metadata and controls

134 lines (98 loc) · 4.85 KB
Raw

Piper-Plus WebUI

Gradio-based web interface for piper-plus inference and training.

There are two separate WebUI implementations: a local development version with inference and training tabs, and a Docker version optimized for standalone inference deployment.

Quick Start (Local)

Requirements

  • Python 3.13+ recommended (3.11+ supported)
  • piper-plus installed
  • ONNX models downloaded

Installation

# Install WebUI dependencies
uv pip install -r src/python_run/requirements_webui.txt

# For training functionality (optional)
uv pip install ".[train]"

Running

cd src/python_run
python -m piper_plus.webui --data-dir ../../test/models

# Or with custom settings
python -m piper_plus.webui \
  --data-dir /path/to/models \
  --host 0.0.0.0 \
  --port 8080 \
  --debug

Options

Option Default Description
--data-dir ./models Directory containing ONNX models
--host 127.0.0.1 Host to bind to
--port 7860 Port to run on
--share off Create a public Gradio link
--debug off Enable debug logging

Features

Inference Tab (both implementations)

  • Model Selection: Auto-detects all .onnx models in the data directory
  • 6言語マルチリンガルモデル対応 (ja, en, zh, es, fr, pt)
  • 言語自動検出: テキスト入力から言語を自動判定
  • 話者選択: マルチスピーカーモデルでの話者切り替え
  • Template System: Language-specific templates (English, Japanese, Chinese, Spanish, French, Portuguese)
  • Speed Control: 0.5-2.0x
  • Noise Parameters: Expressiveness and phoneme width variation
  • Audio Output: Play and download generated speech

Training Tab (local only)

  • Dataset path validation with structure check
  • Base model selection or new model training
  • Quality selection (low/medium/high)
  • Training parameters (batch size, learning rate, epochs)
  • Start/Stop controls with real-time log display

Architecture

There are two independent WebUI implementations:

Local development version (src/python_run/piper_plus/webui.py) -- uses PiperVoice runtime for inference and includes a training management tab via training_manager.py.

Docker version (docker/webui/app.py) -- a standalone Gradio app that uses piper_train.infer_onnx and piper_train.ort_utils directly for ONNX inference. Inference only, no training tab. Includes session caching and warmup via create_session_with_cache.

src/python_run/piper_plus/
├── webui.py              # Local WebUI (inference + training)
├── training_manager.py   # Training management backend
├── sample_texts.py       # Sample text collections
└── requirements_webui.txt

docker/webui/
├── app.py                # Docker WebUI (inference only, standalone)
├── Dockerfile
├── docker-compose.yml
├── entrypoint.sh
└── run.sh

Key Design Decisions

  • Gradio Framework: ML-optimized UI components with built-in audio playback
  • Language Detection: Automatic model-to-language mapping with template adaptation
  • Lazy Model Loading: Models loaded on synthesis, not on startup
  • Two implementations: Local version depends on the piper-plus runtime package; Docker version depends on piper_train directly, avoiding the runtime dependency

Docker Usage

The Docker image uses docker/webui/app.py (not the local webui.py).

# Build
docker build -t piper-webui -f docker/webui/Dockerfile .

# Run
docker run -p 7860:7860 -v ./models:/models piper-webui

# Or docker-compose
cd docker/webui && docker-compose up

Environment Variables

Variable Default Description
MODELS_DIR ./models Host path to model directory (docker-compose volume)
OUTPUT_DIR ./output Host path to output directory (docker-compose volume)
PIPER_PLUS_MODEL (none) Specific model to load (passed to entrypoint)
PIPER_PLUS_MODEL_DIR /models Model directory inside the container
PIPER_PLUS_OUTPUT_DIR /output Output directory inside the container (used by entrypoint.sh)

Note: docker-compose.yml sets GRADIO_SERVER_NAME and GRADIO_SERVER_PORT, but app.py launches Gradio with explicit --host / --port CLI args (defaults: 0.0.0.0 and 7860), so those env vars have no effect. To change the bind address or port, override the entrypoint command args instead.

Troubleshooting

Issue Solution
No models found Check --data-dir path (local) or volume mount (Docker); ensure .onnx and .onnx.json pairs exist
Import errors uv pip install -r src/python_run/requirements_webui.txt
Port in use Use --port 8080 (local) or change the port mapping in docker-compose
Docker issues Check volume mounts and port availability