Practical overview
When it fits
Use it when you need to run Orpheus-based text-to-speech behind a web interface or an OpenAI-compatible speech endpoint, and can also provide the required model inference service.
Who it is for
- Developers integrating speech generation into applications through an OpenAI-style endpoint.
- People who want to operate a speech-generation server and use its web interface.
- Teams building multilingual text-to-audio workflows with control over voice and expressive tags.
Documented uses
- Convert submitted text into WAV speech.
- Choose a voice and adjust speech speed through the API parameters.
- Generate speech in the listed languages and add supported expressive tags.
- Provide long-form audio generation using the project's described batching and crossfading.
Requirements & limits
The repository is public and identifies the project as Apache-2.0 licensed. The primary service requires a running Orpheus inference server; Docker Compose can orchestrate one with the FastAPI service.
- The service depends on a separate Orpheus model inference server unless using the supplied Docker Compose setup.
- The README documents CUDA, ROCm and CPU Compose configurations, with NVIDIA GPU requirements specified for the CUDA Docker path.
- Native installation requires Python 3.8–3.11; Python 3.12 is documented as unsupported.
- The documented speech response format is WAV.
- The README lists 24 voices across eight languages and identifies selectable voice, speed and emotion-tag controls.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Start the speech service with Docker Compose
Set up the FastAPI service alongside its Orpheus model inference server using a documented Docker Compose configuration.
Before you start
- Docker Compose
- A machine suitable for one of the documented GPU or CPU configurations
- For NVIDIA GPU Docker support: Linux or Windows, CUDA 12.4 or greater, and NVIDIA Container Toolkit
Steps
Create a local .env file from .env.example.
For a language-specific model, set ORPHEUS_MODEL_NAME in .env to the desired model filename.
Start the appropriate configuration with docker compose -f docker-compose-gpu.yml up, docker compose -f docker-compose-gpu-rocm.yml up, or docker compose -f docker-compose-cpu.yml up.
Expected outcome
The Compose setup coordinates the FastAPI audio service and inference server, and downloads the selected model before service startup.
Guide limitations
- The NVIDIA GPU configuration requires CUDA 12.4 or greater and NVIDIA Container Toolkit.
- The project provides separate GPU and CPU Compose files; the appropriate option depends on the machine and GPU stack.
Prepare and start a native FastAPI installation
Install the documented Python service and start its local web interface and API.
Before you start
- Python 3.8–3.11; Python 3.12 is documented as unsupported
- PyTorch installed for the intended CUDA or ROCm environment
- An Orpheus inference server, such as llama.cpp or LM Studio
Steps
Clone the repository and enter its directory.
Create and activate a Python virtual environment.
Install the matching CUDA or ROCm PyTorch packages, then install dependencies from requirements.txt.
Create the outputs and static directories, then start the service with python app.py.
Expected outcome
The README gives localhost addresses for the web interface and API documentation after the server starts.
Guide limitations
- The inference server must be running an Orpheus model and expose an endpoint the FastAPI application can use.
- The documented native prerequisites exclude Python 3.12.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 13 October 2026
- Original website title
- Lex-au/Orpheus-FastAPI · Official homepage title · retained in its source language