Applio

AI & Audio

Train your own voice model and convert audio on your own GPU.

Applio is a voice conversion tool built on RVC (Retrieval-based Voice Conversion). You can train a voice model from a few minutes of clean speech, then re-voice any recording so it sounds like that model, entirely on your own GPU and without sending audio to a cloud service. The web interface covers training, single and batch conversion, text-to-speech with voice conversion, model blending and real-time conversion. Use it with your own voice or with the clear permission of the person whose voice you are modelling. It is a self-hosted alternative to paid voice-changer and voice-cloning services.

License

MIT

GitHub stars

3,814

Official website

https://applio.org

Source code

https://github.com/IAHispano/Applio

Compatibility

Architectures

amd64

Database

Stateless

Resource footprint

6144 MB RAM · 4 CPU

Needs an NVIDIA GPU for practical training. In our test LXC it idled at about 0.7 GB of RAM, and training a voice model on 7 minutes of audio (200 epochs, 40 kHz) finished in roughly 25 minutes on an RTX 3090 inside a 6 GB memory limit. Converting a 15-second clip takes a few seconds. The Docker image is about 17 GB, so plan for 40 GB or more of disk.

Proof of Work Gallery

Inference
Conversion
Training

Deployment

# Needs an NVIDIA GPU, the NVIDIA driver and nvidia-container-toolkit on the host.
# The first build downloads PyTorch with CUDA and takes several minutes (image is ~17 GB).
services:
  applio:
    build:
      context: .
      dockerfile_inline: |
        FROM python:3.12-trixie
        RUN apt-get update \
         && apt-get install -y --no-install-recommends git ffmpeg libportaudio2 \
         && rm -rf /var/lib/apt/lists/*
        RUN git clone --depth 1 https://github.com/IAHispano/Applio.git /app
        WORKDIR /app
        RUN python -m venv .venv \
         && .venv/bin/pip install --no-cache-dir --upgrade pip python-ffmpeg \
         && .venv/bin/pip install --no-cache-dir -r requirements.txt \
         && .venv/bin/pip install --no-cache-dir --no-deps torchvision==0.26.0 pedalboard==0.9.16
        ENV PATH="/app/.venv/bin:$PATH" \
            LD_LIBRARY_PATH="/app/.venv/lib/python3.12/site-packages/nvidia/cu13/lib"
        EXPOSE 6969
        ENTRYPOINT ["python3"]
        CMD ["app.py", "--server-name", "0.0.0.0", "--port", "6969"]
    container_name: applio
    restart: unless-stopped
    runtime: nvidia
    environment:
      - NVIDIA_VISIBLE_DEVICES=all
      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
    shm_size: 2gb
    ports:
      - "6969:6969"
    volumes:
      - ./applio-logs:/app/logs
      - ./applio-models:/app/rvc/models

Alternative to

ElevenLabs, Respeecher, Voicemod