- Home
- AI Sound Effect Generators
- Sony Woosh
On the recordRead 5 Oct 202618 facts
Sony Woosh
On the record
- Additional models
- Release v1.0.1 provides SonicCLAP-AR for audio-text retrieval and SonicCLAP-MOS for alignment with human perceptual judgments.github.com · read 5 Oct 2026
- API
- A self-run API server is provided for inference, but currently supports Woosh-DFlow only.github.com · read 5 Oct 2026
- Audience
- The repository describes inference scripts, local demos, and an API server for running sound effect generation models.github.com · read 5 Oct 2026
- Audio models
- Woosh-AE provides audio encoding and decoding, while Woosh-CLAP provides text-audio alignment for diffusion conditioning.github.com · read 5 Oct 2026
- Compute options
- Installation instructions offer either CPU or CUDA support.github.com · read 5 Oct 2026
- Downloads
- Pretrained model weights and sample media are available through GitHub releases.github.com · read 5 Oct 2026
- Inference options
- The repository provides installation options with CPU or CUDA support and inference test scripts for each model.github.com · read 5 Oct 2026
- Licensing
- Most repository code is MIT-licensed, code for Woosh-VFlow and Woosh-DVFlow is Apache 2.0 licensed, and released model weights are CC-BY-NC licensed.github.com · read 5 Oct 2026
- Local demos
- The repository includes local Gradio demos for Woosh-Flow and Woosh-DFlow, accessed in a browser on the same machine.github.com · read 5 Oct 2026
- Model components
- The release includes an audio encoder/decoder, a text-audio alignment model, text-to-audio and video-to-audio generators, and distilled versions of the generators.arxiv.org · read 5 Oct 2026
- Model weights
- Open weights for models trained on public datasets are available from the repository’s releases.github.com · read 5 Oct 2026
- Open weights
- The repository provides inference code and open weights for models trained on public datasets.github.com · read 5 Oct 2026
- Organization
- The repository identifies Woosh as a public release of a sound effect foundation model by Sony AI.github.com · read 5 Oct 2026
- Purpose
- Woosh is Sony AI’s publicly released sound effects foundation model, with inference code and open model weights.github.com · read 5 Oct 2026
- Support
- The repository welcomes pull requests and asks contributors to open an issue before making major changes.github.com · read 5 Oct 2026
- Text-to-audio
- Woosh-Flow and Woosh-DFlow generate audio unconditionally or from a text prompt.github.com · read 5 Oct 2026
- Use case
- The models are optimized for sound effects and are presented as tools for audio research and building approaches or baselines.arxiv.org · read 5 Oct 2026
- Video-to-audio
- Woosh-VFlow generates audio from a video sequence and can take optional text prompts.github.com · read 5 Oct 2026
