Generates and decodes sound effects from text prompts or video sequences using open foundation models.
Sony AI's open foundation models for audio encoding, text-driven sound effects, and video-synced audio.
Woosh is a sound effect foundation toolkit built by Sony AI for text-to-audio and video-to-audio synthesis. It offers latent diffusion models alongside an audio encoder and decoder for sound designers and game developers. You can run inference scripts, launch local browser interfaces, or integrate generation into custom pipelines via an API server.
The repository provides a suite of models including Woosh-AE for latent audio encoding, Woosh-CLAP for text conditioning, and latent diffusion models like Woosh-Flow and Woosh-VFlow. It synthesizes audio directly from descriptive text prompts or aligns generated audio to input video files. You manage and execute the tools through Python scripts using the uv environment manager.
It suits game developers, technical sound designers, and AI audio researchers looking to generate original audio assets locally. It is less suited for non-technical creators who require a finished digital audio workstation plugin or immediate commercial rights.
Synthesizes custom sound effects directly from text descriptions using standard and distilled diffusion models.
Creates audio tracks aligned with input video sequences using multimodal latent diffusion.
Compresses raw audio into neural latents and reconstructs high-quality waveforms with Woosh-AE.
Provides web-based user interfaces to test text-to-sound generation in your local browser.
The source code is free and open under MIT and Apache-2.0 licenses. However, the pre-trained model weights use a non-commercial CC-BY-NC license. Running the project requires your own local GPU compute hardware.
896 open remote AI-training roles, updated November 16, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
These reviewed AI tools do the same job instantly — from our scored directory.
The repository source code uses Apache-2.0 and MIT licenses, but the open model weights provided on GitHub releases are licensed under CC-BY-NC, restricting commercial usage without separate arrangements.
No. Once you install the dependencies and download the model checkpoints locally, all audio generation runs offline.
See Zekai first in Google