- Home
- AI Sound Effect Generators
- AudioGen
AudioGen
What it is
AudioGen is a text-guided audio generation model provided through AudioCraft. It generates sounds from written descriptions, with options for conditional or unconditional generation and continuing audio from a prompt. Sampling controls include greedy, temperature, top-K, and top-P methods. The listed pretrained model is facebook/audiogen-medium, with 1.5 billion parameters; an example API can generate a sample and save it as WAV. The implementation combines an autoregressive Transformer with a 16 kHz EnCodec tokenizer. The project provides API instructions and a Jupyter notebook demo for local use. Local inference with the medium model requires a GPU with at least 16 GB of memory. Installation requires Python 3.9 and PyTorch 2.1.0, and ffmpeg is recommended. AudioGen was trained on English descriptions and is less capable in other languages. It cannot produce realistic vocals, and prompt engineering may be needed for satisfactory results. The code is MIT-licensed, while model weights use CC-BY-NC 4.0. Training datasets are not provided, and the model card calls for risk evaluation and mitigation before downstream use.
For and against
For
- Generates sounds from text descriptions.
- Supports continuation from an audio prompt.
- Offers multiple sampling controls.
- Provides a local Jupyter notebook demo.
- Example output can be saved as WAV.
Against
- Local medium-model inference requires a GPU with 16 GB memory.
- Trained on English descriptions; other languages perform less well.
- Cannot generate realistic vocals.
- Training datasets are not provided.
The verdict
AudioGen offers text-driven sound generation with prompt continuation and several sampling methods. Consider its hardware requirements, language limitations, and model-weight license before choosing it for a project.
Who it suits
It suits audio, machine learning, and AI researchers, as well as amateurs learning about generative models. Local use is best suited to people with the specified GPU memory and software requirements.
Plans and prices
AudioGen (self-hosted)
Pretrained medium model · local inference requires a GPU with at least 16 GB memory
Read 3 Oct 2026 · github.com
On the record
- Audio continuation
- The generation stage supports conditional or unconditional sample generation and audio continuation from a prompt.github.com · read 3 Oct 2026
- Available model
- The page lists one pretrained AudioGen model, facebook/audiogen-medium, with 1.5 billion parameters.github.com · read 3 Oct 2026
- Generation
- The example API generates sound from text descriptions and shows saving output as WAV audio.github.com · read 3 Oct 2026
- Generation modes
- The training documentation describes conditional and unconditional generation, audio continuation from a prompt, and greedy, temperature, top-K, and top-P sampling.github.com · read 3 Oct 2026
- Hardware requirement
- The instructions say inference with the medium-sized models requires a GPU with at least 16 GB of memory.github.com · read 3 Oct 2026
- Installation
- AudioCraft installation requires Python 3.9 and PyTorch 2.1.0; ffmpeg is also recommended by the repository instructions.github.com · read 3 Oct 2026
- Intended users
- The model card names audio, machine learning, and AI researchers, as well as amateurs learning about generative models, as primary users.github.com · read 3 Oct 2026
- Language limit
- The model card says AudioGen was trained on English descriptions and performs less well in other languages.github.com · read 3 Oct 2026
- License
- The repository says its code is MIT-licensed and its model weights use CC-BY-NC 4.0.github.com · read 3 Oct 2026
- Local demo
- The maker provides a Jupyter notebook demo that can be run locally with a GPU.github.com · read 3 Oct 2026
- Model architecture
- The released model combines EnCodec audio tokenization with an autoregressive Transformer language model and has 1.5 billion parameters.github.com · read 3 Oct 2026
- Model availability
- The AudioGen instructions list one pretrained model, facebook/audiogen-medium, and a local Jupyter notebook demo.github.com · read 3 Oct 2026
- Model design
- The provided reimplementation is a single-stage autoregressive Transformer trained over a 16 kHz EnCodec tokenizer with four codebooks sampled at 50 Hz.github.com · read 3 Oct 2026
- Model distinction
- The provided models are not the original models used to report results in the AudioGen publication.github.com · read 3 Oct 2026
- Output limit
- The model card says AudioGen cannot generate realistic vocals and may require prompt engineering for satisfying results.github.com · read 3 Oct 2026
- Prompt generation
- The documented API generates audio samples from text descriptions, with an example configured to generate five-second samples.github.com · read 3 Oct 2026
- Purpose
- AudioGen is a text-to-sound generation model provided through AudioCraft.github.com · read 3 Oct 2026
- Responsible use
- The model card advises against downstream use without further risk evaluation and mitigation.github.com · read 3 Oct 2026
