Google DeepMind describes a music generation model and experiments intended to explore new ways to create music.
StemGen: A music generation model that listens
Examples for ICASSP2024 paper “StemGen: A music generation model that listens”.
About StemGen: A music generation model that listens
StemGen is an end-to-end music generation model trained to listen to musical context and respond appropriately. It uses a non-autoregressive language-model type architecture similar to SoundStorm and VampNet. This website accompanies the ICASSP 2024 paper and presents example outputs from three model variants: slakh, internal, and mingus. The examples cover test-set generation, where context audio is built from a test dataset partition and the model generates a single stem with conditioning chosen at random, and iterative generation, where a generated stem is mixed with existing audio and then used as context for further generations. The variants use target instrument category or genre/stem conditioning and 32 kHz or stereo 48 kHz Encodec tokenizers, with around 250M parameters. Mingus is noted as not presented in the paper.
Key features
- Generates a single instrument stem from musical context audio.
- Conditions on target instrument category or genre/stem category.
- Supports iterative layering using generated stems as new context.
- Demonstrates rhythm following without initial harmonic or melodic information.
- Uses 32 kHz or stereo 48 kHz Encodec tokenizers by variant.
- Includes variants around 250M parameters trained on Slakh2100 or internal data.
Best for: Readers and researchers who want to hear example outputs from the StemGen architecture described in the ICASSP 2024 paper.
Good for
- Provides audio examples for test-set and iterative generation scenarios.
- Details each model variant's dataset, conditioning, tokenizer, and parameter count.
- Includes iterative examples of chat-like music construction and rhythm following.
Watch out for
- The site is an examples page for a paper, not described as an end-user generation service.
- Mingus is noted as not presented in the paper.
- The project is not open source, and no pricing information is provided.
StemGen: A music generation model that listens at a glance
- Category
- AI & Machine Learning › AI Music Generation
- Type
- Website
- Pricing
- Not confirmed. Check the site
- Platforms
- Web
- Open source
- Not stated
- Language
- English
- Last checked
- Oct 5, 2026 · Online when checked
- Listed
- Oct 5, 2026
StemGen: A music generation model that listens alternatives
Tools and sites you could use instead, ranked by how closely they match.
ACE-Step is a local music generation model supporting Mac, AMD, Intel, and CUDA devices.
Open source
24 alternatives
Stable Audio provides customizable models and tools for iterative music and sound creation.
AIVA provides an AI music generation assistant for creating original music.
Paid
24 alternatives
Suno AI Music generates songs from text prompts and offers simple, custom and instrumental modes alongside a selectable model.
Ecrett Music lets users create royalty-free music tracks through its online service.
Popular in AI Music Generation
Well-known listings from the same topic.
MusicCreator AImusiccreator.aiGoogle Flow Musicflowmusic.appMubertmubert.comAudioXaudiox.appAisongmakeraisongmaker.io
StemGen: A music generation model that listens FAQ
What is StemGen?
StemGen is an end-to-end music generation model trained to listen to musical context and respond appropriately. The website presents example outputs from models of its architecture.
How does iterative generation work on this site?
A context audio is provided, a new stem is generated and mixed with the existing audio, and the new mixed audio becomes context for generating another stem; the process repeats.
Which models are shown?
The page presents slakh, internal, and mingus. Slakh uses Slakh2100; internal uses 500 hours of human-played music in individual instrument stems; mingus was pretrained on synthetic data and fine-tuned on 2 hours of human-composed and produced music. Mingus is noted as not presented in the paper.
Is something wrong or out of date?
Tell us and an editor will check it.
Own this website?
Claim this listing
Claim the listing and suggest updates yourself. An editor checks every claim first.