DADABOTS
DADABOTS is an AI tool specifically designed for music production that leverages neural networks to automate the music creation process. By generating raw audio...
Last verified:
What is DADABOTS?
DADABOTS is an AI music research project and experimental band that uses neural networks to generate raw audio, specifically focusing on extreme music genres like death metal, black metal, math rock, skate punk, and beatbox. Founded in 2012 at Music Hack Day MIT by musicians CJ and Zack, the project combines band performance, hackathon-style development, and research lab activities to eliminate human involvement in music creation through deep learning. The team runs a famous 24/7 endless death metal livestream on YouTube that streams neuralネットワーク-generated death metal to infinity.
The tool uses modified SampleRNN architecture, a hierarchical LSTM network that generates raw audio in the time domain rather than MIDI or symbolic music. This approach captures timbre and space as compositional elements, which is essential for modern extreme music styles. The system trains on raw acoustic waveforms of metal albums, guessing the next fraction of a millisecond millions of times over several days. DADABOTS has also developed DadaGP, a tokenized GuitarPro dataset with 26,181 songs across 739 genres, and contributed to Stability AI's Stable Audio Open model for on-device audio generation.
DADABOTS is designed for extreme music enthusiasts, AI researchers, experimental musicians, and developers interested in neural synthesis. The project appeals to underground music cartographers who want to explore new sonic territories, programmers who want to build AI music tools, and artists seeking collaborative AI instruments. Their open-source approach includes releasing code on GitHub, publishing scientific papers at conferences like NIPS and ISMIR, and collaborating with avant-garde musicians like Mike Patton, Reeps ONE, and bands like Krallice and Lightning Bolt.
Key features include the infinite death metal livestream, Genre Cannon for cramming multiple genres into minutes, prompt jockeying for live DJing with AI, on-device audio generation through their Arm partnership, and multiple releases like the WOW LOL EP (Kawaii Deathcore Drill and Bass) and Relentless Doppelganger (Neural Technical Death Metal inspired by Archspire). The project emphasizes absurdism, open-source code release, and artist collaborations while treating AI as an instrument to break and rebuild in real time.
DADABOTS pricing
Pricing model: Free
DADABOTS is completely free and open-source. All code is released on GitHub including DadaGP encoder/decoder, Stable Audio Tools, and Dadabots-modified SampleRNN. Research papers are published free on arXiv. The 24/7 death metal livestream on YouTube is free to listen. Generated music albums are available on Bandcamp. The Stable Audio Open model is available on HuggingFace. No paid tiers or subscriptions are offered - the project operates as open research with collaboration-based funding.
DADABOTS pros
- Generates raw audio in time domain instead of MIDI for authentic timbre
- 24/7 endless death metal livestream on YouTube available forever
- Specialized in extreme music genres like black metal and math rock
- Open-source code released on GitHub for community use
- Publishes peer-reviewed scientific research at major conferences
- DadaGP dataset includes 26,181 songs across 739 musical genres
- Collaborates with avant-garde artists like Mike Patton and Reeps ONE
- On-device audio generation works offline on Arm mobile devices
- Stable Audio Open runs 30x faster on smartphones than previous versions
- Genre Conditioning allows mixing and blending any bands together
- No human involvement needed for music generation once trained
- Can generate 3-minute songs in 10 seconds with Stable Audio
- Creates unique characteristic artifacts of neural synthesis as artistic feature
- Suitable for music therapy and giving cultural voice to underserved groups
- Hyperparameter tuning allows control over output style and quality
DADABOTS cons
- Audio quality is lo-fi and lower fidelity compared to commercial tools
- Training requires expensive V100 GPUs initially prohibitively costly
- Original SampleRNN runs in Theano and is outdated and difficult to install
- Generating 1 minute of audio takes 1 hour with older models
- No control over specific musical content or sections in output
- Requires curation tool to explore 10+ hours of generated audio
- Models can overfit to learn only one part while ignoring rest of music
- Raw audio generation needs heavy-duty hardware only recently available
Frequently asked questions about DADABOTS
What is DADABOTS?
DADABOTS is a cross between a band, a hackathon team, and an ephemeral research lab. They are musicians who do the science, engineer the software, and make the music all in one project. Founded in 2012 at Music Hack Day MIT by CJ and Zack, they use neural networks to generate raw audio and eliminate humans from music. Their ethos includes absurdism, releasing open-source code, and collaborating with artists and students.
How does creating music with SampleRNN work?
SampleRNN is a hierarchical LSTM network trained on raw acoustic waveforms of metal albums. It listens and tries to guess the next fraction of a millisecond, playing this game millions of times over a few days. After training, it hallucinates 10 hours of music which is then curated using the D.O.M.E. tool to find and arrange the best bits into albums for human consumption.
Why do you focus on math rock and black metal instead of mainstream genres?
Mainstream music is described as dead and rigid, while the underground is home to real explorers and scientists of music. Math Rock and Black Metal are the music the founders love with a special place in their hearts. New black metal bands like Krallice push the genre to new places never felt before, making the research fresh rather than rehashing old sounds like publishing scientific papers on the same old experiments.
What is the difference between your approach and MIDI-based computer generated music?
DADABOTS trains completely unsupervised with no knowledge of music theory, no MIDI, and nothing - just raw audio. MIDI is only 2% of what there is to love about music and cannot capture atmosphere, timbre, or modern styles. Raw audio is unruly and became tractable only recently with heavy-duty hardware and cleverer algorithms from DeepMind and MILA's text-to-speech research.
Do human creators need to fear for their jobs?
If designed with people in mind, AI is a creative tool, but full automation means some creative professionals will be replaced, especially easily commoditizable work. The advice is to automate your job and be the one running the machine. DADABOTS' aim is 100% human augmentation rather than replacement, with several collaborations in the works with bands like Lightning Bolt, Artificial Brain, and Krallice.
How did you get started working on DADABOTS?
On the first day they met in 2012, CJ said Zack felt like he'd known him his whole life. They formed a hackathon team at Music Hack Day MIT intrigued by the pointlessness of machines generating crappy art. They announced they would destroy SoundCloud by creating an army of remix bots that spidered SoundCloud for music to remix, posting hundreds of songs an hour. They kept getting banned but kept working around it.
What inspired you to program AI to create music?
They saw the potential of deep neural networks when image style transfer was released, watching photographs transform into impressionist oil paintings. They had been researching ways to model musical style and generate music with a target timbre. Deep learning seemed like the tool they were looking for once Wavenets could synthesize human voices in multiple languages.
Will your AI be able to generate new music purely on its own at some stage?
They would need to add more functionality for AI to choose its own music to generate. Having access to more kinds of music and a way to condition between artists or styles would allow something resembling self-evolving musical taste. The AI would need to pick what music to create on its own based on audience feedback, critics' feedback, or maximizing novelty and curiosity.
Why does the audio quality sound low and why don't you generate high fidelity?
The desire for high fidelity is described as never-ending chase-the-dragon sensory hedonism that encourages centralization of media to only groups with the largest budgets. That kind of world is non-participatory and disempowers DIY media, whereas the most important voices come from disenfranchised people. Lower fidelity makes the technology more accessible and participatory.
What is next for the DADABOTS project?
They want to build a spaceship for navigating the cosmos of all possible music, mixing bands together and blending anything with anything to discover music never before imagined. The goal is to make it so easy for kids to invent a new genre and set up bots that make music in that genre forever, creating a spaceship for exploring all possible musical territories.