AudioLM
No estimate
No hardware requirements for this model
The weights for this model have not been published, so it cannot be downloaded or run on your own hardware at any size. It is reachable only through its provider, and no graphics card changes that.
On record
Full specification
Everything on record for this model. Most of it describes how it was trained rather than how it runs — useful context for judging how much work went into it, and how it compares with models built at a different scale.
Origin
Who built this model, where, and when it was published.
- Organisation
- Google Research
- Organisation type
- Industry
- Country
- United States of America
- Published
- 26 July 2023
- Authors
- Zalán Borsos, Raphaël Marinier, Damien Vincent, Eugene Kharitonov, Olivier Pietquin, Matt Sharifi, Dominik Roblek, Olivier Teboul, David Grangier, Marco Tagliasacchi, Neil Zeghidour
What it does
The problem areas the model was built for. A model can carry several of each.
- Domain
- Audio
- Task
- Audio generation
- Approach
- Self-supervised learning
Size
How large the model is and how much data it was trained on. Parameters are the figure that decides whether it fits on a given graphics card.
- Parameters
- 1.5B
- Training data
- 135,000,000,000 tokens
"We use identical decoder-only Transformers in all stages, with 12 layers, 16 attention heads, embedding dimension of 1024, feed-forward layer dimension of 4096 and dropout of 0.1, together with T5-style relative positional embeddings [38], resulting in a model parameter size of 0.3B per stage." Three stages (figure 2), and 300M per stage. Plus 600M parameters for w2v-BERT XL, so 1.5B total
60k hours of English speech 13680*60000 = 820800000 words https://docs.google.com/document/d/1G3vvQkn4x_W71MKg0GmHVtzfd9m0y3_Ofcoew0v902Q/edit#heading=h.sxcem9l5k3ce
Training compute
The arithmetic performed to train the model, measured in floating-point operations. It is a measure of what the training run cost, not of how fast the finished model answers you.
- Training compute
- 3.9 × 10¹⁸ FLOP
- How it was established
- Operation counting
"We train each stage on 16 TPUv4s with batch size of 256 for 1M steps." That's for the 900M-param transformers If there's 256 passes in each batch, then using 6ND that's 900m * 256m * 6 = 1.3e18. sanity check: 16 tpu4s is 4.4e15 FLOP/s. 1.3e18 FLOP / 4.4e15 FLOP/s is 295 seconds. adjusting for utilization it would be ~1000 seconds or 15 minutes? probably too short, so 1.3e18 seems too low. upd there are 3 stages -> 1.3e18*3 = 3.9e+18 (Speculative due to reasoning above)
The training run
What it physically took to train: which chips, how many, for how long, and what that drew from the wall.
- Training hardware
- Google TPU v4
Availability
Whether you can obtain the model and run it on your own hardware, which is what decides if any of the graphics-card figures on this page apply.
- Weights
- Closed — provider access only
- Model access
- Unreleased
- Training code
- Unreleased
How it is classified
Labels the source dataset applies when tracking notable models, and how confident it is in the entry.
- Foundation model
- Yes
- Why it is tracked
- SOTA improvement
- Record confidence
- Speculative
- Citations
- 917
Compared to other systems without text supervision, AudioLM achieves the highest sWUGGY scores across both splits. Similarly, it also attains the highest score in the sBLIMP metric, improving by 8% relative over the previous state-of-the-art (CPC-BERT [59]).
Sources
Where this record came from and when it was last checked.
- Reference
- AudioLM: a Language Modeling Approach to Audio Generation
- Last updated
- 25 May 2026
What the numbers mean
Where it came from
AudioLM was published by Google Research, in United States of America, in July 2023. It comes out of industry.
It works in Audio, and is recorded as doing audio generation.
Its weights were never published, so it can only be reached through its provider. No graphics card changes that.
What went into building it
Producing it required around 3.9 × 10¹⁸ FLOP of arithmetic, on Google TPU v4, which is a statement about the training budget rather than about inference.
Around 135,000,000,000 tokens went into training it.
Its inclusion criterion is sOTA improvement.
Answers
AudioLM — common questions
Who created AudioLM?
AudioLM was published by Google Research, based in United States of America, categorised as industry.
When was AudioLM released?
AudioLM was published in July 2023. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.
What is AudioLM used for?
AudioLM works in Audio, and is recorded as handling audio generation. These are the areas it was designed around; they describe intent rather than a hard boundary.
How much compute was used to train AudioLM?
Around 3.9 × 10¹⁸ FLOP, on Google TPU v4. That measures what producing the model cost and says nothing about how quickly it answers once trained — inference speed comes from memory bandwidth, not from the training budget.
What GPU do I need to run AudioLM?
None. AudioLM is a closed model — its weights were never published, so it cannot be downloaded or run on your own hardware at any price. It is reachable only through its provider.
Is AudioLM open source?
No. AudioLM has not had its weights published, so it exists only as a service controlled by its owner.
How many parameters does AudioLM have?
AudioLM has 1.5B parameters. "We use identical decoder-only Transformers in all stages, with 12 layers, 16 attention heads, embedding dimension of 1024, feed-forward layer dimension of 4096 and dropout of 0.1, together with T5-style relative positional embeddings [38], resulting in a model parameter size of 0.3B per stage." Three stages (figure 2), and 300M per stage. Plus 600M parameters for w2v-BERT XL, so 1.5B total. That figure is the total, and it is what decides how much memory the model needs — roughly half a gigabyte per billion at the compression most people use.
The other direction
Looking at it from the other side?
This page starts from the model. If you already own a card and want to know everything it will run, start from the hardware instead.