Avalon: ASR for Human–AI Interaction
| Author | Jack McIntire, Finnian Brown (Aqua Voice) |
|---|---|
| Published | August 2025 |
| Format | Spec |
| Original | Read the original |
TL;DR — The model card for Avalon, Aqua Voice’s speech-recognition model tuned for people talking to computers and AI, especially about software and code.
A five-page model card from the San Francisco startup Aqua Voice. It describes Avalon, a speech-recognition model trained on a new large dataset of real and synthetic human–AI interactions, set against the background of large-scale pretraining approaches like wav2vec 2.0 and Whisper.
It reports benchmark results against other systems, including the company’s own technical-vocabulary benchmark, and claims the biggest gains in software and coding contexts.
What it is: a company-published model card: vendor-reported results, not peer-reviewed.
Where this came from
- A copy saved in October 2026.
- Published by Aqua Voice: https://app.aquavoice.com/research/avalon-model-card.pdf (linked from https://aquavoice.com/blog/introducing-avalon)
5 pages. A copy is archived locally against link rot; the header links the original source.