File size: 2,829 Bytes
5f89c1a 8671f98 5f89c1a 8671f98 5f89c1a 8671f98 5f89c1a 8671f98 ace1db2 8671f98 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 | ---
title: Number of audio-to-audio models on the Hub
emoji: 🎛️
colorFrom: yellow
colorTo: gray
sdk: static
app_file: index.html
pinned: false
short_description: Monthly audio-to-audio models on the Hub, with landmark rele
---
# Number of audio-to-audio models on the Hub
Every public model repo currently tagged `audio-to-audio` on the Hugging Face Hub,
counted by the month it was created, with landmark releases pinned along the
timeline as their lab's logo.
Hover a column for that month's figures, or click it to keep the box open.
Hover a logo for the model; click it to open the repo on the Hub. Hovering a
column gathers that month's logos underneath it.
Nothing runs on the Hub: this is a static Space, so no hardware is assigned and
no container is started. The browser renders a page built ahead of time, and the
figures sit in it as two `<script type="application/json">` blocks.
## Files
| File | Role |
|---|---|
| `index.html`, `style.css` | the page, pre-rendered |
| `task.json` | which tag this Space charts, and its title |
| `data.json` | monthly counts: `[month, new, releases with 25+ likes]` |
| `releases.json` | the landmark releases and their lab logos |
| `fetch_data.py` | re-reads the counts from the public Hub API |
| `select_releases.py` | picks the landmarks and resolves their logos |
| `build_static.py` | writes `index.html` + `style.css` |
| `refresh.py` | does all three, then pushes what changed |
## Staying current
A single scheduled Hugging Face Job refreshes every audio-task Space in this
family, this one included. It pulls each Space, re-reads the Hub API, rebuilds,
and pushes back only files whose bytes changed — so a quiet day leaves no commit.
```bash
python3 refresh.py # or run it by hand
```
## How landmarks are chosen
Ranked by likes, which is the only engagement signal the listing API returns,
then filtered twice: repos that repackage another team's weights (GGUF, ONNX,
MLX, quantisations) are dropped as formats rather than releases, and one owner
may hold at most four slots so a single lab cannot crowd out the field. A
`pins.json`, where present, is a hand-curated list merged in ahead of those.
## Caveats
**This is a survivorship count.** The Hub API returns only models that exist
*today* and carry the tag *now*. Repos deleted since, or created under another
tag and retagged later, are invisible, so early months are understated.
**The window starts at 2022-04** — it 2022-03 is the Hub's createdAt backfill, not activity: 75-100% of it lands on 2022-03-02 across hundreds of unrelated owners, so those repos predate the field. Cumulative figures in the hover box still count from the true start of the series.
**Landmark figures are all-time**, so older releases have had longer to
accumulate downloads and likes than recent ones.
|