# A few hours of audio teach speech models languages they never heard

A large preprint fine-tunes 24 open speech recognisers on 50 low-resource languages. Every model improves sharply, but the output is still a draft that needs human correction.

- Category: Vision & multimodal
- Readiness level: II (Benchmarked) on Science Underway's four-step scale, as proposed
- Published: 2026-11-14
- Editor: Stefan Perdal
- URL: https://scienceunderway.com/articles/a-few-hours-of-audio-teach-speech-models-languages-they-never-heard

## The study

Low Language Transfer Speech-to-Text (preprint, not peer reviewed). arXiv 2611.16260. https://arxiv.org/abs/2611.16260

## At a glance

- What it shows: Off-the-shelf speech-to-text models do badly on rare languages they were never trained on. After training on about an hour and a half of speech per language, every model tested got roughly nine in ten letters right.
- What it does not show: It does not show transcripts good enough to use without checking, since about a third of words are still wrong. It also does not show how the models do on recordings unlike the ones they trained on, beyond a small check in four languages.
- Why it matters: Most of the world's languages have very little recorded speech. This suggests a small, affordable recording effort can give such languages a usable first-draft transcriber, built on freely available models.

## What the study shows

The study tests speech recognition, software that turns spoken audio into text. It takes 28 open models that each handle many languages. It scores them on 50 low-resource languages: languages with little recorded and transcribed speech. These cover 11 writing systems, with 25 test clips per language.

Unchanged, most models did poorly. Many wrote the wrong alphabet entirely. The author then fine-tuned 24 of them. Fine-tuning means continuing to train a model on new examples. Each language had at most 1,000 clips, about 82 hours of audio across all 50 languages.

The main measure is character error rate (CER): the share of letters a model gets wrong. Every model improved, by between about 17 and 53 percentage points. The best main run, Meta's MMS 1B, fell from about 31% wrong to 11%. A larger model from Meta's omniASR family reached 9.5%.

Words are harder than letters. Even with an add-on that checks which words exist, the best word error rate was 34%. The author calls the output a draft to correct, not a finished transcript.

Several controls back the main claim. The same model designs trained from scratch, without prior training, ended 60 or more points worse. So the gain comes from what the models had already learned from other languages. On 12 languages that no model claims to know, every model started at 46% or worse and the best ended near 12%.

The study also reports costs. Fine-tuning made the median model worse on English, Spanish, French and German, by about 6 points. Lightweight adapters, which train only about 4% of the model, came within about a point of full training. On two of three models they also forgot less.

This is a preprint from a single author at Scicom, a Malaysian company. It has not been peer reviewed.

## Our reading

In our reading, the most useful finding is how little data it takes to get started. Twenty-five clips per language, about two hours across all 50, already cut one model's error rate by about a third. Gains slowed after that, but did not stop.

The study is careful about what the gain is made of. The from-scratch controls and the 12 unlisted languages make a convincing case that pretraining on other languages does the heavy lifting. A practical example: a public broadcaster with a few hours of transcribed regional-language audio could get a draft transcriber. Editors would then fix about one word in three, rather than typing from nothing.

The main caution is that test and training clips come from the same collections. The models may partly learn how a collection sounds, not just the language. A check on a separate public dataset in four languages kept most of the gain, but four languages is a small sample. On our readiness scale this is level II, Benchmarked: results on public test sets, but no released training code or fine-tuned weights that we could find.

The rest of the article (what it means, limits, what to watch) is free for members: https://scienceunderway.com/articles/a-few-hours-of-audio-teach-speech-models-languages-they-never-heard

## How this was made

An AI model read the study and wrote this article in the structure Science Underway uses. An editor reviewed it and approved it for publication. Check the original study before you rely on the details.

---
Source: Science Underway, https://scienceunderway.com/. Please link to the article when you cite it.
