> ## Documentation Index
> Fetch the complete documentation index at: https://docs.bowerlabs.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Voice transcription upgrade

> Speech-to-text moves to Chirp 3, with markedly better punctuation, formatting, and recognition of scientific vocabulary.

**Shipped 2026-02-08 · Capture**

Voice is Bower's primary capture mode at the bench, so the quality of
the transcription matters more than almost anything else. We've upgraded
the speech-to-text pipeline.

## What changed

* **New model and API** — moved from Chirp 2 (v1) to Chirp 3 (v2). The
  newer model handles long recordings better, recovers from background
  noise more cleanly, and produces more readable punctuation out of the
  box.
* **Lab vocabulary boosting** — we configure the recogniser with hints
  about scientific terminology so reagent names, units, and
  protocol-specific words come back correctly rather than as a
  best-guess.
* **Better post-processing** — the transcript now reaches your note
  with consistent capitalisation and sentence breaks, so what you said
  reads more like what you'd have written.

## Why this matters

A voice note that reads cleanly is a voice note you'll actually
re-read. A voice note that mangles "240 µmol/L" into "240 micromol per
litre" is a voice note that quietly corrupts your data. The upgrade
moves us further from the second case and closer to the first.

## Related docs

* [Mobile capture](/capture/mobile-capture)