Uploading Audio

Voice notes and sound recordings pushed into the anarchive — the upload twin of Downloading Audio, and Phase A of Upload Media. Audio goes first because it fits through the easiest door: a Telegram voice note is an opus file well under the Bot API's 20 MB download limit, so the whole channel works with no public endpoint at all.

## Voice First

The gesture is Graham's: think of something worth keeping and talk for two or three minutes — on a walk, at a table, birds and trains welcome. Send it to the bot with a caption. The caption is the intent: a title, a target, an instruction.

Two bots now serve the gesture. Max's hosted bot — webhook, cloud — forwards what it receives to the Pi5's public endpoint; the Pi5's own bot long-polls from behind the router. Either way the audio lands in the same inbox and whisper transcribes it on the spot — no iCloud relay, no iMac.

## The Pipeline

1. **Receive** — voice note (`.ogg` opus) or audio file, plus caption, into the inbox 2. **Stub** — a stub page is created at the final URL before transcription begins; the link travels back to Telegram at once, and the page says check back and refresh 3. **Transcribe** — whisper on the Pi5; the Mac picks up long recordings when awake 4. **Summarise** — qwen3 on the Pi5 writes the opening paragraph and a title suggestion; the full transcript keeps its place on the page 5. **Page** — summary, transcript, an audio item, and the original in `# Assets`; the opus kept as sent, plus a web-friendly copy 6. **Route** — private to the Pi5 wiki by default; `public` in the caption sends it to the anarchive and the pin queue

## Captions as Instructions

The caption rides the audio as plain language. Absent, the page is private, titled from the transcript. Present, it steers:

public — a Field Recording from Finsbury Park

A first word of `public` routes to the anarchive; a phrase becomes the title; anything else becomes guidance for the summary. No syntax to memorise — the model reads it.

## See - Upload Media — the doors, ingress and authentication this channel uses - Spoken Deeds, Sound of Law and Contributing Sound — the first uses - Uploading Video — the heavier sibling - Downloading Audio — the pull twin, and the podcast anarchive - Media Plan — the wider channel map

# Assets

uploading-audio