Skip to content

Recording

astar-server can keep a recording of every transmission that passes through the node: each time someone keys up, on any link, the node writes what they said to its own audio file. Later you can play those files in a browser, download them one at a time or all together, or feed them to unamblify.

Recording is off until you turn it on.

Turning it on

Add a [recording] section to node.toml. Here it is with every key at its default, so enabled = true is the only line you have to change:

[recording]
enabled = true                       # the default is false
dir = "/var/lib/astar/recordings"    # where the recordings go
max_bytes = "2 GB"                   # the most the folder may hold
max_age_days = 30                    # how long a recording is kept
frames = true                        # also keep the codec frames, for unamblify
Key Default What it means
enabled false Record every transmission.
dir "/var/lib/astar/recordings" The recordings folder. It must be an absolute path, and the node creates it if it is missing.
max_bytes "2 GB" The most the recordings may take up. At least "100 MB". KB, MB and GB count in thousands ("2 GB" is 2,000,000,000 bytes); KiB, MiB and GiB count in 1024s; a bare number is bytes.
max_age_days 30 How many days a recording is kept, from 1 to 3650.
frames true Also keep the codec frames of M17 transmissions, laid out for unamblify (see below).

The node checks these as it loads node.toml and refuses to start with a message naming the bad key, for example a relative dir or a max_bytes under 100 MB.

The settings page has the same rows on its Recordings card: Record every transmission, Folder, Keep at most, Keep for days, and, under the card's advanced rows, Save codec frames for unamblify. Like every key in this section, a change there takes effect at the next restart.

What gets recorded

The node records one file per transmission. A transmission, which hams call an over, starts when audio begins arriving and ends when it stops. Two small rules decide where one over ends and the next begins:

  • A gap shorter than half a second is part of the over. Someone pausing for breath, or a link dropping a few packets, does not split a sentence into two files. The gap is filled with silence.
  • A gap of half a second or more ends it. The file is then closed with one second of silence on the end, so it doesn't stop dead on the last syllable.

Each file is mono FLAC, a lossless audio format that browsers and audio editors play directly, at the conference's own sample rate. Files go into one folder per day, named after the time the over started (in UTC), the network and who was talking:

2026-10-27/20261027-010004-allstar-2001.flac
2026-10-27/20261027-010231-m17-W1ABC.flac
2026-10-27/20261027-011500-node-unknown.flac

Who was talking comes from the node's own events. For AllStar it is the node number of the link that keyed up within the last two seconds. For M17 it is the callsign the M17 monitor heard. When the node can't tell, the network reads node and the talker unknown. When two stations talk over each other, the file is named for the first one, and the list on the Recordings page shows both.

A node in parrot mode records nothing: there is no conference to listen to.

What recordings contain

The recorder listens to the conference the way one more listener would. It is a member of the conference that can only hear, so it can never put anything on the air. What it hears is the whole mix: every AllStar link, the M17 monitor, and a Discord bridge if one is running. Whatever someone says in the Discord voice channel ends up in the recordings too.

Keeping those recordings, and who you share them with, is your decision as the node's operator. If the people on your net, or in your Discord channel, would expect to be told, tell them.

Keeping the disk in check

The node keeps the folder within the two limits you set:

  • Any recording older than max_age_days is deleted.
  • Then it adds up the recordings and deletes the oldest first until the total is under max_bytes.

It checks both limits when the node starts, after each over is written, and once an hour, so a node that stays quiet for weeks still lets old recordings age out. A check never runs in the middle of an over: one that falls due then waits for the over to end.

It only ever counts and deletes files it wrote itself. It keeps a list of them, index.jsonl, in the folder, and the codec-frame files are listed in the manifest.jsonl files under captured/. Anything else you put in the folder is never counted and never deleted.

Separately, the node keeps at least 500 MB free on the disk under the folder, whatever max_bytes says. With less than that free, it doesn't start a new file, and a long over that is already being written is cut short (what was written is kept). The Recordings page shows a red banner while this lasts: "Under 500 MB free on the recordings disk — nothing new is recorded until there is room again." Recording picks up again by itself once there is room.

The Recordings page

Open Recordings → at the top of the node page, or go straight to /admin/recordings. It is admin-only, like the rest of /admin: a listener who is signed in is refused, and on the control port the node's own loopback is the console admin as usual (see Signing in). The /admin prefix exists only on the public listener; the control port serves the same pages without it, so there the page is at /recordings.

The page lists every over, newest first, with its time, length, talker, network and size. Each row has three buttons:

  • Play plays the recording in the browser. You can drag to any point in it, because the node sends just the part the player asks for. The button turns into Pause while it plays.
  • Download saves the .flac file.
  • Delete asks "Delete the recording from … ? This cannot be undone." and removes the file once you confirm.

The badge at the top says what the recorder is doing: recording an over, idle, waiting for a conference, or stopped for lack of disk. If recording is off in node.toml, the page says so and links to the Recordings card in Settings.

Downloading everything

Download all, at the top of the page, downloads every recording in one file: a .tar archive from /admin/recordings.tar. The node writes the archive as it sends it, one file at a time, so a big folder needs no extra disk space and recording and pruning carry on meanwhile. A recording deleted while the download is running is left out of the archive.

Downloads and Play use the same per-client stream slots as the live audio and event streams: on the public listener one address may hold 4 at once, and everyone behind one proxy or NAT address shares those 4.

Add ?since= and an RFC 3339 time to fetch only what started at or after it, for example everything from 26 October on:

/admin/recordings.tar?since=2026-10-26T00:00:00Z

From the node itself, curl against the loopback control port is the easiest way, since the loopback is the console admin and needs no sign-in:

curl -o recordings.tar 'http://127.0.0.1:8730/recordings.tar?since=2026-10-26T00:00:00Z'

From your own computer, open an SSH tunnel first (ssh -N -L 8730:127.0.0.1:8730 <user>@<node-host>) and run the same command there. A since the node can't read is refused with "since must be an RFC 3339 time".

Inside, the archive looks like this:

recordings/index.jsonl
recordings/2026-10-27/20261027-010004-allstar-2001.flac
captured/codec2-3200/manifest.jsonl
captured/codec2-3200/astar-air/20261027-010231-m17-W1ABC-rx.ambe
captured/codec2-3200/astar-air/20261027-010231-m17-W1ABC-rx.flac

recordings/ holds the audio and its list. captured/ holds the codec frames, and appears only with frames = true and some M17 traffic.

Feeding a download to unamblify

Digital networks send speech as codec frames, small packets of compressed speech that a vocoder decodes back into sound. M17 uses Codec 2 at 3200 bit/s. With frames = true, the node keeps each M17 over's frames exactly as they arrived, before decoding, beside the decoded audio, in the layout unamblify reads its own captures from:

  • captured/<mode>/astar-air/<key>.ambe is the raw frames, back to back, with no header.
  • captured/<mode>/astar-air/<key>.flac is the same over decoded, 8 kHz mono.
  • captured/<mode>/manifest.jsonl has one line per over.

To use a download with unamblify:

  1. Unpack it into an empty folder: mkdir air && tar -xf recordings.tar -C air.
  2. Point unamblify's data root at the folder that holds captured/, either with --data-root air or by setting UNAMBLIFY_DATA.
  3. Check the files: unamblify verify --data-root air --mode codec2-3200.

unamblify reads the dstar, ysf-dmr and codec2-3200 folders. The nxdn and dmr72 folders that the macOS app can write are kept, but unamblify skips them.

For a debugging log of the raw packets rather than the audio, see Packet log.

These are off-air recordings, so there is no clean original of what was said to pair them with. They are for listening tests and for re-decoding, not for training pairs.