v2.0.0

The first release of the rewritten converter: one service that turns HLS
(.m3u8) Internet radio streams into a plain, continuous HTTP audio stream a
radio can play.

Your radios need no changes. http://<pi>:8123/<percent-encoded-m3u8-url>
is byte-for-byte the URL format version 1 used, so anything already configured
keeps working.

Highlights

• One pipeline per station, shared by every radio. Ten radios on one station
  costs one decode, not ten. Measured on a Raspberry Pi 3: about 165% of one
  core for a 192 kbit/s station, unchanged whether one radio is listening or six.
• FFmpeg is not needed for some stations. MP3 and AAC-in-MPEG-TS are
  forwarded through a pure-Python demuxer, so those stations cost almost no CPU
  and lose no quality. radioconv probe <url> tells you which engine a station
  will use before you point a radio at it.
• A radio that connects mid-stream starts playing at once. HLS arrives in
  segments and the encoder finishes a segment faster than realtime, so a new
  listener used to wait for the next one. Time to first byte on a real station
  went from 8.08s to 0.02s, and each listener is now handed recent audio to
  cover the gap between segments.
• Encrypted streams work: EXT-X-KEY:AES-128, including mid-stream key
  rotation.
• It stays up. Every socket has a timeout, every buffer is bounded, every
  FFmpeg child is killed on the way out, and a wedged stream is restarted rather
  than left sitting there serving nothing.

Watching it

http://<pi>:8123/ turns any .m3u8 URL into the exact string to type into a
radio. http://<pi>:8123/dashboard shows what every station is doing right now -
listeners, CPU, memory, buffer depth, restarts and the last hour of traffic - and
every figure and column heading explains itself on hover. Silence is measured
from the listener's side, so a station that is simply idle between segments is
not reported as a fault.

The same numbers are available as JSON at /metrics.json and in Prometheus
format at /metrics, which is what a script or a home-automation rule should
read.

Installing

git clone https://github.com/clifarr/HLS-Radio-Converter
cd HLS-Radio-Converter
sudo ./packaging/install.sh

(main is this release. To pin the tag instead, add --branch v2.0.0 to the
clone, which leaves you on a detached HEAD - fine for installing, but
git pull then does nothing until you switch back to main.)

Needs Python 3.9+ and FFmpeg. On 64-bit ARM, x86_64 and Apple Silicon an FFmpeg
binary ships with the dependencies; on 32-bit Raspberry Pi the installer adds the
distro package. radioconv check reports which one is in use.

A PyInstaller spec is included for a machine with no Python at all - see
CONTRIBUTING.md.

Coming from version 1

• Radios need no changes.
• Settings were renamed: ffmpeg_location → ffmpeg_path, ip_address → host,
  buffer_size → chunk_size. bitrate and port are unchanged and the
  defaults match, so audio quality is the same out of the box.
• If your stations are MP3-in-TS you can now drop FFmpeg entirely.

Full instructions, troubleshooting and the hardware notes are in the
README.