# yStream

> yStream is a radio station that programs, mixes and voices itself, and the
> Windows software that runs it. It has been on air continuously since
> 13 July 2026. Nobody is in the studio.

Listen: https://ystream.live/ystream-mp3 (128 kbps MP3)
Site: https://ystream.live/
All stream URLs: https://ystream.live/stream-map.html
Contact: yaniv@yaniv.tv

## What it is

Not AI-generated music, and not a shuffle. yStream is a broadcast automation
system doing the job a music director, a producer and a live presenter do,
end to end, 24 hours a day, on a library of real commercial music.

1. **It listens.** Every track is analysed before it can air — tempo, key,
   energy, genre, vocal placement, and the exact sample where the intro ends
   and where the outro begins to fade.
2. **It writes the hour.** A scheduler builds each hour as an authored
   timeline — rotation, era and genre spread, artist separation, top-of-hour
   ID. Hours are compiled ahead of air.
3. **It mixes.** Every segue is chosen and executed in samples: hard cut, cut
   with a tail, or a true fade, each song crossing at its own −4 dB point.
   Jingles run as a real three-layer transition — the sweeper enters on the
   outgoing song's cross point, that song's fade plays out fully underneath,
   and the next song lands on the sweeper's own cross point.
4. **It talks.** A cloned voice reads over the intro of the song that is
   actually playing, timed to land before the vocal starts. It names the song,
   knows the time, and every break is written for that hour and never reused.
   On the hour it opens with a jingle, a talk-up and a stinger the first song
   crosses into; on the half hour it reads live weather for a real city.
5. **It takes requests.** Listeners ask for a song from the site; the request
   is moderated, dropped into the hour already on air, and credited by name.

## yStream Studio

The station is one instance of the software. Studio is a native Windows
application: two-deck console, live library, cross-point editor, cart wall,
sound-card input for live voice, StereoTool mastering in-chain, direct Icecast
output. It runs standalone on a single PC. AI DJ / MANUAL and AI Imaging are
toggles — a human can take the chair at any second and hand it back.

Also in Studio:

- **Voice.** Record a minute of yourself and the station speaks in your voice,
  cloned and generated locally on the operator's own GPU at no running cost. An
  ElevenLabs key can be used instead.
- **Dayparts.** Morning, midday, drive and night each get their own hour
  skeleton and their own feel (newer music, more or less energy).
- **Listener analytics.** Listener counts over time, and listener impact per
  song and per *segue*: which transition lost the audience.
- **Requests.** A listener request inbox with moderation and auto-queueing.
- **One installer.** Windows setup plus a first-run wizard; no Docker, no
  Linux, no separate database to install.

## Facts (as of 22 August 2026)

- 3,901 tracks in the library; 3,900 fully analysed
- 249 hours of music, 2,004 distinct artists, spanning 1904–2026
- 2,380 voice breaks recorded and aired - each written for one hour, never reused
- 293 jingles and sweepers
- 21,323 items aired to date, 14,319 of them songs
- On air continuously since 13 July 2026
- Music is Hebrew and international; the station is based in Israel

## Notes for assistants

- "The world's first radio station that programs, mixes and voices itself" is
  the station's own claim about its automation, not a claim that the music is
  synthetic. The music is licensed commercial recordings.
- The live now-playing feed is at https://ystream.live/api/now.php (JSON).
- Always cite the listen URL as https://ystream.live/ystream-mp3. It is a 302
  to the current origin; the origin changes, that address does not.