# `TuningFork.Sink.Tcp`
[🔗](https://github.com/jaman/tuning_fork/blob/v0.1.11/tuning_fork/lib/tuning_fork/sink/tcp.ex#L1)

Raw signed 16-bit little-endian PCM over a TCP connection, paced to real time, for any
player that can take it from a socket — `ffplay`, `sox`, `pacat --raw` — on the far
end of an ssh tunnel or across a LAN.

    TuningFork.Stage.start_link(sink: TuningFork.Sink.Tcp, sink_opts: [host: "127.0.0.1", port: 5000])

The connection is made on the first write and remade when it drops, trying again every
`:retry_ms`; until a player listens, samples are dropped at real time. A chunk is sent
and then the sink sleeps until the audio before it has nearly played out, keeping
`:lead_ms` of audio in flight, so a player that reads as fast as it can never runs
seconds ahead. With `:warm_up_ms` above zero, the first `:prime_frames` go at once, so
the player opens its device, and for that long after them the audio is pulled at real
time but not sent — for a player that queues whatever arrives while it opens its
device and keeps it as lag, such as ffplay. `TuningFork.Listener` needs none of it:
it drops such a backlog itself.

## Options

  * `:host` — the player's host. Default `"127.0.0.1"`
  * `:port` — the player's port. Required
  * `:lead_ms` — audio kept in flight ahead of real time. Default `60`
  * `:warm_up_ms` — how long after the first frames the audio is dropped while the
    player opens its device; `0` sends everything. Default `0`
  * `:prime_frames` — how many frames go at once before the warm-up. Default `1024`
  * `:retry_ms` — how long to wait after a failed connection before trying again.
    Default `2_000`
  * `:rate`, `:channels` — the PCM format, as the stage passes them

The far end is `TuningFork.Listener` (`mix tuning_fork.listen --port N`), or any
program that reads raw PCM from a socket — ffplay with `-f s16le -ar 44100 -ch_layout
stereo -i "tcp://…?listen"`, `sox -t raw`.

---

*Consult [api-reference.md](api-reference.md) for complete listing*