DatCaptainHorseandClaude Opus 5.5
d0b4a6ce76
The guest's audio stream ran about 3.5% short of realtime: 193 packets a second where 200 were due, with no network loss at all. The client plays at its own 48 kHz, so its buffer drained and underran every few seconds, idle or loaded. The graph was running 128-frame (2.67 ms) cycles. The 256 floor for VMs matched on `cpu.vm.name`, which PipeWire reads from DMI and nesbox does not provide, so it never applied, and neswire's 240-frame request was rounded down to a power of two. A busy guest missed ~20 of those cycles a second (pw-top ERR on neswire), and each missed cycle's audio never reached neswire while the graph clock moved on. - PipeWire is fixed at 240 frames, power-of-two rounding off: one cycle, one 5 ms Opus packet, twice the headroom of before. - neswire reads the graph position each cycle and fills exactly the frames a missed cycle skipped, as silence, so the stream stays at 48 kHz whatever the scheduler does. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
neswire
A PipeWire sink that Opus-encodes guest audio and sends it to
neshub.
Audio in the guest has no speakers to reach, so neswire registers itself as an output device and takes what anything plays into it. From the application's side it is an ordinary sink.
Running it
cargo run --release --bin neswire
Then select the neswire sink as the system output, or point an application at it.
| Flag | Env | Default | Description |
|---|---|---|---|
--ipc-path |
NESWIRE_IPC_PATH |
/tmp/nestri-audio.sock |
Where to send Opus packets |
--channels |
NESWIRE_CHANNELS |
2 |
2 stereo, 6 for 5.1, 8 for 7.1 |
--packet-duration-ms |
NESWIRE_PACKET_DURATION_MS |
5 |
Opus frame size in ms |
--bitrate-per-channel |
NESWIRE_BITRATE_PER_CHANNEL |
64 |
kbps per channel |
RUST_LOG takes a standard tracing filter, e.g. RUST_LOG=neswire=debug.
Sample rate is fixed at 48 kHz, which is what Opus wants and what every consumer device runs at anyway.
Testing without a hub
neswire's only output is a Unix datagram socket that neshub binds, so running
it on a desktop means it has nothing to talk to — it retries connect forever
and there is no way to see what it would have sent. hub-stub binds that
socket and reports what arrives:
# terminal 1
cargo run --bin hub-stub
# terminal 2
cargo run --bin neswire
It decodes the Opus rather than counting bytes, and that distinction matters more than it looks. Opus codes digital silence in about two bytes a packet, so a sink that is receiving nothing but zeros still produces a steady ~3 kbps and looks correct on every meter downstream. Peak amplitude is what tells a working sink from a silent one.
cargo run --bin hub-stub -- --channels 8 # match neswire's channel count
hub-stub mirrors ipc_listener::run_audio_listener in neshub: same bind,
same 0o666, same stream-type and codec checks. Where the two disagree,
neshub is right and the stub should be corrected.
Layout
src/
├── main.rs CLI, wiring, the sink→encoder→IPC chain
├── sink.rs the PipeWire sink itself
├── encoder.rs Opus encode, including the multichannel mappings
└── bin/
└── hub-stub.rs
Licence
Apache 2.0. See LICENSE.