Mihonarium/StreamToSpeaker

Stream To Speaker adds a virtual audio output to Windows and streams it over the network — to Sonos, IKEA SYMFONISK, KEF, Denon HEOS, MoOde and Volumio over UPnP / OpenHome, and to HomePod, AirPort Express and shairport-sync over AirPlay. Windows volume and the speaker's own buttons stay in sync. It also has experimental support for Sendspin (Music Assistant's streaming protocol): it can stream to Sendspin speakers and offer the PC's sound to Music Assistant as an input (optional, off by default).

Install

Directly or from the Microsoft Store.

⬇ Download and run it. Needs Windows 10 1809+ / Windows 11, x64 (not ARM64 or Windows Server).

Use

1. Set Stream To Speaker as the Windows audio output — or route a single app to it in the Volume Mixer.
2. Pick a speaker in the app.

Sonos speakers grouped in the Sonos app show up as one entry ("Living Room + Kitchen"); audio plays on the whole group and the volume slider moves the group volume. To stream to a single speaker, ungroup it in the Sonos app.

Sound lagging behind video → −25 / −100 ms. No sound → Resync (Ctrl+Shift+R). Speaker missing → ↻ Rescan; check it's on the same network as the PC.

Sendspin support is experimental and off by default: tick Enable Sendspin in the Sendspin and Music Assistant card to list Sendspin speakers and offer the Music Assistant input. Sendspin speakers play directly, without Music Assistant. One that asks to be paired shows or says a code (or has one printed on it): type it in when the app asks; it's remembered.

Music Assistant input (experimental): tick Offer this computer as a Music Assistant input and enable the Sendspin Source plugin in Music Assistant. The first time, open this computer in Music Assistant's player settings and run its setup; the app shows a 6-digit code to type in there. After that, select Music Assistant in the speaker list to send the PC's audio there; it gets audio only while that is selected. Music Assistant chooses the speaker: set Automatically play line-in on in this computer's player settings there, or play the input from Browse.

Closing the window minimises to the tray; quit from the tray menu.

Speakers that stream over UPnP fetch the PC's audio from a plain HTTP URL on the PC. By default any device on the same network can play whatever the PC plays from that URL, from the moment the app starts (before you pick a speaker) and whichever speaker you play on — AirPlay, HomePod, Sendspin and Music Assistant included — unless you turn streaming off. Advanced → Privacy mode serves that URL only to the speaker you're streaming to and the other speakers of its Sonos group (as grouped when you picked it): their addresses are allowed for as long as the session lasts, and switching speakers cuts off anything still listening from the old one. A refused device is named in a message, so you can tell a phone from a speaker; if a speaker stays silent (one grouped after you picked the group, say), pick it again or turn privacy mode off. AirPlay, Sendspin and Music Assistant playback doesn't use the URL, so privacy mode never affects it. Disabling streaming or forgetting the speaker closes the URL to every other device either way.

Web UI: to switch speakers or nudge latency from your phone, click Enable web UI in the Web UI card and open the link it shows (Copy link, or Open in browser on the PC). The link holds a secret key, so anyone with it can control the app on your network; Make a new link replaces it, and the old one stops working. Running headless or as a service, with no window to show the link, stream-to-speaker.exe --web-link prints it; run it as the Windows account the app runs under. Scripts send the same key with every API call — see TECHNICAL.md.

Once a day the app asks GitHub whether a newer release exists and, if so, shows a link to it. Nothing is downloaded or installed by itself. Turn the check off under Advanced if you'd rather not.

CLI flags

--headless                No GUI; run as a CLI service.
--no-tray                 GUI window only, no tray icon.
--web                     Enable the web UI + HTTP/JSON API, both behind a
                          secret link (shown in the window; printed once at
                          start with --headless). Off by default; when off,
                          only /stream.raw is served.
--web-link                Print the web UI's link and exit — for headless and
                          service installs. Run it as the account the app
                          runs under, with the app's --port / --advertise-ip.

--source <auto|driver|wasapi-loopback|sine>
                          Audio input. Default 'auto': kernel driver, falling
                          back to WASAPI loopback.

--player <name-or-ip>     Speaker to use: name substring or IPv4. Omit to pick
                          in the GUI (or the terminal picker in --headless).
--no-interactive          Skip the terminal picker even in a TTY.
--list-speakers           Print discovered speakers and exit.
--no-discovery            Skip discovery; only serve HTTP.

--port <N>                TCP port. Default 5901.
--bind <ip>               HTTP bind address. Default 0.0.0.0.
--advertise-ip <ip>       IP to give every speaker and to search from.
                          Default: per speaker, this PC's address on that
                          speaker's network; search on every interface.

--initial-buffer-ms <N>   Prebuffer hint sent to the speaker. Default 50.
--silence-pace-ms <N>     Pacing of silence packets. Default 10 (real-time);
                          higher drains the speaker's buffer between tracks.
--rate-fudge-ppm <N>      Clock-skew compensation. Default 0.
--latency-adjust-step-frames <N>
                          Frames added or dropped per packet while adjusting
                          latency. Default 4.
--no-silence-injection    Don't inject a low noise floor during silence.
--silence-packets-threshold <N>
                          Consecutive silent packets before quiescence.

--ssdp-interval <N>       Minutes between re-discoveries. Default 5.
--log-level <level>       error / warn / info / debug / trace. Default info.

Tuning the last group: TECHNICAL.md.

Verifying a download

gh attestation verify StreamToSpeakerSetup.exe -R Mihonarium/StreamToSpeaker

proves the file was built by this repository's CI from a specific commit. Offline: --bundle <file>.sigstore.json, shipped with each release.

The driver is certified by Microsoft.

From the next release on, the Microsoft-signed driver inside each release installer is checked against a rebuild from this source by the Driver reproducibility workflow before the installer is signed (the installers of v0.1.6, v0.1.9 and v0.1.10 were checked afterwards by the same workflow), and each installer build logs which driver package it bundled. (Pre-releases built with the test-signed driver, which needs testsigning mode, skip that check.)

Developers

Architecture, building from source, HTTP API: TECHNICAL.md. Driver signing pipeline: docs/driver-signing.md.

License

Source MPL-2.0. Released binaries: free to install and use, no redistribution or modification — LICENSE-BINARIES.md.

Driver scaffold derives from Microsoft's sysvad sample (MIT); UPnP/SSDP patterns from swyh-rs (MIT).