aiodisklavier

Async Python client for the Yamaha Disklavier ENSPIRE local HTTP API.

Talks to the piano directly over your own network. Verified against firmware 5.24.00 on a Disklavier ENSPIRE PRO grand.

Install

pip install aiodisklavier

Use

import aiohttp
from aiodisklavier import Disklavier, SongGroup

async with aiohttp.ClientSession() as session:
    piano = Disklavier("192.168.1.50", session)

    info = await piano.async_get_current_info()
    print(info.song_title, info.playback_status, info.position_seconds)

    # Fuzzy title search runs on the piano itself.
    await piano.async_play_search("Clair de lune")

    # Play one song and stop, rather than continuing through the library.
    await piano.async_play_song(24, SongGroup.DOWNLOADED_SONGS, single=True)

Finding the piano is a plain SSDP M-SEARCH for urn:schemas-upnp-org:device:Disklavier:1; the library exposes that device type as UPNP_DEVICE_TYPE.

What it covers

Firmware behaviours worth knowing

These are properties of the piano, not of this library, and each is easy to get wrong. The full reasoning, with provenance for every claim, is in docs/enspire-api.md.

• There is no stop state. stop leaves playback_status reading pause at position zero. Use CurrentInfo.is_stopped rather than looking for a stop constant.
• Waking takes about twelve seconds, during which power_status reads wakeup and the piano ignores commands. The HTTP API answers normally while asleep, so reachability tells you nothing about power state.
• Empty libraries are an error, not an empty list — HTTP 200 carrying {"status": "error", "error_info": "no song"}. The browse methods translate that envelope back into the empty list it denotes, so callers just see [].
• State reads can come back truncated while a song is playing, because the daemon rewrites those files in place. Reads retry automatically. Payloads may also carry a trailing \n\0, which is stripped rather than retried.
• State lags a command. Reading current_info straight after a load_song or reselect returns the previous song. Allow a short settle before trusting a post-command read.
• Radio's interaction with transport commands is not established. There is reason to think playback behaves differently while a radio channel is playing, but it has not been exercised on hardware — treat transport during radio as unknown.
• async_play_test_chord makes a sound — a C major triad for one second. It goes to the MIDI daemon rather than the sequencer, so it will not disturb a loaded song.

Two APIs, one preferred

The piano exposes a versioned open API at /api/1.0/<command> and an internal, unversioned set of endpoints under /ctrl/ that its own web UI drives. This library uses the open API wherever possible and drops to /ctrl/ only for what the open API cannot do: seeking, repeat and shuffle, the extended state block, reindexing, and the test chord. Both funnel XML into the same Unix socket inside the piano.

The open API takes some finding: nothing the piano normally serves links to it, and neither the phone app nor the piano's own web UI calls it. The one client-side trail is /ctrl/api_test.html, a test harness Yamaha ships on the device — that is where the /api/1.0/ form is visible. /api/api.php?_com=<command> is the same surface by another name, verified equivalent down to the error codes.

Development

python3 -m venv .venv
.venv/bin/pip install -e ".[test]"
.venv/bin/python -m pytest

Tests run against a real aiohttp test server that imitates the piano, so no hardware is needed and the suite does not depend on any mocking library's grip on aiohttp internals. Several tests encode behaviour found only on real hardware — those are commented as such, because they look arbitrary otherwise.

Licence

MIT