Skip to content

Device Text to Speech

Arie J. Godschalk edited this page Nov 15, 2026 · 2 revisions

Home / Devices

Text-to-Speech

Text-to-Speech device artwork

Device artwork from Google Services.

Make speech with Google Cloud voices. The app creates an audio file and supplies a temporary local URL; it does not play audio by itself.

Before you start

Complete cloud speech setup, save the JSON and choose a default voice in App settings. Add Text-to-Speech in Homey. Your player must accept the chosen audio format and reach Homey's local address on TCP port 47333.

Device settings

Setting Meaning and default
Speed % Speech rate, default 1; range 0.25–4. A value of 1 is normal.
Pitch Default 0; range −20 to 20 semitones.
Delay for each file to (finish) playing in ms Extra time added to the measured audio duration for the completion timer; default 1000 ms. A zero value currently falls back to 1000 ms.
Device Profile Audio effect profile, default No Profile. Choose the closest output device; see Google audio profiles.
Volume Gain (dB) Default 0; range −96 to 16. Start low; more than +10 dB generally brings little benefit.
Prefix Pause (ms) / Suffix Pause (ms) Default 0; range 0–60000. Prefix can make room for a speaker's startup sound. Test suffix pauses with your own text before relying on them.
Always use SSML Default off; wraps plain input in speak tags. SSML is Google's speech markup language.
Encoding MP3 by default; LINEAR16 produces WAV. Choose a format your player supports.

Card inputs override the corresponding settings when provided, but the implementation treats zero pitch or gain as a fallback to the device setting. To reliably use zero, also set that device setting to zero. Voice defaults to the app-wide voice unless selected on the card.

Audio, timing and privacy

Cloud audio URLs use plain HTTP on your local network. Anyone who can reach the URL can listen while it remains cached; don't expose the port to the internet or use sensitive announcements on untrusted networks. Homey keeps up to 100 recently used recordings in memory; old URLs stop working after eviction or app restart. Matching requests can reuse audio. The duration token is milliseconds and is an estimate based on the generated file, not a confirmation from the speaker. New speech on the same device replaces its pending completion timer.

“Get URL” conditions update the device URL and duration but do not trigger playback. “Start saying … and finish” conditions trigger the URL route and wait for the estimated duration plus delay. This is not player feedback. The Finished trigger's current listener checks an ID that the trigger does not supply, so don't depend on it firing. See the announcement example.

For a chime followed by speech, and for spacing out sentences, see Play a sound before speech. Independent requests are not automatically queued.

SSML example

Google accepts supported SSML, for example:

<speak>Hello.<break time="1s"/>Dinner is ready.</speak>

Send this as the Text argument. Use a valid single <speak> root and test with your chosen voice; unsupported markup can fail.

Flow cards

Open Text-to-Speech Flowcards · All cards · Examples · Troubleshooting

Clone this wiki locally