Voice assistants
Wake words run on the device. Audio goes only where your Assist pipeline sends it.
Two assistants
Section titled “Two assistants”Assistant 1 and Assistant 2 each have their own wake word, Assist pipeline, sensitivity, sounds and timing. Use one for Home Assistant’s own assistant and the other for, say, a cloud LLM. A slot with no wake word is off.
Set it up
Section titled “Set it up”-
In Home Assistant, open the device’s Assistant 1 device.
-
Pick an Assistant (Assist pipeline) and a Wake word. Custom wake words added to Home Assistant are listed too. The EchoLocal integration adds a wake word library for managing them.
-
The device downloads the wake word and starts listening.
On an Echo Show you can also pick the wake word under Settings → Voice assistants. That needs Allow the device to perform Home Assistant actions turned on for the device in the ESPHome integration; without it the row is dimmed.
Starting without the wake word
Section titled “Starting without the wake word”- Echo Dot: tap the action button for assistant 1, hold it for assistant 2.
- Echo Show: Talk now on the assistant’s settings page.
- From Home Assistant: the Wake button on the assistant’s device (
wake_assistant_1,wake_assistant_2).
On the screen
Section titled “On the screen”On the Echo Shows, Settings → Voice assistants has a page per assistant:
-
Talk now
-
Wake word
-
Wake word sensitivity: higher wakes more easily, with more false wakes. In Home Assistant the same setting is a threshold (0.5–0.99, default 0.85) where lower wakes more easily
-
Wake sound: a chime, a custom tone, or none
-
Response playback: Stream starts sooner and can stutter on a slow network; Download first waits for the whole answer
-
Visuals: a visual for listening, thinking and replying, in the panel or full screen
Timing, follow-up and recordings
Section titled “Timing, follow-up and recordings”These are only in Home Assistant, on each assistant’s device:
| Setting | What it does | Default |
|---|---|---|
| Follow-up time | Keep listening after a reply, so you can ask again without the wake word (0–30 s) | 0 |
| Max listening time | Stop listening after this long (5–60 s) | 15 s |
| Max thinking time | Give up waiting for Home Assistant after this long (5–300 s) | 90 s |
| Reply buffer | How much of a streamed reply to collect before playing (0–3000 ms) | 650 ms |
| Recordings kept | Keep the audio of the last N requests (0–10), for the card’s Activity tab | 0 |
The Last wake word, Last heard and Last reply sensors show the most recent turn.
On the Echo Dot, the ring shows each stage with its own effect: Wake word effect, Thinking effect and Replying effect, per assistant. See Echo Dot: buttons and light ring.
Say “stop”, no wake word needed, tap the screen on an Echo Show, or tap the action button on an Echo Dot. It stops, in order: a ringing alarm or timer, the assistant, an announcement, then pauses music.
Stop word sensitivity in Home Assistant (0.5–1, default 0.7) sets how sure it has to be; 1 turns it off.
Voice triggers
Section titled “Voice triggers”Two kinds of things you can say, each with its own switches under Settings → Voice triggers on an Echo Show and on the voice triggers device in Home Assistant. Everything is on to start with.
Actions are said to an assistant, after the wake word. Assist hands them to the device:
- “Call the kitchen”, “video call the kitchen”: an intercom call
- “Broadcast”, “broadcast dinner is ready”: a broadcast
Actions need the EchoLocal integration. When it’s missing or out of date, the Voice triggers page says so, with a code to scan that opens it in HACS.
Commands are heard on the device itself, without a wake word (Echo Show):
All actions and All commands turn a whole kind off; each one has its own switch too.
Timers
Section titled “Timers”“Set a timer for ten minutes” starts a timer, with a countdown on an Echo Show. See Messages, timers and alarms.