Make OVOS Answer Only Your Voice¶

In a nutshell

Anyone in the room can say the wake word. A speaker verifier checks who said it. You record a few clips of each person in your household once. After that, when someone else says the wake word, the assistant stays silent. The voice samples stay on your device as numbers, never as audio.

The speaker verifier is the plugin ovos-ww-verifier-plugin-speaker. It runs after the wake-word engine fires and before the assistant starts to listen. It compares the voice in the wake-word audio with the voices you enrolled. If no enrolled voice matches, the listener drops the activation. See Wake-word Verifiers for how verifiers fit into the listener. The listener can run more than one verifier: the Silero VAD verifier ovos-ww-verifier-silero, from ovos-vad-plugin-silero, rejects wake words with no speech in them. The blog post Only you can order the cheeseburgers: Speaker verification comes to OVOS shows the verifier in use.

Alpha channel only

Only the prerelease channel of ovos-dinkum-listener loads verifiers. Install the listener with pip install --pre ovos-dinkum-listener.

Before you start¶

A working wake word. See Change your wake word.
Three or more recordings of each person, 16 kHz mono WAV. Ordinary sentences work. The wake word itself is not necessary. Five to thirty seconds of speech per person in total gives a better profile.

Set it up¶

Install the plugin into the environment that OVOS runs in:
```
pip install --pre ovos-ww-verifier-plugin-speaker
```

Enroll each person. Give a name, then the clips:

ovos-speaker-enroll Alice alice1.wav alice2.wav alice3.wav
ovos-speaker-enroll Bob bob1.wav bob2.wav bob3.wav

The command prints the stored profiles:

Enrolled 'Alice' from 3 clip(s).
Profile dim=256  norm=1.000000
Profiles saved to: /home/<user>/.local/share/ovos_speaker_verifier/profiles.json
Enrolled speakers: ['Alice']

The first run downloads the speaker model. The profiles file holds one vector of 256 numbers per person and no audio.

Add the verifier to ~/.config/mycroft/mycroft.conf:

{
  "listener": {
    "ww_verifiers": {
      "ovos-ww-verifier-speaker": {
        "threshold": 0.45,
        "fail_open": false
      }
    }
  }
}

Restart OVOS:
How to restart your assistant

OVOS is not one program — it's several separate services (the listener, ovos-core, ovos-audio, the messagebus, and more) that need to pick up a config change together. How you restart them depends on how OVOS was installed:
- systemd install (the ovos-installer or a RaspOVOS image): restart the whole stack with one command —
```
systemctl --user restart ovos.service
```
  ovos.service is a --user target unit that the other services (ovos-messagebus.service, ovos-listener.service, ovos-audio.service, ovos-core.service, ovos-phal.service, and — if installed — ovos-gui.service) declare as PartOf=, so restarting it cascades to all of them. Restarting a single service instead (e.g. only the listener after a wake-word change) works the same way: systemctl --user restart ovos-listener.service.
- Docker/container install: restart the equivalent service(s) in your compose file, e.g.
```
docker compose restart <service-name>
```
  (or docker compose restart with no name to restart everything).
- Manual/development run: stop the running ovos-core process (Ctrl-C) and start it again.
Say the wake word. The listener log has one INFO line for each decision, with the name of the best match, the score and the threshold:
```
speaker verifier accepted Alice (score 0.738, threshold 0.45)
speaker verifier rejected unknown speaker (best match Alice score 0.116, threshold 0.45)
```
These two lines come from a test with three synthetic voices and Alice enrolled. Alice's "Hey Jarvis, what time is it?" scored 0.738. The two other voices scored 0.116 and 0.170.

What happens to a rejected wake word¶

A rejected speaker removes the wake word event itself. The bus gets no recognizer_loop:wakeword and no recognizer_loop:record_begin, so no skill sees the activation.

Speaker	Profiles enrolled	`fail_open`	Result
An enrolled person	yes	any	The assistant listens
Anyone else	yes	any	The assistant stays silent
Anyone	none	`true`	The assistant listens
Anyone	none	`false`	The assistant stays silent

Installing the plugin turns it on

The listener runs every installed verifier unless its config sets "enabled": false. With fail_open at its default of true, the verifier accepts everyone until you enroll the first person. Enroll first, then set fail_open to false.

Tune the threshold¶

The threshold is a cosine similarity between 0 and 1. A higher value rejects more guests and also more of your own wake words. A lower value does the opposite. Lower it for a distant or noisy microphone.

The default of 0.45 is for the default model, wespeaker-resnet34. Each model has its own score scale. If you change model, find the threshold for that model: compare the scores in the log for your own voice and for a guest, and choose a value between them. Set a different value for one person with per_profile_thresholds, for example {"Bob": 0.5}.

Configuration keys¶

Key	Default	What it does
`model`	`wespeaker-resnet34`	Speaker model: a `speakeronnx` alias or a path to an `.onnx` file
`threshold`	`0.45`	Lowest similarity that counts as a match
`fail_open`	`true`	Accept every wake word when no profile is enrolled
`profiles_path`	`~/.local/share/ovos_speaker_verifier/profiles.json`	Where the profiles are stored
`per_profile_thresholds`	`{}`	A threshold for each name
`enabled`	`true`	Set `false` to keep the plugin installed but inactive

The plugin README lists the other speaker models and the audio format keys.

Limits¶

The verifier checks a voice, not a password. A recording of an enrolled person can pass.
Enroll clips recorded on the microphone that OVOS uses, so the profile matches what the verifier hears.
The verifier sees only the wake-word audio. A short wake word gives less speech to compare.