Speaker identification with Raven

Hey!

Raven is able to identify the speaker based on the provided templates (each model is speaker specific).

For those using Precise or Porcupine, it might be possible to repurpose it as a speaker identification step behind the wake word detection.

Just an idea.

:grin:

4 Likes

This is what I do since snips with custom wakewords.

In jeedom, we have the wakewordid and can filter actions depending on who is speaking. For example my son can ask to let its bedroom lamp on, but until a defined time, it can’t and tell him to ask his parents :rofl:

1 Like

This is not about Raven as a wakeword detection system but as a step behind a universal wakeword system like Porcupine or Precise. Anyone can speak the wakeword but Raven can then match the detected wakeword WAV buffer to a specific template (pre enrolled speaker).

1 Like

@KiboOst Interesting. How do you do that? Do you run multiple instances of the snips wake word detector simultaneously? One for each person? If so, how did you make that work?

Nothing special just three wakeword in the toml.

I don’t use it with rhasspy though, still have my snips full setup running.

But Raven do the same, I have three custom in Raven and when intent reach my Jeedom scénario I have the wakeword id with all other tags.

Here is a simple way to check it in Jeedom

Hi, I am very interested in this. I am running Rhasspy with a satellite setup on RPi4b+/3b+ in docker.
How would I integrate Raven as a speaker identification stage in my setup?
Any suggestions?

Best,
Ck