Home Assistant Voice Preview Edition: Integrated but not working

Configuration Voice Assistant
, ,
1

Hey there,

I am a newby to home assistant (but not to home automation - using shelly and raspberry pi for years).

I set up the Home Assistant Green with two Home Assistant Voice Preview Edition without the cloud soly in my local network.

All my devices are in the system I defined all the areas and floors.

The Voice PEs are also integrated and I can wake them successfully with “Hey Jarvis”. But anything I ask Jarvis afterwards is not being recognised. No matter if I ask for the time ("Sorry, I dont understand) or simply ask for switching on the lights (“the area is not defined” - it is and the lights are also defined in this area).

image
image1352×1151 151 KB

Any ideas?

Thanks for your help, guys.

Frank

3

Setting >> voice assistant >> 3 dots next to the
Voice assistant you use >> debug

Look at what the voice assistant recognizes your speech as. It likely is capturing incorrect speech.

HA remote is more accurate
Speech to phrase STT provider makes fewer mistakes

First steps is looking at debug to see what it interprets your speech as

4

Thanks @tmjpugh

I found those entries. I speak German and it looks like that it tried to hear English.
I am pretty sure that everything is set to German. Any ideas?

INFO:wyoming_faster_whisper.sherpa_handler:
INFO:wyoming_faster_whisper.sherpa_handler: Yeah.
INFO:wyoming_faster_whisper.sherpa_handler: Licht an embarrass.
INFO:wyoming_faster_whisper.sherpa_handler: Nicht Aus in the Roche.
INFO:wyoming_faster_whisper.sherpa_handler: What is this?
INFO:wyoming_faster_whisper.sherpa_handler: We feel worse is
INFO:wyoming_faster_whisper.sherpa_handler: Schalt das nichtimber wo aus.
INFO:wyoming_faster_whisper.sherpa_handler: Schalt do palp diegu bureau decketzwei aus.
INFO:wyoming_faster_whisper.sherpa_handler: Schalt das nicht aus
INFO:wyoming_faster_whisper.sherpa_handler: Bureau Licht zwei Auschalten
INFO:wyoming_faster_whisper.sherpa_handler: Schultz liicht in the wars.
INFO:wyoming_faster_whisper.sherpa_handler:
INFO:wyoming_faster_whisper.sherpa_handler: List an embure.
INFO:wyoming_faster_whisper.sherpa_handler:
INFO:wyoming_faster_whisper.sherpa_handler: I think

faster-whisper is set to German and at the moment my only option. Could I change this?

Thanks for your help
Frank

5

I can highly recommend using vosk for German. Faster Whisper gave me very bad STT results and was relative slow.

6

you have 2 wake word configured and it look like 2 different Voice Assistants set. Are both Voice Assistants set for german (de)?

I looked at the speech to phrase I recommended and it look like German translation is slightly incomplete. Point is, maybe the SpeechToText (STT) provider you have choosen does not have good language support for german. You may want to verify which support it best.

Personally after my own difficulty with STT I do think Speech to Phrase is long term best as it eliminates most incorrectly recognized phrases and reduces the “I did not understand that” replies. I would reach out to them and help them with getting the rest of german translation completed. I honestly believe it would not take much time and will give best long term experience.

In case you do not know, STT is responsible for taking the words you speak and translating them to text that may be used to determine command you desire. this is the most important part of Voice Assistant in my opinion.

7

You launched the add-on with default settings (auto) and are using Parakeet v3 (a multilingual model without explicit language specification). Change the engine to faster-whisper.

8

Thanks @Jowe1

Vosks seems to work. At least some commands can be executed. I guess I need to configer the hardware.

9

Thanks @tmjpugh

I’ll check and experiment with vosk and speech to phrase.

10

Thanks @mchk

As I came from there and was very sure to have everything set to German, I also tried the others. Getting more comfortable with Home Assistant and learning fast, I’ll give it another try.

Thanks for your help.