Rhasspy offline voice assistant toolkit

FunkyBoT · October 7, 2020, 5:54pm

After updating Rhasspy to 2.5.6 (hassio in docker) it stopped to work.
Getting the following:
TrainingFailedException: file is not a database

synesthesiam · October 7, 2020, 6:07pm

If you’re using fuzzywuzzy, you may need to delete your existing intent_examples.json file and retrain. A fix for this should be coming soon.

FunkyBoT · October 7, 2020, 6:21pm

That’s the case. Done and working again. Thanks for your fast response.

Jiran · October 8, 2020, 7:58am

OK, will try that. Now I wait form my PS3 EYE camera and bought raspberry pi4 as I now run HASS on raspberry 3 (hassbian). Hassbian i snot supported anymore and I´m not able to upgrade to higher hass version than 0.107.5.
So I will move hass config to hass.io (which is something I was always aware) on rasspy4 and will use raspy3 for Rhasspy. Is raspberry pi3 enough for this?
After that I will install Rhaspy etc. and will start playing with voice commands.

hassdd2 · October 8, 2020, 11:54am

I have encountered to the error with train in VI profile. I edit the sentences, then click train. But error show in the bottom as picture.

For EN profile. It works very well. But for Vietnamese (VI) profile, the system encountered that problem after downloading missing file requirement and train.

synesthesiam · October 8, 2020, 2:00pm

Yes, Rhasspy has no trouble running on a Pi 3. You could also run it as a Hass.io add-on.

Sound good I don’t trust Google Translate with the default voice commands, so it’s always nice to have a native speaker volunteer to translate them.

synesthesiam · October 8, 2020, 2:37pm

OK, thanks. I have a fix for this problem that I’ll roll into the next update.

Before then, you can fix the problem by going into your Rhasspy profile and edit these two files:

profiles/vi/model/phones/optional_silence.txt
profiles/vi/model/phones/silence_phones.txt

All you need to do is make everything upper-case. So “sil” becomes “SIL”, “spn” becomes “SPN”, etc.

hassdd2 · October 8, 2020, 3:19pm

Many thanks for your guidance. It’s perfect. Now all things smooth. Thank you so much to create rhasspy.

synesthesiam · October 8, 2020, 4:09pm

You’re welcome! I’d like to improve the Vietnamese speech model at some point by re-training it with public speech data.

If you have a good microphone, I’m also looking for volunteers to create new text to speech voices

Jiran · October 8, 2020, 5:53pm

So you think raspy4b 8GB will be enough for both hass.io and complete rhasspy? Isn´t is better to run rhasspy in own docker or own separated HW? What is the peak HW spec. Rhasspy needs RAM or CPU?

Do you have info if ps3 camera as mic is attacheable to hass.io docker?

synesthesiam · October 8, 2020, 9:13pm

Probably more CPU depending on which wakeword system you’re using. But even then, it won’t be using much CPU or RAM, especially on a Pi 4.

I’ve successfully used it with Hass.io, yes. You just have to select the input device on the Rhasspy add-on configuration page.

hassdd2 · October 8, 2020, 11:18pm

Blockquote
You’re welcome! I’d like to improve the Vietnamese speech model at some point by re-training it with public speech data.
If you have a good microphone, I’m also looking for volunteers to create new text to speech voices

I’m excited to contribute to improve Vietnamese speech model. I’m newbie for Home Assistant and Rhasspy using.
So, I’m learning how to use, train Rhasspy, connect Rhasspy to control Home Assistant by intent, event, etc. I’m not sure I can immediately train Rhasspy Vietnamese speech model well, but I’m exicted to contribute.

I had 2 type of microphone.
First one (USB cable) I think it’s quality not so good

Second one (Bluetooth), this is motorcycle bluetooth intercom, the quality of record is very good, it can eliminate the noise, wind, etc. well. But I don’t know how to connect to Rhasspy.

I think discussing on this public post will trouble for other one who need to ask for other question. It’s better if we can work by private inbox or email, etc. How can I contact with you to work to train Rhasspy.

synesthesiam · October 11, 2020, 12:20am

Good news, @Jiran! I’ve created a Czech profile for Rhasspy. It will be included in version 2.5.7, which should be released in the next few days

Jiran · October 14, 2020, 3:26am

Great, waiting. I also found another Czech TTS http://epos.speech.cz/

synesthesiam · October 16, 2020, 2:03pm

2.5.7 is now available via Docker and Hass.io

Jiran · October 16, 2020, 6:56pm

Great, tomorrow will test IT

FunkyBoT · October 19, 2020, 11:45pm

Updated to 2.5.7.1 and it stopped to work again.
Running rhasspy under hassio/Docker/Ubuntu 18.04/Proxmox.
I have no clue what is going on.
It detects the hot word… Problem with Pocketsphinx again?

platup · October 20, 2020, 1:57pm

There is a .1 version? I was on 2.5.7 and rolled back to 2.5.6

jonaspaulo · October 20, 2020, 4:29pm

Same. Was on 2.5.5. Upgraded to 2.5.7 and the stop word ceased to function.

Rolled back to 2.5.5

FunkyBoT · October 20, 2020, 5:29pm

Sadly, my hassio snapshot is from only 4 days ago. I haven’t noticed that Rhasspy was broken. I just back it up with all addons updated… Will have to wait for a fix.