Mimic 3 Text to Speech Launch!

Hi everyone! Today is the day that Mimic 3 is officially launched :partying_face:

The official docs talk about all the different ways you can try it for yourself :slight_smile:

I will be making an update to Rhasspy soon to add Mimic 3 as a TTS option, which I would recommend over Larynx now. Especially for the Pi 3 and 4, I’d say this is a big win for open source offline TTS!

Let me know what you think, and have a good day.

15 Likes

This is what I think: Awesome :smiley: :rofl: :+1:

2 Likes

This would be a great addition to Rhasspy!

2 Likes

Glad to hear about plans for an update for Rhasspy :slightly_smiling_face:.

I recently also did some testing with Mimic 3 as a drop-in replacement for MaryTTS and can now confirm this method to work flawlessly.

May I kindly ask for plans to integrate the most recent version of the “Thorsten” dataset to Mimic 3 as well? ( Thorsten-22.05-neutral)

3 Likes

Definitely :slight_smile:

1 Like

And don’t forget harvard-glow_tts please. :slightly_smiling_face:

1 Like

:astonished: :sunglasses: :astonished: :scream: :sunglasses: Seems I didn’t get deep enough into all the details…?
Afai now found out, the “thorsten_low” voice is already based on the (preliminary) version now named 22.05-neutral. So most likely atm there’s no specific action required.

1 Like

Glad to hear it’s coming to Rhasspy! I will probably go through some efforts in getting it added in Home Intent for 64bit pi os’s. The bit of experimenting I’ve done has been incredibly promising, the realtime factor is really good. Thanks for putting it out there!

1 Like

I installled mimic3 as a docker image:
(Before I had google wavenet)
https://mycroft-ai.gitbook.io/docs/mycroft-technologies/mimic-tts/mimic-3

After that, I changed in my rhasbian Server (I have Server / Satelllites):
preferences → Text to Speech to MaryTTS
After save and restart, I changed
URL to http://10.2.254.251:59125/process (Thats the IP of the docker home)
After save and restart, I refreshed available voices
Now I could pick up my language/voice.

This works! :grinning:
Great !! Thank You

Problems with german language: :frowning_face:
“.” vs. “,” :
11.00 → Elf tausend (eleven thousand)
11,00 → Elf komma Null (eleven point zero)
(So I add a replace command in Home Assistant to temperature values)

Time:
11:30 → elf dreißig (eleven thirty, wrong)
11:30h → elf Uhr dreißig (right)

Beginning and End of sentences:
If rhasbian speaks a sentence, there are missing words in the beginning and at the end.
With home assistant there is a strange sound at the end, if you finish the sentence with a point (.) No words missing

1 Like

I see it supports the use of CUDA acceleration…how difficult would it be to support Google Coral?

It depends on the kind of networks that Coral supports. Last I checked, they don’t support transformer networks :frowning: