Credit: @yannikw

This article will show you how to change the configuration on the Herika server to make her speak German.
This should work for any other language in a similar way.

Services covered in this article (in the order of appearance in the configuration):
⦁ChatGPT
⦁TTS: Azure & MIMIC3
⦁STT: whisper

OpenAI / ChatGPT
The only “language setting” you can make for ChatGPT is by providing a PROMPT_HEAD as well as HERIKA_PERS in the desired language.

⦁My PROMPT_HEAD was a direct translation of the English default text: “Wir spielen ein Rollenspiel in der Welt von Skyrim. Ich bin [Spielername] ein [Rassenname]“

⦁In HERIKA_PERS I added two sentences after the default text. I added something along the lines “The only language you can speak is German. All your answers must be in German.” 

“Du bist Herika, eine Bretonin die Witze und sarkastische Kommentare liebt. Du bist [Spielername]s Begleiter. Die einzige Sprache, die du sprechen kannst ist Deutsch. Alle deine Antworten müssen auf Deutsch sein.“

I didn’t think this was necessary at first since my text input was German and I expected a German answer, but sometimes I still got an English answer.
Adding these lines really worked wonders.
I found a guide for French Herika that suggests to do something similar (https://www.julien-richard.fr/skyrim-nextgen-herika/)

TTS: Azure

You will need to choose a TTS Azure voice that can speak/pronounce your desired language, so that the text generated by ChatGPT is read correctly.

You can find the available languages for Azure in the Voice Gallery of Azure: https://speech.microsoft.com/portal/voicegallery.
Here you can listen to the available languages.

From there you can also get the identifiers like “de-DE-SeraphinaNeural” to insert for TTS Azure voice.
Once you selected your voice, click on Sample code next to the Try Now Button (might be hidden behind three little dots).

Somewhere close to line 10 you should find:
config.SpeechSynthesisVoiceName = "de-DE-SeraphinaNeural"

This is what you enter to TTS Azure voice.
By accident I found out that depending on the chosen region for the Speech Service there are other voices available.
Choosing europewest gave me the most voices.


TTS: MIMIC3

Same as for Azure, you will need to choose a TTS MIMIC3 voice that can speak/pronounce your desired language.
If you open the MIMIC3 server configuration (or https://mycroft.ai/mimic-3/), you will find up to three dropdowns to make a selection for a voice.

The three dropdowns make up the identifier.
The basic “rule” is 1/2#3 (see screenshot).
You can also use the inspector in your web browser to find the strings.

For my server, I use TTS MIMIC3 voice de_DE/m-ailabs_low#angela_merkel

TTS Azure/MIMIC3 Settings Screenshot

STT: Whisper

Set SST WHISPER LANG to your language code (in my case “de”). SST WHISPER TRANSLATE must be set to False.
This is important.
We don’t want our spoken text to be translated to English.
This would likely enforce an English answer from ChatGPT.

Make sure to check & save your configuration.
Pro tip: Use the Troubleshooting Tab on the server to Test ChatGPT Connection, Test Azure TTS Connection, Test MIMIC3TTS Connection. On all of these you should receive an output in your desired language (for TTS it is going to be mispronounced English text).



I made a video in German that shows the entire process.
You can find it here: https://youtu.be/mprZ882wZvs?si=k8vRHWxcNrGHWc31 

You’re not prepared for Herika Merkel at the end!

Enjoy Herika in your own language!