Generating Asterisk sounds with AI (including new German sound files)

Hello everyone,

I have created a little CLI tool that translates the original Asterisk transcriptions and generates new sound files based of that. It uses ChatGPT and DeepL for the translation and the Google Text-To-Speech API for the sound files. There is also a transcription for sounds that are missing in the original transcriptions.
Here is the link to the Github repository:

Currently the tool provides new Asterisk sound files in German. Pull requests for other not well supported languages are welcome.

New German Sound Files (99,9% Coverage)
There are a lot of missing German sound files in FreePBX. This makes a lot of modules like the Voicemail unusable. This is the reason why I have created this tool.

Download German Sound Files

FreePBX Installation Guide


Have these been validated by a native speaker?

I’d also note that Google Translate is also AI.

The German sound files have been validated by myself (native speaker). I have tested all available FreePBX feature codes, created Call Flow Controls, Call Recordings and Voicemails. I have manually improved a lot of translations. But most of the lines are translated by ChatGPT. ChatGPT has one big advantage over DeepL or Google Translate, it has a better understanding of the context. So for example Google Translate will translate “All circuits are busy now.” to “Alle Schaltkreise sind jetzt belegt.” which is not wrong but is missing the context of a phone system.
This is just a tool to speed up the creation of unsupported languages. But as it says in the README “… the translations still needs some manual work.”.
Are the German sounds ready for a productive environment? I would say yes.

1 Like

That might be a problem for repeatability, as it may have learned a context specific to your account. I seem to remember reading that it can give different answers to the same question.

I have used the OpenAI API with the model “gpt-3.5-turbo”. Unlinke the API is not keeping the context. You always have to set the context again per request. So this will not have anything to do with my account. But it is true the responses will always vary a little bit.

tested it as a native speaker.

Definitly good enough for a production enviroment.

Sometimes there are small issues, like the echo test.
It says “machine” instead of “device”

1 Like

Thank you for testing. If you find any sound file which needs improvments, feel free to create an issue or pull request on github.
How to contribute to this projetct

Thank you for making this. The borderline unusable German audio files in FreePBX were always an annoyance for me.

Thanks…great job…I just installed the package…I am a native speaker too :slightly_smiling_face:

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.