It uses the new ChatGPT API, Whisper, and Amazon Polly.
The result is a virtual chatbot inside Telegram Messenger, that works similar to the ChatGPT website. The major difference is that you can just talk to the bot rather than having to write. And it can respond using text-to-speech voice synthesis as well.
This seems like a minor feature, but being able to ramble on about some things and having it understand you is quite a surreal experience.
I use it for all kinds of things such as walking around and turning a spoken brain dump into an organized list. Or asking it to write an SQL query for me when I’m not exactly sure how to phrase it properly. I just ramble on and it understands me well enough to come up with a useful answer.
Will add more functionality over time. Let me know what you think!
I have been working on a telegram bot too. Mine has these functionalities at the moment:
Model GPT Turbo 3.5
Memorize prompts & answers (reset by /command in telegram)
Vocal message analysed by whisper and answer returned by voice (same memory used so you can use alternatively voice or text
Image generation via DALL-E (requests memorized as well)
Error handling with messages in telegram
GPT answers split to be compliant with telegram characters max size.
I will add funtionalities over time as well but no more idea or something like games…
I haven’t quite yet figured out a reliable way to count tokens (OpenAI’s libraries are limited to a handful of programming languages I think), so that might be causing it.
