ATTA is a voice chatbot. It gets audio and returns proceed audio by LLM. For now everything is running on Fast API.
The program is currently using my tunned version of whisper-base model "marcsixtysix/whisper-base-pl" to speech recognition. You can see the model in my github repository: https://github.com/mryt66/Speech-recognition-pl
The output from the speech recognition model is processed by a language tool that formats the text before sending it to the large language model "marcsixtysix/gemma-3-4b-it-pl-polqa" via Ollama. For this project, I have fine-tuned the Gemma-3-1B-IT model to function as a Polish-language Q&A system. You can find it here: https://huggingface.co/marcsixtysix/gemma-3-1b-it-pl-polqa-GGUF
Once a response is generated by the LLM, it is passed to the edge_tts, which converts the text into voice output.
Example of program usage
Api's endpoints