harrolee/drive-gooder

don't read and drive

drive-gooder's Issues

Drop Coqui.TTS in favor of Transformers.pipeline

Use Coqui's XTTS model on HF.

Use Transformers.pipeline.

This will radically reduce the size of your container.

Simplify the appUser setup

If you drop Coqui.TTS as a dependency, you won't need to give appUser a password.
You only did that so that you could login as appUser and have the $HOME env var change.
(Coqui.TTS used $HOME)

In addtion, appUser may not need root permissions. I think that was a Coqui.TTS thing too? Unsure.

Allow user to change these values

...without rebooting the app

These env vars are defined at app start.

summarize_model = environ["SUMMARIZE_MODEL"].lower()
speech_to_text_model = environ["SPEECH_TO_TEXT_MODEL"].lower()
text_to_speech_model = environ["TEXT_TO_SPEECH_MODEL"].lower()
question_model = environ["QUESTION_MODEL"].lower()

Use chat3.5-turbo. Upgrade LangChain

The OpenAI api changed. Use the 3.5-turbo chat model.

Try using a newer version of langchain too. The current version doesn't know that OpenAI changed.

Do you need langchain? Atm, you're not using an agentic workflow.

Docker container only: coqui tts cannot save audio file

When running outside of container, in macos, generates this path: /Users/lee/projects/drive-gooder/api/backend/llm/../../output/oj5rhjoyf0bk7r5d.wav

I suspect ../../ works in macos but not linux. The docker container uses debian (from the python:3.10-slim-bullseye docker image)

Maybe the issue is the path generated by _build_temporary_file_path()?