wtf is pronounce-assess?

pronounce-assess is a Python library that checks how well someone pronounces words in real time. The author built it because there were not many open tools for this, pointing to a couple of paid cloud services as the closest existing options. You install it with a simple pip command. To use it, you create an assessor object, which loads the underlying model and automatically decides whether to run on your graphics card or your regular processor. You give it a reference sentence, which it converts into the expected sounds a speaker should make, then stream in chunks of audio as someone speaks. For each chunk, it returns the sound it detected, a label, and a score, so you can build a live pronunciation feedback experience such as a language learning app. The library notes that real time assessment works much better with a graphics card, and while it still works on a regular processor, it will be noticeably slower. The lower level building blocks used to load the model, convert text to sounds, and stream results are all available separately if you want more control than the simple assessor object gives you. The audio input can come from a file, a live microphone, a websocket connection, or any other source that provides small chunks of numeric audio data. The project includes example scripts for trying it with a live microphone and with a Gradio based web demo, along with basic instructions for setting up a development environment and running its test suite. The license is MIT.