Voice search with OpenAI Whisper, DuckDB, and Metaphone

This repository is a companion to Using Metaphone to handle bad transcriptions in voice search for sommeliers . Read the blog post for an explanation of how this code works.

This a Streamlit app. You can check it out on Streamlit Cloud.

Running the Streamlit app locally

You'll need to place your OpenAI API key in .streamlit/secrets.toml:

OPENAI_API_KEY = "<your-openai-api-key>"

Then run poetry run streamlit run main.py.

This demo uses OpenAI whisper, which costs $0.006/minute as of May 2023. It's cheap.

Navigating this codebase

• index.py prepares all the indexes in wines.duckdb.
• search.py implements querying our indexes.
• main.py is our Streamlit app.

The blog post walks through our search indexes and how we query them in detail.

Contributing

This codebase is purely educational. If you see errors, please open a PR. If you want to make improvements, I encourage you to fork it and talk about your changes -- let me know if you do!