This project is to provide speaker embeddings from podcasts. These embeddings can be used for a variety of cases, not limited to - generating local conversational assistants that are "experts" in the topic that the respective podcast covers, comparing speaker/podcast similarity, etc.

This project also takes in user inputs to search for podcast episode titles (given a Podcast) based on keyword search with three types of matching - exact, fuzzy, and similarity. This service also takes in a podcast name inputted by the user, scrapes the respective RSS feed, and adds the podcast to a Transcription DAG that gets triggered on each episode, determined by the delta in the time between episodes from the XML file for the RSS feed. Embeddings of speakers from podcasts are stored in a PostgreSQL Database. Work in progress.