Earkind is an AI-powered podcast generation tool that automatically creates full podcast episodes from a list of news articles and research papers. The pipeline combines large language models (LLMs) with neural expressive text-to-speech (TTS) and programmatic audio editing. It starts with a .txt file containing selected news and arXiv paper URLs; titles and abstracts are crawled from arXiv, and additional information is extracted from raw PDF text using the ChatGPT API. Custom system and user prompts generate scripts for each section (intro, outro, transitions, sponsor) organized as conversations between characters—host Giovani Pete Tizzano, analyst Robert, and research expert Belinda. The scripts are then parsed for speaker markers and fed into Azure's TTS for recording. Finally, audio editing with Pydub combines narrations, jingles, sound effects, and background music, automatically adjusting volumes and overlaying sections. The tool also generates podcast descriptions with timestamps and titles using ChatGPT. The first show, "GPT Reviews," is a daily mix commenting on AI news and research papers.
