View Proposal

---

Proposer

John See

Title

Automatic Podcast Summarisation and Information Extraction Application

Goal

The aim of this project is to develop a web-based application to automatically summarise podcast (audio) episodes. Among the tasks include implementing an end-to-end pipeline involving speech transcription, text processing, and AI-driven summarisation. Key information such as topics, discussion points and takeaways are expected to be generated.

Description

This project will develop a web-based application that automatically generates concise and structured summaries from podcast episodes. Users can upload an audio file or provide a podcast episode, after which the system transcribes the conversation and extracts useful information such as: (a) Episode summary, (b) Key discussion points, (c) Important insights, (d) Topics discussed, (e) Timestamped sections, (f) Keywords/topics. Optional tasks: - The application could also allow users to select different summary formats, such as brief overview, detailed summary, or bullet-point highlights. - The project can evaluate different summarisation approaches and investigate factors such as podcast length, number of speakers and type of conversation on summary quality.

Resources

Open-source tools and libraries are available.

Background

Podcasts have become an increasingly popular form of digital media, covering a wide range of topics including education, news, technology, entertainment, business and personal experiences. Unlike short-form content, podcast episodes can often last from 30 minutes to several hours and may contain a large amount of information presented through extended conversations and discussions. While this format provides listeners with rich and detailed content, the length of podcast episodes can make it difficult for users to quickly identify the main topics, important points and useful insights without listening to the entire episode. Recent advances in automatic speech recognition (ASR) and large language models (LLMs) have created opportunities to automatically process and understand spoken content. Speech recognition systems can convert podcast audio into text, while modern language models can analyse the resulting transcript and generate concise summaries. These technologies make it possible to transform a long podcast episode into more accessible forms of information, such as summaries, key discussion points, topics, keywords and actionable takeaways.

Url

Difficulty Level

Moderate

Ethical Approval

Full

Number Of Students

1

Supervisor

John See

Keywords

audio transcription, text processing, text summarisation, web application

Degrees

Bachelor of Science in Computing Science