As both avid podcast enthusiasts and busy students, we struggle to keep up with our favorite shows. Listening to podcasts is incredibly time consuming, so we wanted to reduce the time commitment involved.
What it does
Our web-based application allows users to upload a mp3 file, which is used to return a summarized version of the document, which keywords highlighted and selected phrases become useful links.
How we built it
The web-application was built using Django Python, we used various python libraries for file handling and various NLP libraries and a combination of Microsoft Azure and Google Cloud for the analysis of keywords.
Challenges we ran into
Transcribing the audio files accurately was an interesting problem. Our largest challenge was the summarization of the text, as we still wanted the summary to be readable. In the end we used an extractive technique
Accomplishments that we're proud of
Transcribing turned out well and efficient use of NLP libraries and tools. Additionally, leveraging Azure and GCP was a new and unique experience. It helped us realize you don't need full knowledge of ML in order to be able to use these tools.
What we learned
How to fit the various API's and ML algorithms together with a Django server and how important having a dedicated plan for our project was.
What's next for TLDL
The design is a little lacklustre, so that could be spruced up a little. We would also want to add user account in the future so people could store the summarized text. In the future we have hopes to add a lecture component so students can summarize the lectures they have to watch. Additionally, we hope to add more interactive components in the future next time.