Inspiration
At times, students seek ways to get a condensed version of a lecture or other video in order to save time or to find relevant information. We thought Ncapsulate would be the ideal solution to summarize longer videos and for client's to quickly catch up on the media of their choice.
What it does
It lets users upload a video of mp4 format of any time length, and the website returns a short summary of the whole video, as well as the full transcript.
How we built it
We used python libraries like "SpeechRecognition", "Vosk", and "ffmpeg" for converting .mp4 format into the text file. Then we use the generated text file and natural language processing library "Spacy" to create a short summary for the video.
Challenges we ran into
One of the main challenge that we ran into was improving the accuracy of the generated transcript and improving the speed. At first, the accuracy was very low and for the most part showed incorrect transcript of the video. The program worked slowly too. Approximately, it took 10 minutes to generate a transcript for a 10 minute video. But, at the end of the day we were able to improve the accuracy to 90-99%, and now it also takes about 40 seconds for a 10 minute video (including the nlp). Initially, we wanted to use a topic-model, but then we realized that it wouldn't fit with our design idea for our project. Although the topic-model did work for us, we realized that it wouldn't work for the client as we intended. As a result, we slightly modified our idea to return a summary of the text instead of a list of the key topics of the text.
Accomplishments that we're proud of
We are proud of how quickly we came up with this idea and a solution. And, how we quickly optimized the program and fixed the challenges mentioned above. We are also proud of overcoming dependency bugs and modifying our idea to be the most practical for the client without abandoning our original idea.
What's next for Ncapsulate
The next step for our project is to decrease the wait time even further, containerize our application, enable the website recording functionality, and to launch a user-friendly website to the Internet. As the world gears towards more digital media, we hope that Ncapsulate can act as a multi-purpose platform where users can get any form of media summarized in a short and concise manner.
Built With
- css
- html
- javascript
- natural-language-processing
- python
- speechrecognition
Log in or sign up for Devpost to join the conversation.