Speech recognition and synchronised video generation
An AI pipeline that recognises Quranic audio, aligns it with the text and builds the finished video automatically.
What this project is
My final-year project at UET Taxila was an AI model for recognising Quranic recitation and synchronising it with the text. It automates a task that used to take about five hours of manual video editing, and now takes about five minutes.
The project became the basis of a peer-reviewed paper at the 4th International Conference on Digital Futures and Transformative Technologies (ICoDT2) in 2024.
What needed solving
Producing a video with text synchronised to a recitation was done by hand. An editor had to listen to the audio and place each piece of text at the right moment, which took about five hours of manual editing per video and does not scale to many recitations.
What I did
- Built the AI model for recognising Quranic text and synchronising it with audio.
- Automated the video generation step so the finished video is produced from the audio.
- Published the work at ICoDT2 in 2024.
How it works
Recognise
Speech recognition turns the recitation audio into text.
Align
Forced alignment matches the text to timestamps in the audio.
Generate
Automated video generation builds the synchronised video from the aligned text.
Deliver
A finished video in about five minutes instead of about five hours.
Tools and methods
What came out of it
| Measure | Result |
|---|---|
| Production time per video | About 5 hours → about 5 minutes |
| Publication | ICoDT2, 2024 (published) |
The production-time improvement is reported in the published paper.
Interested in similar work?
Tell me about your computer vision or deep learning problem and we can see if it's a good fit.
👋 Let's talk