Raj’s Sunday morning bliss moment.
If you know me, you know I’m frugal—not only do I love saving money, but I’m always looking to boost my productivity. To do that, I use any and all available tech, as long as it’s free (or cheap). Let me share with you how I saved some $$ and time using an LLM (Large Language Model).
As many of you know, I love creating video content on cutting-edge tech topics. Over the past 8 years, I’ve published 14 video courses! But there’s one thing I absolutely hate about making these videos: working on the subtitles. You know, those lines of text that show up at the bottom of the video 🙂
Let me break down the process without getting too technical. Basically, you create a text file (usually with a .srt or .vtt extension) and upload it to the video platform. This file includes the timing info and the text to display, and the video player displays the text on-screen at the right time.
Pain of subtitles in early years
My journey as a content developer began back in 2016 with a course on IBM Bluemix. This course ended up being about 10 hours of video, and I manually wrote the transcript (.srt file) for the entire thing. In total, I must have spent close to 100 hours! But that wasn’t even the worst part—any time I made edits to the video, I had to manually update the subtitles file all over again.
Freelancers meet my accent
In 2017, I created a course on REST API and decided to hire a freelancer to handle the subtitles. It was still a manual process—someone had to go through the entire video and write down the transcript. Even though I wasn’t the one typing out every word, I quickly realized I’d still have to review the whole thing. The worst part? The subtitle quality was pretty rough. When I raised my concerns, the freelancer blamed it on my accent 🙁
Still, it was better than doing it all myself. For a 10-hour course, it set me back around $400, and I spent about 20 hours fixing the quality issues myself.
From Freelancers to Foul Language
By 2019 I had published 5 more courses and had become a pro at negotiating the best prices with subtitle free lancers. But then I found out about online services that provided automated subtitles generation. Interestingly cost of using these services was more or less the same as the free lancers but they promised good quality (time saving for me). For one of my courses I tried the service and was pretty impressed UNTIL a couple of my subscribers came back to me with a question on “WHY ARE YOU USING THE CURSE WORDS IN YOUR VIDEO?” Wow it looked like my accent was playing its part again. Long story short, the automation generated curse words and I did not do a quality review as I trusted the technology 🙁
Enough of my rant, let’s fast forward to this weekend.
Gen AI to my rescue
For the last two years, I’ve been telling everyone how Generative AI is going to transform the lives of anyone who dares to give it a whirl. So when I recently published a course on Generative AI, and faced the dreaded task of transcribing the entire 20+ hours of video content, I decided to see if I could leverage an LLM for generating subtitles. I started working building a automation—literally—just 12 hours ago. And all my videos are already subtitled with AMAZING QUALITY !!!
My investment : 45 minutes of research and coding
ROI: Saving of roughly $300 + a ~20 hours of quality check time 🙂
Here’s how I whipped up a video subtitles generation utility in less than 60 minutes!
- Decided to use OpenAI open-source model called Whisper. This model is trained on 680,000 hours of audio/video data. It is great at carrying out multiple tasks such as transcription and translation!!
- Created a Python script with roughly 100 lines of code
- Ran the code on my Laptop (CPU) on 20 hours of video. Kicked off the code last night on all the .mp4 files and this morning all subtitles are done with AMAZING quality.
In my entire career, I’ve never encountered a technology as transformative as Generative AI. If someone like me can invest just a little time and reap so many benefits, imagine what large enterprises with millions of dollars at their disposal could achieve!
Would you like to check the quality of subtitles? Try this YouTube video.
Interested in checking out the code. Here is a link to the GitHub repository.
Full disclosure: I am the author of the course “Generative AI application design & development“.
#GenerativeAI #VideoSubtitles #LLM #Whisper #OpenAI #OpenAIWhisper #ContentDeveloper
