Lilia Quotes Compilation Video

Lilia maintains a list of memorable quotes from her Twitch livestreams. The quotes looked so funny I decided to try and find the original video sources, extract the quotes, and compile them into a single video.

TL;DR I made a video with 58 Lilia quotes.

In a previous post I wrote about extracting (YouTube) video snippets where a certain word was spoken, and compiling them into a single video. The result was a a video where Lilia said “cheese” or “cheesy” 300+ times. But now I would be looking for entire quotes from Twitch livestreams.

Lilia’s quote list contains the following information: an integer ID (1-218) for the quote, the quote itself, the username of the person who recorded the quote, a text describing the stream (e.g. the title of the game Lilia was playing), and a timestamp. For example:

2 “I won’t need a key if I have an axe, right?” lilia_live Conarium 10/19/2019 12:39 AM

Unfortunately, the timestamp was not much use because it’s constant for all quotes with ID 1-210. Weird. The text describing the stream was more helpful, but a lot of times the text was simply “Retro” or “IRL” instead of a game title.

So there was not much to go on except the quotes themselves. I’d watched a fair number of Lilia’s videos but I couldn’t remember if I’d heard any of the quotes before. And watching all of Lilia’s content from start to finish to find the quotes was just not feasible.

It seemed to me that Lilia had uploaded the vast majority of her Twitch broadcasts to YouTube, so I decided to search for the quotes in the auto-captions, which YouTube generates automatically.

It was easy to grab the auto-captions using youtube-dl. The file format presented some problems since it mixed the text I wanted to search with timestamp information and other meta information, but that was a relatively minor issue.

The big problems were: A) the auto-captions were wrong quite often, B) some videos didn’t get any auto-captions at all, and C) the recorded quotes didn’t always match what Lilia said in the video. I’m not going to dwell on C, it’s simply not reasonable to expect them to be verbatim all the time. I’ll elaborate a bit on A and B, though.

  • A. The auto-captions were often wrong when the audio was low, and when Lilia’s voice overlapped with the sounds from the game. Nothing mysterious about that. But sometimes the audio-to-text conversion was simply bad, resulting in missing words / sentences, or completely different words.
  • B. Videos with “too poor” audio quality did not get auto-captions at all. This was extremely rare though, I noticed it for one or two of the very earliest videos. It was more common that “too lengthy” videos didn’t get any auto-captions. I think all videos shorter than three hours all got auto-captions, while none that were four hours or longer did (or was it five?).

In the end I could only find 58 of the 218 quotes. Disappointing, but considering the number of difficulties I encountered I have to be happy with that, for now.

I decided to compile the video snippets with OpenShot, with the ambition to include some fancy transitions, but that failed. It was too much work for a subpar result, I should’ve just used ffmpeg. Also, the video quality bothers me, but 720p is still technically HD I suppose…

The video with 58 of Lilia quotes is on YouTube, and here’s a list of the quote ID:s in the video, and their source videos:

This post is part of the Lilia Quotes project.