Show HN: Scryer - deep search for video and podcast content

by prohobo • 4 years ago

8 8 4 years ago

prohobo4• 4 years ago

Heya, I'm working on creating a metadata search engine (using transcripts) for YouTube, Spotify, etc. Basically any video/audio content.

Everyone always complains about how bad multimedia search tools are, so this is something that can address that. Some people tried to do stuff like this back in 2006 (Blinkx) but never made it very far. This takes those attempts a step further than basic transcript searching, and uses machine learning to extract topics, entities, and keywords that can be used to filter on the data sets - as well as show "gists" of the content in any piece of media.

It's very much a work in progress. Currently it doesn't have a mobile layout, it's buggy, and it's quite - uh - "overwhelming", but hopefully someone can see the utility of it.

HanClinto3• 4 years ago

This is great! Literally just yesterday I was looking for something like this.

I've used the Chrome extension YouTube Captions Search ( https://chrome.google.com/webstore/detail/youtube-captions-s... ) and that works quite well... but only for one video at a time. I was just thinking that it would be great if there were a public repository of YouTube CC content that we could search through, and lo-and-behold -- this pops up on my feed today.

Very nicely done with this! Are you using the YouTube-provided CC files as a data source for this, or are you creating your own transcripts from scratch?

prohobo4• 4 years ago

Thanks, actually I'm looking for any feedback I can get on this and then make it as open source as possible (while still earning something) so people can index their own content feeds and use it.

Right now, yes we're taking the transcripts from Google - just because it's easy. But I'll implement our own transcriber once we start incorporating other sources.

HanClinto3• 4 years ago

Yeah, that makes a lot of sense!

Is there any way that we can contribute? Like, if I watch a video with CC turned on, it feels like it would be easy for me to upload the CC file(s) to a central server, keyed with the ID of the video. Any user watching a video once would be enough to add a video to the database, and we could crowdsource indexing this content to help burden some of the load?

polyterative• 4 years ago

cool stuff

Show HN: Scryer - deep search for video and podcast content

Related Stories