I kind of blew a gasket over the 2 GB figure in combination with the upcoming broadband speeds you quoted.

My brash reaction was way out of the line. I apologize for that.

Anyway, 2 GB IS unrealistic for one episode, especially for streaming purposes. Imagine sustaining those ~12mbps to several streaming clients at once, and serving those 2 GB to a couple hundred fansub watchers for ~8 episodes.

Also consider that the HQ “to-download” encodes are only in the 300-400 MB range as well, and of arguably sufficient quality. this is what youtube serves as well by the way – or at least what they served last time I checked – 720p at 2mbps, which would probably be about 400MB per episode of 24 minutes, including audio.

Reply

…actually, let me address some more of your points properly.

As for YouTube (again): I wish they’d use 720p to its fullest first. Disregarding that the vast majority of user-generated content doesn’t do that either, YouTube video compression is rather ugly. Bumping up the bitrate will certainly help there, but I wouldn’t be surprised if the 1080p content then ends up suffering from the same bitrate starvation the 720p videos do now.

As for the target audience: You do have a point about the original question being what we would like to see. I got worked up over the unrealistic figures you named. I took a more general approach to the question in my mind right away since I see the whole thing more from an encoder’s point of view than from that of a potential streaming client. This made me miss the point in a way.

Generally, I suppose that people who care enough about visual quality to go for very high bitrate HD streaming would just as well download a large file and delete it later on. Though I can appreciate the convenience of being able to “just go and watch” an episode of an ongoing show in full quality. However since full quality encodes are readily distributed through other channels anyway, I’d see the biggest “market” in people who just don’t care all that much about visual quality or are under strong bandwidth constraints that prevent them from using bittorrent or even large direct/irc downloads.

If the bandwidth was available to test which of those two audiences is bigger, it’d be an interesting experiment to serve both very high and very low quality encodes, but the fact that the current beta streaming page only serves 200p-400p encodes at low bitrates (Railgun 6 is less than 130 MB large) seems to point that they’re at least aiming for the latter audience, or maybe generally don’t have the bandwidth resources for HD streaming.

As for resolutions: while at least h.264 seems to deal rather well with high resolutions at low bitrates, what you suggest is a bit too high. But a choice like 768×432 or 704×400 (or 640×480 for 4:3) will still work and look nicely well under 100 MB target size for the video part of the stream.

As for audio: You seem to overestimate the bitrate requirements for decent audio compression. With a modern encoder like Nero’s AAC encoder you can get away with 60-80 kbps for a very listenable stereo signal that usually covers frequencies up to 15-16 kHz (above which the human hearing quickly diminishes as well), all this for 11-14 MB per episode of 24 minutes. And very good quality can be had at twice the bitrate (or size) already, so there’s no real reason to compromise much in the audio department unless you aim for almost absurdly low total file sizes – which will definitely not look pretty any more.

PS: Don’t let other people discourage you from posting anywhere, ever. I mean whatever am I gonna do if I don’t like what you have to say? Stab you in the face via TCP/IP? ;)

PPS: I wanted to write most of this in a more general comment to the original blog post, but I guess this works as well… hope SpiegelEiXXL can even be arsed to read through this lengthy rant.

Reply