[CELT-dev] Opus for audiobooks etc

[Daniel Jensen]

OK, I hadn't realized that the higher-delay SILK modes were only saving 
on frame and packet/transport overhead; I'd assumed that they'd be 
getting some kind of prediction benefit.

> The low bandpass can be objectionable for the music parts in mixed content.
>
> Keep in mind that communication codecs are usually do a wideband at 16KHz,
> which sounds clearly and obviously worse for speech. (Although not
> objectionable)

Right; I was imagining that some variety of SWB would be what one 
usually wants for audiobooks.

> Here's Opus at 20 kbps beating AMR-WB, and at 32 kbps getting close to
> transparent (at 16 kHz samplerate, you may note):

Those were the results to which I was referring when I said "Some mono 
tests seem to have given 32kbps Opus rather high marks." It's nice to 
see that it beats AMR-WB -- IIRC Audible.com's low-bitrate codec is also 
ACELP based and thus should be very comparable to AMR-WB. But AMR-WB+ is 
a rather different beast, and since there's no royalty for mono AMR-WB+ 
decoding unless you're using it on a realtime channel, AMR-WB+ seems 
like a credible competitor if you can stick to mono. I guess outside of 
USAC it's the only answer I have to Greg's "disadvantage compared to what?"

Good to hear that those results may have been slightly anomalous and 
that the stereo has seen major improvements since then.

> If your comparison points are Vorbis, MP3, Speex (or other
> pure-communication codec), or AAC it should be no contest.

I imagine so. It'd be nice to see how wide that gap is; has anyone 
looked into doing an Opus vs. Vorbis, MP3, or HE-AAC test at lower 
bitrates, esp for speech?

> For your application the improvements for encoder VBR and automatic speech
> detection that we're currently working on will probably be relevant. This
> use case would probably also benefit from additional look-ahead in the encoder,
> (and potentially two-pass rate control)

Great to hear. Is two-pass planned for after 1.0?

A couple of months ago I was in touch with Chris Lascelles, the guy 
behind the Ear Bible (www.earbible.com), suggesting he keep an eye on 
Opus because of its improvements over Speex. In his response he 
mentioned that though they used Speex for their original product they 
switched to Vorbis for a second version because they can use much 
cheaper hardware that way. (I don't own one, so I can't check the 
bitrates, but the original had 1GB of flash and a recorded bible is 
anywhere between 55 and 85 hours, usu. in the 70s, so that's probably 
~32kbps; I don't know whether the Vorbis version has more flash.) Does 
anyone know of people who are looking at developing an Opus hardware 
decoder?