Opus is now RFC6716, version 1.0.1 released!
Reply #31 – 2012-09-22 22:56:41
I would assume they may get better results with one of the experimental psychoacoustic tunings discussed here. If they were to use 48 to 64 kbps (or maybe 80 to 96 kbps), possibly the experimental psy tunings would help, as I believe they're focused on the CELT mode, and the full-range bass would also be welcome. It seems likely that 128 kbps is going to be pretty excellent / near transparent in either mode and hard to distinguish between main or experimental branch in normal listening. The 24 kbps stream (Super Wideband at 24 kHz sampling rate, I think) is below the level where CELT dominates (I think most of the experimental psychoacoustic tunings have been developed for the CELT layer) and instead I believe Opus always uses Hybrid mode with SILK encoding the signal up to 8kHz and CELT encoding the spectrum between 8-12 kHz, I believe, so I don't expect there would be much difference between main and experimental there either. I guess the modes chosen are based on use-cases. Unlikely to include a PSTN modem, but maybe includes cellular streaming to mobile devices at sporadic and variable rates of GPRS, which is often about 28 kbps in rural areas of the UK on some networks. Absolute Radio labs' 128kbps Opus stream sounds excellent but at times the 24 kbps Opus was really lacking in bass, I believe - I compared to their AAC+SBR 24kbps stream. On the other hand, at times Opus sounded like it had less high-frequency harshness than the AAC+SBR version but also at times it had less HF sparkle due to its presumably 12kHz lowpass. For a speech-oriented encoder setting, Opus24 is very music-friendly. I'd probably pick the AAC+SBR at 24kbps, but the artificial-sounding nature of it might grate with me eventually and I might choose Opus instead. I know there's an adaptable HF filter in the SILK layer, so presumably its effect will vary with the audio source. It would be lovely if bass could be preserved by Opus when music is detected at 24 kbps without losing too much fidelity due to the bits used there. As Opus is defined by the decoder, which seems to simply add the SILK and CELT decodes, I'm not sure if there might be scope in hybrid mode to choose to use CELT to encode the area below the high-pass filter that removes non-speech background noise from the SILK layer but might be amenable to encoding using longish blocks.