Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Song difficult for opus before 64 kbps. (Read 5850 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Song difficult for opus before 64 kbps.

Build: libopus 06.07.2019
Sample: I know it's compressed into oblivion, even if is a FLAC source.
It's last 5 lines of the hook before verse 2 (after the clap there is a bass that loses energy and texture) under 64 kbps.
Song: Z - nobody's better
I don't know why this bassline is difficult to compress, I am not an audio engineer.


Another sample (flac) maybe the problem is intensity stereo
Muse - dig down (under 54 kbps). The bass at the start make a strange hissing noise, why, is intensity stereo?
FFmpeg opus doesn't have this artifact.


This are only examples, I think there are many samples that have the same problem or artifact.

Re: Song difficult for opus before 64 kbps.

Reply #1 –
Opus is improving, but bass is muddy before 48 kbps, I think due to its nature of low delay audio codec. Core aac high efficiency (Apple) is better at low bitrate except in the highs.

Re: Song difficult for opus before 64 kbps.

Reply #2 –
hopefully in a future version of opusenc and opusdec, the opus team can implement an AI machine learning audio restoration algorithm in both encoding and decoding. This would be a perfect test for opus, considering that it already has most of the information of the original file encoded in the stream (it goes up to 20khz)

Re: Song difficult for opus before 64 kbps.

Reply #3 –
Crunchy bass are a problem for opus.

Re: Song difficult for opus before 64 kbps.

Reply #4 –
Quote from: dwalkerdon on
hopefully in a future version of opusenc and opusdec, the opus team can implement an AI machine learning audio restoration algorithm in both encoding and decoding. This would be a perfect test for opus, considering that it already has most of the information of the original file encoded in the stream (it goes up to 20khz)

Machine learning will be a huge factor in increasing efficiency of audio codecs in the future.

Simple things like detecting speech-only content vs musical content, and adjusting quality based on that is still an unfulfilled promise, although OPUS is leading the way in some respects at sub 32kbps bitrates. For example, having the option to encode all speech-only audio at 28kbps mono (which is near transparent using OPUS), and then switching to 64kbps (or much higher) stereo for all musical audio within the same file... that capability alone will result in an unprecedented increase in efficiency especially for podcasts and other streamed content.

Machine learning also has the potential to assign bits with much greater precision and granularity, to achieve clarity where it matters on the audio spectrum in any given frame, while using sophisticated synthetic reconstruction on less critical frequencies. You mentioned bass — future codecs might ideally boost bits based on volume in each frequency band, so if there is a lot of energy in a low frequency band, it will render those frequencies more faithfully in that instant, while relying more on intelligent reconstruction elsewhere. With enough temporal resolution and agility of the encoder, I imagine you would be able to achieve huge increases in efficiency this way.

I imagine also that a next-gen version of intensity stereo will one day be able to render convincing spatial audio rivalling the true stereo source for very little expense vs mono. All this will be more CPU intensive than it is today, but increases in core counts and energy efficiency will make all of these improvements practical and affordable, IMO.
 

Re: Song difficult for opus before 64 kbps.

Reply #5 –
It wasn't intensity stereo problem. In another topic they told me it was the music/speech detector.
Opus uses only a few kbps to encode high frequencies and it is a low delay codec.
Opus 1.3 was fine at 48 kbps, a bit muddy and distorted with some EDM which has a lot of white noise and high frequencies but not much of difference to 64 kbps, which is recommended.
Now 41 kbps libopus beta sounds very similar to 64 kbps opus 1.3. 7-8 kbps gain but it's something.
I can't wait for 1.4 version.
A positive thing is that they are expanding for lower bitrates.
Probably I should have submitted a free sample and not open threads without any samples.