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: [major Did Not Do the Research] From: IETF Opus codec now ready for te (Read 7438 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

[major Did Not Do the Research] From: IETF Opus codec now ready for te

Reply #25
A transform doesn't bring any compression gain without quantization and zeroing low magnitude coefficients.


We don't zero low magnitude coefficients. Sucks to be other formats. 

But there are tradeoffs. Opus really can't do lossless, not at any rate, not even if you substitute the transforms and filters with perfect reconstruction integer alternatives and carefully controlled rounding.


IIRC Opus has folding mechanism and doesn't need to apply lowpass.  It has a similar idea like  SBR but rather does it different way.

If You allow me a parallels with video compression,  there are some codecs which preserve a good quality of picture but don't keep some small details like a film grain. A film grain , other small or high frequencies details don't necessary require a high rates and can be code at sane bitrates  with a good quality. I think You already know which video encoder does that.   

Furthermore high frequencies >15-16 kHz are rather perceived by feeling than hearing. The absence of them will be still noticeable but Opus preserves them at  extremely low bitrate cost. It's a great achievement.

[major Did Not Do the Research] From: IETF Opus codec now ready for te

Reply #26
But is 250-300kbps really the "Transparency Bitrate"?...

And saratoga, i though VBR needed to have a limit?...

What i want is a VBR mode that will achieve Transparency alone, without any bitrate indication, it will just try it´s best to achieve transparency...

And that PCM isn´t Lossless of analogue, do you mean that it isn´t Perfectly Calculated the analogue signal?
As i don´t think digital can make an Exact Replica of Analogue ( i think it´s pretty much impossible as analogue isn´t limited to herz,bits etc right?).
Yes, the best Opus and AAC encoders as of now should both be transparent by 192kbps, though the possibility remains that some exceptional listener somewhere may be able to ABX some killer sample when using ideal equipment.

It's true that normal "unconstrained" VBR does still have an upper limit, determined by the format, for how much data can be encoded for a single frame. That's a very loose constraint in any normal situation- it wouldn't be optimal to output frames larger than that anyways. The option is available to tighten the constraints, but if you had any need to do so (e.g. realtime low-bandwidth network streaming) you would already be aware of that. You don't need to worry about it- just let the encoder make its decisions.

Because individuals' perception, their equipment, and the characteristics of sound samples vary so widely, "make it transparent but don't use any more bits than absolutely necessary to do so" is at the same time both far too vague and far too precise of a requirement to make of an encoder. Instead, encoders allow you to set a quality level, which corresponds across a wide variety of input (but not necessarily on a single file or group of files) to an average bitrate. It's up to you to determine what quality level meets your transparency and bitrate needs.

An analog signal may contain an infinite amount of information, so in this sense analog-digital conversion is not "lossless." However, audible detail is finite. The sampling theorem says that discrete sampling with infinite precision N times per second perfectly reproduces all frequency components of an analog signal up to N/2 Hz. Since no humans have demonstrated an ability to hear signals significantly above 20kHz, 44.1kHz sampling is sufficient to exactly reproduce all audible frequencies. Finite precision sampling does introduce quantization noise, but 16-bit quantization noise is minimal (-96dB RMS) and should be inaudible in normal conditions, and 24-bit quantization noise (-144dB RMS) is definitely inaudible.

Please take any further questions and confusions about what lossless compression is or isn't, about what VBR is, and about sampling theory/analog-digital conversion to a new thread, and perhaps try reading some basic reference materials.

[major Did Not Do the Research] From: IETF Opus codec now ready for te

Reply #27
Quote from: IgorC link=msg=0 date=
IIRC lossy codecs like Opus, Vorbis and AAC apply Fourier-related transforms which are lossy.

Nope, the transforms are perfectly invertible, hence lossless. The lossy part of lossy waveform coding is the quantization of the spectrum (and in case of MP3, use of a non-perfect-reconstruction filter bank, but that's a detail).


The transform is perfectly invertible only if you 1) assume infinite precision or 2) use an integer approximation. Otherwise you have rounding error and different SDOs deal with that in different ways. The ITU decoders all have a bit-exact definition. OTOH, Opus (MP3 and AAC too IIRC) specified a margin or error. This means that two different encoders can produce slightly different outputs. Without that, float implementations are pretty much impossible.

[major Did Not Do the Research] From: IETF Opus codec now ready for te

Reply #28
One could always assume a specific decoding algorithm and pack a (decoded minus original) sample-wise delta into the encoded file in addition to the lossy stream. But this would make sense only in case the delta information has some specific property which makes it exceptionally easy to compress compared to the original full signal. Otherwise, just use FLAC.

[major Did Not Do the Research] From: IETF Opus codec now ready for te

Reply #29
But is 250-300kbps really the "Transparency Bitrate"?...

And saratoga, i though VBR needed to have a limit?...

What i want is a VBR mode that will achieve Transparency alone, without any bitrate indication, it will just try it´s best to achieve transparency...

And that PCM isn´t Lossless of analogue, do you mean that it isn´t Perfectly Calculated the analogue signal?
As i don´t think digital can make an Exact Replica of Analogue ( i think it´s pretty much impossible as analogue isn´t limited to herz,bits etc right?).
Yes, the best Opus and AAC encoders as of now should both be transparent by 192kbps, though the possibility remains that some exceptional listener somewhere may be able to ABX some killer sample when using ideal equipment.

It's true that normal "unconstrained" VBR does still have an upper limit, determined by the format, for how much data can be encoded for a single frame. That's a very loose constraint in any normal situation- it wouldn't be optimal to output frames larger than that anyways. The option is available to tighten the constraints, but if you had any need to do so (e.g. realtime low-bandwidth network streaming) you would already be aware of that. You don't need to worry about it- just let the encoder make its decisions.

Because individuals' perception, their equipment, and the characteristics of sound samples vary so widely, "make it transparent but don't use any more bits than absolutely necessary to do so" is at the same time both far too vague and far too precise of a requirement to make of an encoder. Instead, encoders allow you to set a quality level, which corresponds across a wide variety of input (but not necessarily on a single file or group of files) to an average bitrate. It's up to you to determine what quality level meets your transparency and bitrate needs.

An analog signal may contain an infinite amount of information, so in this sense analog-digital conversion is not "lossless." However, audible detail is finite. The sampling theorem says that discrete sampling with infinite precision N times per second perfectly reproduces all frequency components of an analog signal up to N/2 Hz. Since no humans have demonstrated an ability to hear signals significantly above 20kHz, 44.1kHz sampling is sufficient to exactly reproduce all audible frequencies. Finite precision sampling does introduce quantization noise, but 16-bit quantization noise is minimal (-96dB RMS) and should be inaudible in normal conditions, and 24-bit quantization noise (-144dB RMS) is definitely inaudible.

Please take any further questions and confusions about what lossless compression is or isn't, about what VBR is, and about sampling theory/analog-digital conversion to a new thread, and perhaps try reading some basic reference materials.


That was some very neat information there, thanks i think i got a good grip of it:)
About what´s lossless and Audible etc, very informative!

And a little question, as AAC and Vorbis is pretty much Transparent at 196 most of the time, what will Opus "192" be?
I know that there isn´t an Exact way to answer that, and that it´s not fully completed, but is there any thoughts about it?

As in low bitrate which Opus it made for, it´s really brilliant!
96 sound glorious, 64 sound like mp3 at 96 or something (which i think is good!), and 32 and lower sound very good for being that low, it´s not Sickly much artifacts so you have to focus to hear the real sound (Mp3 at low bitrate gives that sick artifacts, don´t know what it is, sounds like Space stuff moving around in light speed).