Audio Lossy and Lossless

[Jay123210599]

Will most people be able to tell the difference between lossy audio (AAC, MP3, OPUS) and lossless audio (FLAC, ALAC, PCM)? Which lossy audio format is the most compatible with most devices?

[poisondeathray]

Will most people be able to tell the difference between lossy audio (AAC, MP3, OPUS) and lossless audio (FLAC, ALAC, PCM)?

If you use enough bitrate , most people will not be able to tell the difference with lossy audio

Which lossy audio format is the most compatible with most devices?

What kind of device ? Stand alone player ? Android box ?

newer portable devices like modern smart phones - AAC
legacy portable devices - MP3

[dellsam34]

@Jay123210599: Only to a certain bitrate, anything above 320Kbps is not possible for humans to hear the difference, Although there are audiophools that claim to hear through their skull bones. The reason people use lossless for backup is to preserve the integrity of the files. I personally use FLAC for backing up CDs.

[Jay123210599]

Will most people be able to tell the difference between lossy audio (AAC, MP3, OPUS) and lossless audio (FLAC, ALAC, PCM)?

If you use enough bitrate , most people will not be able to tell the difference with lossy audio

Which lossy audio format is the most compatible with most devices?

What kind of device ? Stand alone player ? Android box ?

newer portable devices like modern smart phones - AAC
legacy portable devices - MP3

I'm using Shutter Encoder, so what should the bitrate be? And I mean the newer/modern devices that people are using nowadays.

[poisondeathray]

I'm using Shutter Encoder, so what should the bitrate be?

For what ? Stereo ? 5.1 ? 7.1 ?

The audio source matters too - the less complex the material, the less bitrate required . The more complex, the more bitrate required . You can try "quality" based audio encoding too . Higher qualtiy levels result in higher bitrates on the same source

I don't use Shutter, but if ever in doubt, just use something very high that Shutter lets you enter

And I mean the newer/modern devices that people are using nowadays.

AAC for portable device
AC3 for home theatre equipment

[Jay123210599]

I'm using Shutter Encoder, so what should the bitrate be?

For what ? Stereo ? 5.1 ? 7.1 ?

The audio source matters too - the less complex the material, the less bitrate required . The more complex, the more bitrate required . You can try "quality" based audio encoding too . Higher qualtiy levels result in higher bitrates on the same source

I don't use Shutter, but if ever in doubt, just use something very high that Shutter lets you enter

And I mean the newer/modern devices that people are using nowadays.

AAC for portable device
AC3 for home theatre equipment

Multi (multiple audio tracks)

[poisondeathray]

Just max it out if you're interested in quality, is there a slider or some drop down menu in shutter ?

Does shutter use ffmpeg libavcodec AAC or libfdkaac ?

It used to be Apple AAC and FDK AAC were rated the highest quality for AAC encoders

There were problems with ffmpeg AAC, but there aren't many high bitrate tests published, only low to mid bitrate range tests where it was always the worst . There were tests on Hydrogenaudio and various other forums, but I haven't seen anything recent . Not sure of the current status.
https://www.reddit.com/r/ffmpeg/comments/ms77y4/libfdk_aac_vs_aac/

[robertazimmerman]

No one can indicate what "most people" can or cannot discern. They can only say what they are capable of hearing (or not).

With the low prices of hard drives and flash drives, it is beyond me why anyone would use lossy compression for audio these days.

[dellsam34]

Sure, no need to use lossy compression anymore, but the placebo effect is real and people who believe a $300 USB cable improves audio quality are real.

[Cornucopia]

Let's put this another way: if you ignore the outliers and exceptions, if you encode most lossy codecs at their *most common* bitrates and compare those head-to-head to losslessly compressed, most people WILL be able to tell the difference, even though they thought there wasn't when listening to only the lossy versions.

Or are you just interested in the outliers/exceptions?


Scott

[dellsam34]

I'm just saying that at certain bitrate level no human can discern the difference, There is nothing wrong with using lossless, I use it all the time.

[poisondeathray]

I'm just saying that at certain bitrate level no human can discern the difference, There is nothing wrong with using lossless, I use it all the time.

Yes, and the same for video . At very high bitrates, nobody can tell the difference if the audio was encoded correctly (some lossy audio encoders apply a pass filter, or Hz cutoff, it must be disabled)

But the point at which a certain % of people can tell is going to vary depending on various aspects such as the source complexity . That "certain bitrate level" is going to be different for different sources . Does the OP want to cover 93% ? 99% ? If you use very high bitrates you cover 99.999% of all cases/scenarios/people .

With the low prices of hard drives and flash drives, it is beyond me why anyone would use lossy compression for audio these days.

I agree in general - but there are usage cases where lossy compression is incompatible . Some of the devices mentioned above for example, cannot play certain types of audio . Other scenarios such as streaming - not all types of lossless audio are compatible

To say there are no valid usage cases for lossy audio would be wrong

[pandy]

At very high bitrates, nobody can tell the difference if the audio was encoded correctly (some lossy audio encoders apply a pass filter, or Hz cutoff, it must be disabled)

Not necessarily - there are types of natural sound that unless codec is specially designed to deal with them they may be perceived as distorted even at reasonable high bitrate in lossy codec. Lossy codec use conversion from time domain to frequency domain and to achieve high compression they need to sacrifice resolution in time domain - classical example is FFT where large FFT size lead to high frequency accuracy but poor time domain accuracy, of course fancy approach may try to adaptively change transformation size but this works up to some point.
Luckily most people is unable to hear difference between 128kbps MP3 and other codecs - they using BT wireless headphones so audio quality is determined by BT codecs anyway.

[poisondeathray]

At very high bitrates, nobody can tell the difference if the audio was encoded correctly (some lossy audio encoders apply a pass filter, or Hz cutoff, it must be disabled)

Not necessarily - there are types of natural sound that unless codec is specially designed to deal with them they may be perceived as distorted even at reasonable high bitrate in lossy codec. Lossy codec use conversion from time domain to frequency domain and to achieve high compression they need to sacrifice resolution in time domain - classical example is FFT where large FFT size lead to high frequency accuracy but poor time domain accuracy, of course fancy approach may try to adaptively change transformation size but this works up to some point.
Luckily most people is unable to hear difference between 128kbps MP3 and other codecs - they using BT wireless headphones so audio quality is determined by BT codecs anyway.

128kbps MP3 is considered low to medium bitrate range - it has easily detectable "warble" . Probably 50-80% of people (not of senior age) can detect this

I wrote "very high bitrates." Something like 320kbps Mp3 for stereo would be considered very high bitrates



If you still think that applies to maxed out, very high bitrates - I disagree

Post a sample which you think is a type of "natural sound" that demonstrates this "distorted" perception when re-encoded properly at very high bitrates. I suggest performing a blind test.

[pandy]

128kbps MP3 is considered low to medium bitrate range - it has easily detectable "warble" . Probably 50-80% of people (not of senior age) can detect this

I wrote "very high bitrates." Something like 320kbps Mp3 for stereo would be considered very high bitrates



If you still think that applies to maxed out, very high bitrates - I disagree

Post a sample which you think is a type of "natural sound" that demonstrates this "distorted" perception when re-encoded properly at very high bitrates. I suggest performing a blind test.

Fast transient natural sounds are usually distorted by lossy encoder as they use relatively large chunk of data and push frequency transformation to at least 512 points - so anything with transients around 512 samples will be unavoidably "blurred" in time domain (512 samples is 11ms long) - you can reduce frequency resolution to improve time domain resolution or try to apply pre-distortion to minimize distortions. This is simple math, you can apply overlapping but still this not solve all issues. Sound samples - triangle, castanets, harpsichord etc. Of course perception is disputable - if you don't know how those sources sounds in real life you probably are unable to hear difference - so going for true acoustic concert with harpsichord to learn you brain how to perceive true sound is unavoidable.

And i bet 128kbps MP3 quality is better than all BT headphones codecs so whole this discussion is affected by reality - nowadays people consuming audio - quality is irrelevant (unless it is really poor). Btw - many years i made test with lame and 'freeformat' where even at 320++kbps there was significant difference between source and encoded MP3 version of signal.

This is my private opinion - being 52 yr old fart and still hearing 17kHz.

[poisondeathray]

Fast transient natural sounds are usually distorted by lossy encoder as they use relatively large chunk of data and push frequency transformation to at least 512 points - so anything with transients around 512 samples will be unavoidably "blurred" in time domain (512 samples is 11ms long) - you can reduce frequency resolution to improve time domain resolution or try to apply pre-distortion to minimize distortions. This is simple math, you can apply overlapping but still this not solve all issues. Sound samples - triangle, castanets, harpsichord etc. Of course perception is disputable - if you don't know how those sources sounds in real life you probably are unable to hear difference - so going for true acoustic concert with harpsichord to learn you brain how to perceive true sound is unavoidable.

The point was about a lossless audio source , and it's very high bitrate lossy re-encode.

Sure, math is good - do you have a practical example that clearly demonstrates a difference ?

Yes, perception is disputable and varies immensely between people. I'm making that claim that very few people would be able tell the difference on a blind test (say 100 blind tests of different types and genres of audio) under normal listening conditions (Not looking at waveforms, or analysis in an editor). A small fractional % of people actual do have super human hearing, but you still cover >99.99% of the population with very high bitrate lossy audio .

Btw - many years i made test with lame and 'freeformat' where even at 320++kbps there was significant difference between source and encoded MP3 version of signal.

And that's what I'm looking for. Examples that show significant differences . Can you dig those up? They would make excellent case studies and learning examples to improve audio algorithms in the future

[pandy]

The point was about a lossless audio source , and it's very high bitrate lossy re-encode.

Sure, math is good - do you have a practical example that clearly demonstrates a difference ?

Yes, perception is disputable and varies immensely between people. I'm making that claim that very few people would be able tell the difference on a blind test (say 100 blind tests of different types and genres of audio) under normal listening conditions (Not looking at waveforms, or analysis in an editor). A small fractional % of people actual do have super human hearing, but you still cover >99.99% of the population with very high bitrate lossy audio .

And that's what I'm looking for. Examples that show significant differences . Can you dig those up? They would make excellent case studies and learning examples to improve audio algorithms in the future

As i've already said - for normal, average listener nowadays 128kbps mp3 is probably OK in terms of quality especially in case where wireless headphone are used (important as modern smartphones acting as personal multimedia device are lacking of headphone connector so only wireless BT headphones can be used to listen sound).
And difficult signals are well known in audiophile society - i'm not innovative in this - some codecs are equipped in techniques to mitigate those effects but at a cost of significant increase in bitrate demand - this is observable for lossy and for lossless codecs - they both require significantly higher bitrate to deal with those sources. Question about what can be easily perceived is open - not arguing on this. Training may help to hear problems.

[dellsam34]

Damn! @ 52 still hearing 17KHz, Must be nice.

[poisondeathray]

As i've already said - for normal, average listener nowadays 128kbps mp3 is probably OK in terms of quality especially in case where wireless headphone are used (important as modern smartphones acting as personal multimedia device are lacking of headphone connector so only wireless BT headphones can be used to listen sound).

But 128kbps stereo MP3 would not be ok for "not being able to tell the difference"

And difficult signals are well known in audiophile society - i'm not innovative in this - some codecs are equipped in techniques to mitigate those effects but at a cost of significant increase in bitrate demand - this is observable for lossy and for lossless codecs - they both require significantly higher bitrate to deal with those sources. Question about what can be easily perceived is open - not arguing on this. Training may help to hear problems.

Yes some posted on hydrogenaudio and similar sites, I was hoping you had some handy

Have you examined high bitrate range OPUS and AAC ? 320-512kbps ? on killer/difficult samples ?

[pandy]

But 128kbps stereo MP3 would not be ok for "not being able to tell the difference"

Yes some posted on hydrogenaudio and similar sites, I was hoping you had some handy

Have you examined high bitrate range OPUS and AAC ? 320-512kbps ? on killer/difficult samples ?

Well... 128kbps is OK if you are listening it on your smartphone trough wireless BT headphones - this is (my observation on streets) approx 90% people listening something - i see sometimes on their screens active Spotify so my assumption is that most of them listening music.

Nope, this well known "difficult" sounds - nothing changed since MP1 times - i recall some of those difficult sounds was part of bitstream suite provided by Philips in early 90's - every codec using transformation from time to frequency domain is affected more or less in similar fashion - you can use some workarounds/hacks but... problem exist.

To be honest i've lost interest in music lossy compression since FLAC get popularity - not use AAC/OPUS at all - if i need to use lossy then it is MP3 usually VBR with highest quality but prefer FLAC over lossy.

[Bwaak]

Damn! @ 52 still hearing 17KHz, Must be nice.

With my Sony MDR-7506 and an app, I tested 12.5 kHz, oops. OTOH, I still hear mosquitoes, are they below this frequency?

I would take 128 kbps MP3 over an average audio cassette any day. No hiss, no wobble, no chewed tape, and hours upon hours without flipping the cassette. Good enough. Same with vinyl, I never cared about vinyl, and the best vinyl and cassettes are mastered off digital anyway. I consider 320 kbps MP3 to be perceptually lossless, I digitized many of my CDs into this bitrate, but I am keeping my CDs