So I've got a question for you /g/ents. I have a .mp3 @ 320kbps and a .m4a @ 128kbps. They both "mostly" cut off at 16 khz except that the mp3 has some signal that goes up to 22 khz. Is the "blue stuff" just noise or is it highs that are part of the song? If it's just noise then the .m4a would be the preferable choice due to size right?
m4a
>>55306454
I don't know anything specific about the mp3 standard, but I am aware that it takes into account human psychoacoustics in effort to eliminate data that the human auditory system can't detect (frequencies higher than ~22kHz, spectra with more timbre than we can distinguish, etc.)
Something tells me that mp3 blue stuff is supposed to be there, and provides potentially better fidelity than the m4a.
Then again m4a is newer, so the standard may have been designed with a better understanding of human psychoacoustics that certainly has developed over the years.
An additional observation, it seems the extra spectra in the mp3 graph turns on and off when needed. Invariably these sections take more information to encode, so perhaps this behavior is an attempt to keep file size low, but keep fidelity high.
AAC looks like it was probably made from the MP3, so while that MP3 is shit and doesn't look to be originally 320k, it's probably better.
But use your ears.
It looks like you can pull better off youtube. What the hell did you do? Use a shitty youtube to MP3 site and then convert that to AAC?
Alright let me rephrase the question. In the .mp3 spectograph, is the stuff above 16 khz noise? Yes or no? Thanks.
>>55306560
>I don't know anything specific about the mp3 standard, but I am aware that it takes into account human psychoacoustics in effort to eliminate data that the human auditory system can't detect
It's much smarter than that and perceptual audio coding is the basic principle for every single lossy encoder. With all of them you eventually hit a point in the bitrate where you start to hear degradation in form of artifacts due to the quantizer failing, generally only in high frequencies. Where this point is and how it manifests itself is dependent on the content being encoded as well as the encoder used aside from the mere bitrate. mp3 and AAC both have multiple encoders - some of them terrible.
>(frequencies higher than ~22kHz
Humans can't hear quite that high. Nyquist frequency is exactly half of the sampling rate(44.1 KHz) which is why that spectra cuts at 22.05 KHz. mp3 is being a bit silly here. There's absolutely no need to encode anything so high if you aim for compression efficiency.
>>55306454
>Is the "blue stuff" just noise or is it highs that are part of the song?
Likely harmonics and very faint ones at that. You shouldn't pay much attention to what happens above 16 KHz. It's nice to have bandwidth there but frequencies that high are mostly masked away by the content present in more sensitive regions of our hearing. This is especially true for full band stereo music. Check out auditory masking and equal loudness contour.
You could just ABX those files and check if you hear any difference. If you do, pick the one that sounds the best or alternatively, get a better source and re-encode it yourself properly. That'll likely yield the best results.
AAC is more compression efficient than mp3, especially at low-ish bitrates such as 128 kbps. However in this case, when judging solely based on the codec alone could do, 320 kbps mp3 is much better than the AAC. It also looks better on the spectra but looks here can be deceiving.
>>55306934
Yes, it's garbage.
>>55306934
sound is just noise
btw, that is a really shitty encode
>>55306454
>not using superior in every way Opus
>year of our lord twenty sixteen
Come on now
Jesus Christ this spam filter.
There you go OP. Learn to youtube-dl.
>>55306560
The issue some people have with the approach mp3 takes is that those "unheard" frequencies create harmonics that are hearable, and overall the exclusion of them affects the final reproduction.
In simple words, master sources are inheently "colored" by their instruments, and a bandwidth limiting format takes away the natural foloring which may well be and often is intended to be heard.
Also mp3 is not as efficient per kbit than other formats, so theres no point to use it anymore.
>>55307029
>have to resample all your music because it doesn't support 44.1
LOL. Opus is trash.
>>55306962
Alright, so if we were talking about a higher quality sample such as this one. Would there be any noticeable difference if I cut this off at say 25khz?
>>55307141
Just encode everything QAAC q127.
>>55307141
Absolutely none unless the resampler somehow manages to effect the parts in the auditory band. It shouldn't, ever.
I'd say you could safely cut that from 19 KHz and hear absolutely no difference, even if you just low pass filter it from 18 KHz and listen only for the remaining 1 KHz high treble band. Refer to that part about auditory masking and equal loudness contour and note what is going on in that spectra within that range.
Human hearing goes up to 20 KHz but degrades with age. You can test your hearing easily by getting a sine sweep which goes up to 20 KHz and this is assuming that your loudspeakers/headphones pass band extends that high. If you are above 20 years old, be glad if your hearing exceeds 18 KHz.
>>55307122
While resampling is essentially a lossy process, this is hardly harmful. If you understand anything about sampling, think about what happens if you upsample.
> Uniornically caring about any frequency above the 6k hz
Why do you guys do this? most music is within the 60-6k hz range anyways. Your ears can't hear.
>>55307267
High pass filter*
>>55307267
Why sample at all when you can use a codec like Vorbis and avoid resampling altogether?
>>55307331
Opus is more compression efficient than Vorbis and the limited sampling rate options simplify the codec, helping out with latency. The 1.95 KHz of extra bandwidth also help a bit with DA-converter anti-aliasing brickwall filters before Nyquist. I guess the Opus developers saw that as the more important factor than resampling.
>>55306454
How do you people screw up mp3s so badly? Also stop using CBR. Even the LAME developers stated that it's deprecated.
>>55307429
MP3 is depreciated, so duh.
>>55307494
(You) are depreciated, so duh.
