From what I understand, CDs are 44.1kHz and DVDs are 48kHz.
- Without up or down sampling, shouldn't the MP3 files that are created from the disc also be 44.1kHz?
- If I have a 48kHz MP3 file, does that mean that someone, somewhere down the line up/down sampled the file?
- Are other CD formats such as SACD or HQCD II 48kHz?
1 Yes
2 MP3 can go up to 48kHz http://en.wikipedia.org/wiki/Comparison_of...chnical_details (http://en.wikipedia.org/wiki/Comparison_of_audio_formats#Technical_details). If the source was a CD so yes, it was upsampled but if the source was 48kHz then the MP3 hasn't been resampled.
3 More info here: http://en.wikipedia.org/wiki/Compact_Disc_..._Audio#Bit_rate (http://en.wikipedia.org/wiki/Compact_Disc_Digital_Audio#Bit_rate) and http://en.wikipedia.org/wiki/Super_Audio_CD#Technology (http://en.wikipedia.org/wiki/Super_Audio_CD#Technology)
I'm not an expert here, this is all I can tell you and still may be wrong. I am sure someone will reply with a better explanation.
48k MP3s are usually either a DVD video rip or (more likely) just some idiot resampling because 'bigger is better'.
hmm...... thanks for the links. kinda confirms what i had preconceived
If the source file is a CD, then 48 kHz is of course resampled.
However CDs aren't the only sources out there. Just about every source except for CD is 48 khz or a multiple of that (96kHz). DVDs, DATs, and nearly every computer sound card is at 48 kHz.
All recordings or transfers from analog that I do personally, I do at 48 kHz, so I actually have quite a bit of legitimate 48 kHz audio in my library.
48k MP3s are usually either a DVD video rip or (more likely) just some idiot resampling because 'bigger is better'.
Not so long ago, many PCs/laptops supported playback only at 48k. Using 48k files, resampling is done only once (saving CPU); also it may be easier to control resampling quality settings this way.
And since sound >22.05k can't be perceived, logically at least, a perceptually coded 48k file shouldn't be any bigger than the corresponding 44.1k.
if i mainly use my computer for music, should i set my sound card to output at 41kHz or 48kHz?
Certainly not 41Khz.
For listening only, this is definitely something you shouldn't worry about. Someone more technically-savyy (I'm not one of them) might tell you why, but IMHO besides the nyquist theorem, common sense dictates there's no reason why you should lose sleep over that for this particular scenario - hence my avatar.
Edit: nyquist theorem added up.
Back in the days of DAT I thought it was odd that that 48Khz was considered a professional master stage before being released on 44.1 considering a second brick wall alias filter stage is then required, and there is a lot of interest in the effect of steep filtering, isn't that more of a detriment than a benefit. Anyone got the reasoning on 48.
Probably something like "bigger is better" followed by a lot of mouth breathing sounds.
Anyone got the reasoning on 48.
Back in the day, it struck the right balance between cost, quality, and storage/band-width for professional audio equipment. To have used 44.1 would have pushed the price up, or the quality down.
When did they start using 48, really?
besides the nyquist theorem, common sense dictates there's no reason why you should lose sleep over that for this particular scenario
Neither is relevant. What are important are the human auditory system (http://en.wikipedia.org/wiki/Auditory_system), and the resulting equal loudness contour (http://en.wikipedia.org/wiki/Equal-loudness_contour). The Nyquist sampling theorem (http://en.wikipedia.org/wiki/Nyquist–Shannon_sampling_theorem) is far more general information theory and is, like the name suggests, about sampling of quantities, rather than hearing. And common sense is hard to apply to something like the limit of your senses. A good way to make informed choices is to conduct experiments, and believe in empirical and statistical results.
Anyone got the reasoning on 48.
Back in the day, it struck the right balance between cost, quality, and storage/band-width for professional audio equipment. To have used 44.1 would have pushed the price up, or the quality down.
I believe that the choice of 48kHz was even more pragmatic. It is technically sufficient, economically viable, but more importantly it was hard to convert into 44100 in the early days (the more fanatic you are, the harder it becomes).
The DAT constraints seemed to be chosen more to limit CD copying than anything else.
-k
I believe that the choice of 48kHz was even more pragmatic. It is technically sufficient, economically viable, but more importantly it was hard to convert into 44100 in the early days (the more fanatic you are, the harder it becomes).
I assume you mean the other way around (copying a CD onto DAT), but DAT didn't hit the streets until 1987, and then the pros had been using 48k for quite some years already: http://en.wikipedia.org/wiki/Digital_Audio_Stationary_Head (http://en.wikipedia.org/wiki/Digital_Audio_Stationary_Head)
I've never found a reliable source to corroborate it, but I remember hearing somewhere that CDDA was originally going to use 48 000 Hz, but it was reduced to 44 100 Hz to increase the runtime from just under 68 minutes to 74 minutes as a way to fit a particular piece of classical music on CD (either by Bach or Beethoven, IIRC). As for why 48 000 and 44 100 were chosen, the numbers needed to be divisible by both 25 and 30, so that digital audio could be stored on tapes in both PAL and NTSC format, which was the cheapest medium to use before CDs became prevalent. As digital audio expanded from the realm of CDs to media with greater capacities (e.g. DVD), using the rounder 48 000 rate became practical, so 44 100 fell out of favour.
AFAIK, there was no significant use of 48k before Sony's DASH tape system (1982).
DASH won out commercially in the pro market over Mitsubishi's 50.4k system (which had launched in 1980) and the AES adopted Sony's 48k as the pro standard in 1985. By 1986, Mitsubishi had switched to using 48k.
There's some info and a video about DASH here: http://www.vintagerecorders.co.uk/sales/pcm3402/pcm3402.htm (http://www.vintagerecorders.co.uk/sales/pcm3402/pcm3402.htm)
I've never found a reliable source to corroborate it, but I remember hearing somewhere that CDDA was originally going to use 48 000 Hz, but it was reduced to 44 100 Hz to increase the runtime from just under 68 minutes to 74 minutes as a way to fit a particular piece of classical music on CD (either by Bach or Beethoven, IIRC).
Disputed. Sources pro et contra at http://en.wikipedia.org/wiki/Compact_Disc_...nd_playing_time (http://en.wikipedia.org/wiki/Compact_Disc_Digital_Audio#Storage_capacity_and_playing_time)
AFAIK, there was no significant use of 48k before Sony's DASH tape system (1982).
Which makes it even harder to understand why 48 came to prevail on the recording and processing side, as - from what I can read - all DASHes also supported 44.1.
Which makes it even harder to understand why 48 came to prevail on the recording and processing side, as - from what I can read - all DASHes also supported 44.1.
Phase distortion in the (necessary before oversampling came along) brick-wall filters was much less with 48k.
Phase distortion in the (necessary before oversampling came along) brick-wall filters was much less with 48k.
That does not explain why one would want to use 48 to record and process something that would be delivered in 44.1 before hitting the DAC.
Certainly not 41Khz.
hmm... cuz setting it to 44.1 will introduce more sample rate conversions, in turn decreasing quality?
He was just making fun of your typo.
Well, to my understanding mp3 is kind of pointless, when you rip a 48khz source...
if one wants better quality, one should make sure to have no elements in the chain, which are very weak... because a chain is just as strong as it's weakest element...
This really has nothing to do with the topic at hand, but there are plenty of legitimate reasons to use mp3. The source's sample rate really has nothing to do with it. 44.1 kHz, 48 kHz, 192 kHz, it makes no difference.
48k MP3s are usually either a DVD video rip or (more likely) just some idiot resampling because 'bigger is better'.
I resample to 48k before encoding mp3. The file isn't any bigger, mp3 file size is based upon bitrate.
I do it because if everything is at 48k then it doesn't need to be resampled upon playback.
Archival lossless I leave how it was, otherwise there is no point to it being lossless, but since sample rate doesn't impact the file size of lossy and the decoded wave form isn't going to be same anyway, resampling before I encode using SoX means it doesn't need a resample during playback on my PC to mix with other sounds (which often are resampled to 48k, I doubt my e-mail alert bloop thing is 48k)
Even speech originally at 22050 I up-sample.
Maybe I'm an idiot, maybe it really isn't necessary, but it isn't because 'bigger is better' - rather it is because same sample rate and bits per sample are needed to mix sounds so why not resample lossy first?
Opus btw resamples just about everything to 48K even if you don't first.
If you upsample the audio to 48 kHz before encoding it using a lossy encoder, and get/choose the same bitrate, the only possible outcome is that the quality is either the same or worse than the respective 44.1 kHz encode. I don't know what you hope to gain. It makes more sense to add the resampler to the DSP of the playback chain, unless you expect the DSP resampler to be audibly bad.
There is a reasonable argument that most of the "tuning" that has gone into the lame mp3 encoder was targeted at 44.1kHz. You could make arguments either way based on temporal resolution, spectral resolution, etc etc but I wouldn't worry about it.
lossy encoding decisions that seemed really important back when most people ripped only to lossy seem unimportant now you can re-generate your entire lossy collection from a lossless source with a couple of mouse clicks (and a little/lot of time!). I don't think I'd choose to slow that process down by adding a decent resampler, but it's up to you.
If someone gave me an mp3 at 48kHz "from a CD", I'd probably wonder if they were being too clever for their own good. There are far too many ways for people with only half the information to mess up something like. It's one of those red flags, like 320kbps mp3s, that makes you wonder. A simple AccurateRIP and standard lame -V2 encode look like someone thought about things enough to do the job properly, but didn't overthink it and potentially do something silly.
Cheers,
David.
If you upsample the audio to 48 kHz before encoding it using a lossy encoder, and get/choose the same bitrate, the only possible outcome is that the quality is either the same or worse than the respective 44.1 kHz encode. I don't know what you hope to gain. It makes more sense to add the resampler to the DSP of the playback chain, unless you expect the DSP resampler to be audibly bad.
I haven't done any ABX testing but it certainly doesn't sound worse to me. It still sounds transparent at 192 kbps.
What I hope to gain, well, if it gets resampled to 48khz after decoding for playback every time anyway then why not resample once. That way it doesn't need to be resampled after decoding. That's all I hope to gain, and it may seem pointless, but it's scripted anyway so it's not like it takes extra effort on my part.
I highly doubt there's is an audible difference. Resampling happens all the time and most people aren't even aware. Since you can perfectly construct an analog wave for all frequencies below half the sample rate, how would taking samples from the reconstruction of the analog wave be worse? I could see resampling to a lower rate losing some frequencies, but not resampling to a higher rate.
I suppose that's a different topic though.
(...) it may seem pointless,
It's your call but it is indeed!
Even more so in the long run, as compatibility issues might only emphasize that.
The issue is not the resampling, which should be audibly transparent, but how well the lossy encoder works at the different sample rates. Think about it - if the encoder has to encode more data points in the same size file, what is it omitting to make room for them?
I do it because if everything is at 48k then it doesn't need to be resampled upon playback.
Until you use a hardware mp3 player at least, which almost always resample 48k back to 44.1k.
Maybe I'm an idiot, maybe it really isn't necessary, but it isn't because 'bigger is better' - rather it is because same sample rate and bits per sample are needed to mix sounds so why not resample lossy first?
I don't know about idiot, but its not a great idea.
The issue is not the resampling, which should be audibly transparent, but how well the lossy encoder works at the different sample rates. Think about it - if the encoder has to encode more data points in the same size file, what is it omitting to make room for them?
I don't think that's the way lossy encoders work. I think they look at what can be omitted that we can't hear, then what can be omitted that we maybe can technically hear but typically don't in our mind (like the mother in law), etc. - so there are more samples in the PCM but at the same time there are more samples that can be thrown away.
The issue is not the resampling, which should be audibly transparent, but how well the lossy encoder works at the different sample rates. Think about it - if the encoder has to encode more data points in the same size file, what is it omitting to make room for them?
I don't think that's the way lossy encoders work.
No, he is right. If you increase the sampling rate, in effect you make each transform block a little shorter (since they are of fixed size). Hence bitrate will probably have to go up a little. The difference isn't huge though, so I think the bigger concern would be how well tuned the encoder is. FWIW, I think at such high bitrates, almost any possible settings will most likely be transparent.
I think the idea is misguided though because its about equally likely on average that 44.1k will be the native sampling rate as 48k once you consider various hardware devices out there. Worse, hardware devices are battery limited and tend to use low quality resampling, whereas PCs are more likely to have a high quality (transparent) resampler.
I think the idea is misguided though because its about equally likely on average that 44.1k will be the native sampling rate as 48k once you consider various hardware devices out there. Worse, hardware devices are battery limited and tend to use low quality resampling, whereas PCs are more likely to have a high quality (transparent) resampler.
Most hardware devices at this point are android or iOS type devices. Not sure what they do, but both do high definition video and I would suspect they do 48k natiively. I'll have to check.
Other than CD red book, does anything else even use 44.1 kHz? I suspect as CD goes away and digital distribution becomes the norm, 48 kHz for digital purchases will become the norm as well. I've already bought several digital downloads that were distributed as 16/48 and I don't buy a lot.
I think the idea is misguided though because its about equally likely on average that 44.1k will be the native sampling rate as 48k once you consider various hardware devices out there. Worse, hardware devices are battery limited and tend to use low quality resampling, whereas PCs are more likely to have a high quality (transparent) resampler.
Most hardware devices at this point are android or iOS type devices. Not sure what they do, but both do high definition video and I would suspect they do 48k natiively. I'll have to check.
Pretty much anything that can do gapless audio (iOS, Android) will have to resample.
Older devices used to run at native sampling rates. The Sandisk players do for instance, at least until you rockbox them, since rockbox is also gapless.
Other than CD red book, does anything else even use 44.1 kHz?
Other than the overwhelming majority of all music
I suspect as CD goes away and digital distribution becomes the norm, 48 kHz for digital purchases will become the norm as well. I've already bought several digital downloads that were distributed as 16/48 and I don't buy a lot.
No, digital distribution is almost entirely 44.1k as well. There is some high res stuff (e.g. 96k) but not much 48k. It wouldn't make sense, since the companies pushing digital distribution use 44.1k hardware devices to make money (e.g. Apple).
Actually it makes a lot of sense. The big players (iTunes, Amazon, Google) all push cloud players, where it is easy to migrate existing collections of purchased music to 48kHz. And they could even spin it as being better (though it's not)
For mastering, I'm not an audio engineer but I suspect going from 96kHz masters to 48kHz is easier resample because you are not creating any samples where they did not already exist.
And for pushing the hidef snake oil songs at a higher price, I'm guessing resampling for playback on your phone is easier at 48 kHz is easier for same reason.
44.1 is really only used for legacy reasons. It's easy to migrate existing purchases on their cloud and video is already 48 kHz and downsampling from 96 I suspect is easier and I suspect upsampling from 32000 is easier as well as many of the sample points fall at same place.
I very well could be wrong, but as far as I can tell, 44.1 kHz really was only used for CD media and legacy is only reason it is still in use with benefits to dropping that legacy by the digital distributors.
Another possibility...
If the audio was extracted from any kind of video file, the sample rate could easily be 48 kHz, depending upon the video file container and format and also the software used to extract the audio.
Since most video files information is lossy encoded, it's probably not high fidelity lossless audio (as a source) even if the source was a DVD (which uses MPEG for audio).
foobar2000 can open MKV files (if you type "*" into the windows explorer Open window). and then you can convert MKV video audio tracks to other audio formats. VLC media player has an audio extraction function. Oxelon Media Converter can extract the audio from videos. GoldWave Editor can extract audio from videos. AVANTI and some other FFMPEG front ends can probably do a lot of different types of audio extraction also. Also MKV Extract can extract individual components of MKV videos. You might think that MKV's aren't very abundant, but WebM videos can be thought of as special MKV files with OGG audio and VP video. And as a container format a lot of snafus of video conversion can be worked around by converting the video to and from Matroska (MKV) during the conversion process manually.
Anyways, hopefully, you get the idea.
Actually it makes a lot of sense. The big players (iTunes, Amazon, Google) all push cloud players, where it is easy to migrate existing collections of purchased music to 48kHz.
This makes zero sense.
I very well could be wrong,
Yes, you are wrong. There is no advantage to downsampling to any particular frequency like you seem to be assuming.
Anyway, your misconceptions aside, there is no reason to convert your mp3s. Its probably not going to result in audible loss of quality, but its not going to help either and it could certainly hurt quality and reduce battery life.
Actually it makes a lot of sense. The big players (iTunes, Amazon, Google) all push cloud players, where it is easy to migrate existing collections of purchased music to 48kHz.
This makes zero sense.
I very well could be wrong,
Yes, you are wrong. There is no advantage to downsampling to any particular frequency like you seem to be assuming.
Anyway, your misconceptions aside, there is no reason to convert your mp3s. Its probably not going to result in audible loss of quality, but its not going to help either and it could certainly hurt quality and reduce battery life.
Well I assume that downsampling from 96 kHz to 48 kHz by the studios for digital distribution would have the advantage of smaller lossless for audiophile sale and allow them to charge more the 96 kHz version. Like they do now when they downsample to 44.1 for digital distribution. Only 48 is exactly one half of 96, 44.1 isn't an integer fraction of 96 so it requires new sample points.
For mastering, I'm not an audio engineer but I suspect going from 96kHz masters to 48kHz is easier resample because you are not creating any samples where they did not already exist.
Like they do now when they downsample to 44.1 for digital distribution. Only 48 is exactly one half of 96, 44.1 isn't an integer fraction of 96 so it requires new sample points.
Your suspicions cannot replace actual knowledge of resampling. If you have any knowledge of high quality resampling algorithms providing inadequate results in the circumstances you mention, please create a thread about them.
Well I assume that downsampling from 96 kHz to 48 kHz by the studios for digital distribution would have the advantage of smaller lossless for audiophile sale and allow them to charge more the 96 kHz version. Like they do now when they downsample to 44.1 for digital distribution. Only 48 is exactly one half of 96, 44.1 isn't an integer fraction of 96 so it requires new sample points.
For the second time: this is nonsense.
Read up on resampling, you have no idea what you're talking about.
Well I assume that downsampling from 96 kHz to 48 kHz by the studios for digital distribution would have the advantage of smaller lossless for audiophile sale and allow them to charge more the 96 kHz version. Like they do now when they downsample to 44.1 for digital distribution. Only 48 is exactly one half of 96, 44.1 isn't an integer fraction of 96 so it requires new sample points.
For the second time: this is nonsense.
Read up on resampling, you have no idea what you're talking about.
This is interesting -
``Note that it's generally preferable for a decoder to output at 48kHz even when you know the original input was 44.1kHz, not only because you can skip resampling but also because many inexpensive audio interfaces have poor quality output for 44.1k.''
https://wiki.xiph.org/OpusFAQ (https://wiki.xiph.org/OpusFAQ)
Opus can only operate at 48kHz, so you must use 48k. That FAQ is telling you that you shouldn't needlessly resample audio files, which is actually what people in this thread are trying to explain to you...
48k MP3s are usually either a DVD video rip or (more likely) just some idiot resampling because 'bigger is better'.
I was joking when I said this, but someone volunteered to prove me right anyway.
48k MP3s are usually either a DVD video rip or (more likely) just some idiot resampling because 'bigger is better'.
I was joking when I said this, but someone volunteered to prove me right anyway.
Well thank you.
My PC where I listen to my music most of the time resamples 44.1 to 48 anyway so I guess doing it once instead of every time makes me an idiot.
Yes I can configure it (pulseaudio) but then video has to be downsampled to mix with other sounds. As I rarely play physical CDs but I do play movies, 48000 makes sense.
Wait your hardware supports 44.1k ? You don't need to resample at all then. Like I said bigger isn't better.
Wait your hardware supports 44.1k ? You don't need to resample at all then. Like I said bigger isn't better.
But it isn't bigger. For lossless I don't resample. With lossy I do, but lossy size isn't determined by sample rate, but to allow mixing of different sound sources together they all need to be same sample rate and bit depth so anything that it isn't the sample rate and bith depth the sound server expects gets resampled.
-=-
I should add that historically pulseaudio resampling was not very well optimized, used a lot of CPU. It probably is better now but I haven't checked.
Yes I can configure it (pulseaudio) but then video has to be downsampled to mix with other sounds. As I rarely play physical CDs but I do play movies, 48000 makes sense.
That is true indeed. Did you consider configuring alternate-sample-rate in your daemon.conf? That way PA would pick either 44.1kHz or 48kHz, whatever starts playing first, and resample the streams which come later.
they all need to be same sample rate and bit depth so anything that it isn't the sample rate and bith depth the sound server expects gets resampled.
Bit depth is not important in resampling decision.
I should add that historically pulseaudio resampling was not very well optimized, used a lot of CPU. It probably is better now but I haven't checked.
PA is using standard resampling libraries (speex or libsamplerate), just like majority of other audio softwares. Look at http://blog.ivitera.com/pavel/linux-audio/...ing-libsoxr-lsr (http://blog.ivitera.com/pavel/linux-audio/pulseaudio-with-ld_preloading-libsoxr-lsr) for integration of the great libsoxr library.
PA is using standard resampling libraries (speex or libsamplerate), just like majority of other audio softwares. Look at http://blog.ivitera.com/pavel/linux-audio/...ing-libsoxr-lsr (http://blog.ivitera.com/pavel/linux-audio/pulseaudio-with-ld_preloading-libsoxr-lsr) for integration of the great libsoxr library.
I've found that using src-sinc-best-quality (from libsamplerate) provides good enough sound quality that I can't hear the difference between a resampled and a non-resampled stream.
And with the alternate-sample-rate option set to 48000hz, as long as I only have my media player or music player running at the same time, there's no need for resampling anyway. I used to be a big Pulseaudio opponent, but I have to admit that it actually works really well now.
If only something could be done about Gstreamer, though...
Libsamplerate output is very good, only its CPU consumption for best-quality poses a real problem on less-powered setups.
But it isn't bigger. For lossless I don't resample. With lossy I do, but lossy size isn't determined by sample rate
For lossy, size is dictated by bitrate and duration only.
However, at a given bitrate, the quality might be improved by
reducing the sample rate. This is fairly easily to hear by playing around with lame in CBR mode at low-ish bitrates (64kbps, 80kbps, maybe 96kbps - all using joint stereo), and comparing the results with a the sample rate of 44.1kHz vs 32kHz.
Please try this. It will show you that your statement is wrong.
Even so, no one can't extrapolate this result to claim an audible difference between 44.1kHZ and 48kHz at high-ish bitrates.
Cheers,
David.