Yes, I found this rather confusing at the start of this video, saying that MIDI has a "quality"! As you no doubt know, it stands for Musical Instrument Digital Interface.
Indeed. MIDI is just a code for the MIDI-compatible hardware to play back using predefined samples on the hardware itself, rather than actual recorded audio, which uses unique samples encoded in the media file. One minute of MIDI is only a few kilobytes, but the audio clarity is on par with CD-audio and can potentially be even better depending on the hardware a MIDI file is played back on.
What I understood from the video :- 1. Humans can't hear every frequency significantly due to vibrating restrictions in the membrane inside of our ears. 2. So we delete this information from the audio file to reduce it's size. 3. We have become better at doing step 2 (?)
Someday you should do a video about the software patent clusterfuck. Audio compression is just one area where software patents created a huge mess. Granted it's an area of law rather than computer science, but it has had a major (mostly negative) affect on software development.
One time, my cousin was bragging about how he converted his songs to a totally lossless compression. He went on at some length about how he could only fit one album on his entire music player because the files were so big. I just nodded and told him, "You might as well have not bothered if you're listening to the music on those crap headphones." I feel bad now for ruining his experience.
I remember quite vividly my earlier online adventures when downloading MIDI files were a completely relevant way of downloading, usually video game music for me at least. Since the MIDIS were little more than "sheet music" in digital form, it allowed you to modify which instruments were played for each track. A MIDI played on one computer could sound completely different on another depending on the sound card was present on the machine. Early MP3s sounded tinny and would be scoffed at in regards to the audio quality by the standards expected today. It wasn't unusual to see MP3s shared at 96-128kbps, which is dreadful, but was necessary to accommodate the limited space available on early MP3 players limited storage - which could be anywhere from 32-64MB. We'd eventually see CD players that would support the MP3 format, which would allow you to store hundreds of megabytes of MP3s on a compact disc. At a time when internet transfer rates averaged 3-5KB/sec and and data storage options were limited in the MB range, MP3s really were a godsend.
As somebody with a background in software dev and IT, this really didn't help at all. It was a very long version of "it doesn't encode the bits you can't hear." I suppose if I really cared I could just research it properly, but I was hoping for a little better laymans description. OK, so, say I have a recording at X Hz and 16 bits per sample for 1 second then I have X*16 bits of data. The way he described it it sounded like uses something like a fourier transform to break into multiple tracks of variable Hz and variable bits. However, even if you broke it down into as little as 5 bands then combined they couldn't have more than a bitrate of (X * 16) / 5, and no single band could be close to the original resolution before the file balloons to a much bigger size. There's a huge gap in how, exactly, separating the frequencies into multiple tracks of differing resolution can then be stored in a way that doesn't produce a larger file. That make sense?
Would have been nice to have more visual demonstration of how compression affects the sound and why you get artifacts. Maybe talk more specifically about LAME and AAC type codecs. This video could have easily been condensed into 2 minutes.
The term 'audio compression' has different meanings for audio engineers and computer scientists so the title could be misleading and/or confusing for some.
Given that the channel is called Computerphile I would guess they don't mean dynamic range compression. While different things the two are related though (i.e. DRC is used in digital telephony to reduce data in transmission )
It was a bit short. I would have liked for him to go more in depth on the different rates of compression for different frequency ranges. Nonetheless still fun to watch :)
Huffman and Morse are actually quite the oppsite: Huffman is prefix code, while Morse needs pauses do know where a letter ends. Huffman is more efficient.
I assume a mp3 encoder just simply removes bits here and there based on the audio pattern and it may also searches for stereo parts into the audio and remove all redundant parts of it, the decoder can then copy paste those missing parts back in from the remaining channel, I don’t think that mp3 encoders use any sound analesis to figure out wich sounds we could hear better then other ones because that’s sound really complicated.
Here's a fun experiment: Find an MP3 that you ripped in iTunes 10 years ago. re-rip it to the same bitrate using a modern version of Lame. Compare the old mp3, new mp3 and CD. I'll bet you a million dollars that, with appropriate blinding, the old MP3 will sound like garbage but the new one will be indistinguishable from the CD. That will demonstrate how far digital compression has come. It will also show you how...lets be charitable and say "encoding-speed-focused" iTunes was when iPods were huge.
The information given about midi and mod files is pretty incorrect, actually. Midi files are first and foremost about sequencing music. They basically tell a computer when a note, and what note, is being played, and by what instrument. Later, a standard called general midi almost every computer a standardized set of instruments, and a minimum amount of playable notes at a time-24 to be exact-that allowed, but wasn't made _specifically_ for sharing music between computers and keyboards. Mod files are essentially a whole different ballpark. They're not sequenced like midi-they're usually made in piece of software called a tracker (such as FastTracker, OpenMPT, etc.). (Though, tbh, I don't know all the subtle differences between mod files and midi.)
I totally forgot about mod files. I remember finding some mod files on a CD borrowed from someone that won the CD on a radio show. It had some demos on it and a dos player for mod files. It was mesmerizing. At the time i was playing Duke Nukem and Descent and the CRT monitor had that funky new PC hardware smell, probably very toxic. We upgraded the PC to 16MB of RAM from 8MB so we could play a demo of MechWarrior. 2 :)
midi is not even a true audio format.. it's the note's for the computer to play.. you could compress the midi with pkzip quite good.. not sure if he knows what he's talking about..
I don't agree with the argument that today we are less interested in bandwidth. If Spotify could save could safe also 5% on bandwidth, their server costs would reduce enough to be an interesting option to explore.
We are less interested in bandwidth though. In the 90's when the average user had a 28Kbps (you would get roughly 3 KB/s ) and the premitive mp3 players had something like 32MB of storage you wanted to reduce size as much as you could.
module files have become quite complex indeed. Many different file formats allowing more and more complex channels. Look up Fear Of Dark - Motorway for an idea on what is possible
Is aptx worth it to go out of the way to buy supporting hardware (phone, headphones, bluetooth adapter, etc.)? Of-course assuming good source quality and hi-fi headphones.
Can we get a discussion of the lowest size-to-quality ratio formats, and what their strengths and weaknesses are? I'm trying to decide which format would be the best to covert my 10,000 or so song music library to that will give me greatest compatibility between devices, flexibility of editing, smallest size on disk or devices, and best overall dynamic quality, and I'd love to see somebody very knowledgeable walk me through that. Is that a super selfish video request? Well...but I'd really appreciate it!
A very unfair description of MOD music which was a genre that was a strong influence on modern music and was not meant to be an emulation of "real audio". Look up the original sound track of games like Star Control 2.
I came here to see about digital audio compression, but you're teaching me how ears work?? It's very unneeded information and I already know it, just tell me how you compress the audio files.
MIDI files represent a score that plays from a pre-determined set of instruments, MOD has a score that plays from a set of samples. They both are score notations rather than digitised audio so are practically the same in that respect and in the terms of how Dr. Mitchell referred to them. If you're going to nitpick, at least be accurate.
Yeah, and music score on paper in front of a conductor is basically the same as well.. You know.. the same as in a Word document and a Emacs document are also interchangeable..
Do you really think the Dr. was talking about the small minutiae of the file format or talking broadly, to a lay audience about the generalities of the way the music is stored? I'll give you a hint, this was a rhetorical question.
Diggnuts It appears you're a bit on the slow side. So I'll try one last time to elucidate it for you. MIDI and MOD are to vector graphics as mp3 is to bitmaps. So you understand yet you obtuse git?
Especially insulting to ignore the fact tracker music was almost a genre unto itself. This video is not up to par with the usual standard of Computerphile.
Check out the audio comparison between 16kbit/s MP3 vs. Opus to hear how much audio compression has advanced: en.wikipedia.org/wiki/Opus_(audio_format)#Quality_comparison_and_low-latency_performance
It's amazing how MP3 (with the amazing LAME codec) can rival much newer formats just because of better psy models, kinda like x264 rivals H.265 codecs for the same reasons. Still, this particular video was _LAME_ for different reasons...
Well, that was awful, even at 2x speed. What content compression are you using? Where's all the fine details? Are you dumbing down for the sheeple that won't watch it in the first place?
@matsv201: not really. I was paranoid it would get overly specific, ex: talking about things specific to MP3 and Vorbis while totally ignoring things like ADPCM (IOW: what most talk about 'audio compression' amounts to...). ... it didn't even get this far ... FWIW: I could apply those descriptions of psychoacoustic modeling and similar to an encoder I have for a customized ADPCM variant... (which is technically almost completely different than something like MP3; and technically if-anything borrows more directly from FLAC). (ADD: though, like other ADPCM's, used fixed-form bit-packing rather than Huffman; mostly because of speed and also because the format needs to be strictly CBR for the use-case anyways; so some additional space-saving may be done externally via Deflate or similar...).
Well.. mod fiiles is really as good as you make them. Yea sure, the original mod files was 8 bit, some later was 16. But because its one isolated sound, it really don´t have that much of a benefit having 16 bit.
Same thing with MIDI, because it's really just a set of instructions rather than an audio compression technique. Case in point: modern movie MIDI soundtracks.
Well.. the quality of the midi is of cause of the data it use to play it and quality of the sample in the hardware. This is quite obvious in the Super Nintendo having rather high quality sample in the ROM of the console. (it can also address extra sample on board the game ROM., so it works like a hybrid between a MIDI and MOD. Looking at the highest quality late module files. They are pretty much CD quality with files in 100-250k range. Even have some where they extracted the vocals and run the lyrics and chorus as different samples repeating and piching them up and down... still making it sound pretty much identical to the original song. Some of the early/Mid 90-tys popular songs was actually made that way to begin with. E-type used normal tracker tools for most of his early stuff
"Fundamental Tennents" - that's the brand of lager that makes you say Tennents (or tenants) instead of tenets? Joking about this slip aside, this is, sadly, nowhere near up to your usual standard.
This video tries to do too much in a single video and ends up really shallow and imprecise in how it covers the material as a result. Also the guy seems to be my age, but doesn't give the impression he was very involved with computers before ~2000.
Isn't it just wrong to say that MIDI has "lower quality" than other formats?
MIDI is playback data, not the same as an audio RECORDING.
Yes, I found this rather confusing at the start of this video, saying that MIDI has a "quality"! As you no doubt know, it stands for Musical Instrument Digital Interface.
Yes MIDI is just the "Score" - which notes and when/how long to play them. Nothing to to with quality, that depends on your audio card, synth etc.
Indeed. MIDI is just a code for the MIDI-compatible hardware to play back using predefined samples on the hardware itself, rather than actual recorded audio, which uses unique samples encoded in the media file.
One minute of MIDI is only a few kilobytes, but the audio clarity is on par with CD-audio and can potentially be even better depending on the hardware a MIDI file is played back on.
Sometimes computerphile does incredible videos. Then there are videos like this, which literally tell me nothing.
Now that was vague. Why not do a video like the ones you did for the JPEG codec, those were really nice
I also had exactly the jpeg compression video in mind when i read the title. i was very disapointed.
The mp3 file synthesis also has a lot of hidden features! Perhaps bring up the topic of filterbanks or wavelets for that matter
A better way to phrase this, instead of whining, would be "Loved the video, would be awesome to have a more in depth one as well!"
The videos all uphold a very high standard. There is nothing wrong with proposing future improvements. Don't create a problem where there isn't one
What I understood from the video :-
1. Humans can't hear every frequency significantly due to vibrating restrictions in the membrane inside of our ears.
2. So we delete this information from the audio file to reduce it's size.
3. We have become better at doing step 2 (?)
Someday you should do a video about the software patent clusterfuck. Audio compression is just one area where software patents created a huge mess.
Granted it's an area of law rather than computer science, but it has had a major (mostly negative) affect on software development.
hey, mod files aren't... uh.. well, close enough I guess
... and midi files weren't ... oh well ... :D
Indeed. Describing the music in Star Control too as a poor emulation of "real audio"...
One time, my cousin was bragging about how he converted his songs to a totally lossless compression. He went on at some length about how he could only fit one album on his entire music player because the files were so big. I just nodded and told him, "You might as well have not bothered if you're listening to the music on those crap headphones." I feel bad now for ruining his experience.
I remember quite vividly my earlier online adventures when downloading MIDI files were a completely relevant way of downloading, usually video game music for me at least. Since the MIDIS were little more than "sheet music" in digital form, it allowed you to modify which instruments were played for each track. A MIDI played on one computer could sound completely different on another depending on the sound card was present on the machine.
Early MP3s sounded tinny and would be scoffed at in regards to the audio quality by the standards expected today. It wasn't unusual to see MP3s shared at 96-128kbps, which is dreadful, but was necessary to accommodate the limited space available on early MP3 players limited storage - which could be anywhere from 32-64MB.
We'd eventually see CD players that would support the MP3 format, which would allow you to store hundreds of megabytes of MP3s on a compact disc.
At a time when internet transfer rates averaged 3-5KB/sec and and data storage options were limited in the MB range, MP3s really were a godsend.
This video was VERY vague in information
I require a detailed explanation of every aspect of digital audio compression, you have 7 minutes and the video must be engaging and easy to follow.
it was sufficient xD
i want more details too btw :) hope they do more of these vids.. i think "how we hear" is an interesting topic
kapa1611 that video should serve as a really broad intro for a more in depth series then
this video was NOT very engaging, to say the least.
As somebody with a background in software dev and IT, this really didn't help at all. It was a very long version of "it doesn't encode the bits you can't hear." I suppose if I really cared I could just research it properly, but I was hoping for a little better laymans description.
OK, so, say I have a recording at X Hz and 16 bits per sample for 1 second then I have X*16 bits of data. The way he described it it sounded like uses something like a fourier transform to break into multiple tracks of variable Hz and variable bits. However, even if you broke it down into as little as 5 bands then combined they couldn't have more than a bitrate of (X * 16) / 5, and no single band could be close to the original resolution before the file balloons to a much bigger size.
There's a huge gap in how, exactly, separating the frequencies into multiple tracks of differing resolution can then be stored in a way that doesn't produce a larger file.
That make sense?
Would have been nice to have more visual demonstration of how compression affects the sound and why you get artifacts. Maybe talk more specifically about LAME and AAC type codecs. This video could have easily been condensed into 2 minutes.
The term 'audio compression' has different meanings for audio engineers and computer scientists so the title could be misleading and/or confusing for some.
I think they changed it to "digital" audio compression.
Given that the channel is called Computerphile I would guess they don't mean dynamic range compression.
While different things the two are related though (i.e. DRC is used in digital telephony to reduce data in transmission )
someone who mixes music. in the studio in live at music venues to ay a few. audio engineer is a broad term though
not nessesarily. compression can be used for tone rather than controlling dynamics.
Automation is sometimes better for leveling out a performance
The best thing about the MOD format is anyone can make samples and easily transfer their work.
The variety of samples to use was never ending.
Yeah i agree, this is a more popular level info dump than typical deep dive we expect from computerphile
I never thought about how the "neighborhood" frequency would affect the ear that way, thanks for a great way to explain it! :D
You forgot about the most important part of audio compression, the discrete cosine transform, also used in jpeg compression
Tenets. Tenants are people who reside in a rented building.
It was a bit short. I would have liked for him to go more in depth on the different rates of compression for different frequency ranges. Nonetheless still fun to watch :)
that topic fascinates me. Woukd love to see more videos about it
why were there no illustrations when he talked about the neighbouring frequency amplitudes and band filtering?
I hope you make a followup about Digital Video Compression!!
I'll come back later and stoned to see his clothes moire
you should really be wearing moire clothes in your profile picture
Interesting topic, needs more love than it got from this short video!
A Computerphile view of Opus Codec would be fantastic!!!
Mod is art.
Huffman and Morse are actually quite the oppsite: Huffman is prefix code, while Morse needs pauses do know where a letter ends. Huffman is more efficient.
yes, but in a way they are similar. just like huffman encoding, morse uses less signals for frequently used letters.
I miss my Windows 2000 two minute MIDIs.
not my time, but yeah of course. that must have been the sh*t back then.
Doom E1M1
I assume a mp3 encoder just simply removes bits here and there based on the audio pattern and it may also searches for stereo parts into the audio and remove all redundant parts of it, the decoder can then copy paste those missing parts back in from the remaining channel,
I don’t think that mp3 encoders use any sound analesis to figure out wich sounds we could hear better then other ones because that’s sound really complicated.
Poor description.
For the love of Benny Hill please use a tripod.
If the video is stabilized then RUclips will also compress it better.
So using a tripod is also more "green" :-)
i watched the whole video and the part that i'm most interested is the audio analytic peugeot picture on the wall
Once again, great video! Thanks for sharing :)
Modern codecs such as Opus are so well done that a 48Kbps opus encoded piece of audio could realistically be streamed over dialup.
Here's a fun experiment: Find an MP3 that you ripped in iTunes 10 years ago. re-rip it to the same bitrate using a modern version of Lame. Compare the old mp3, new mp3 and CD. I'll bet you a million dollars that, with appropriate blinding, the old MP3 will sound like garbage but the new one will be indistinguishable from the CD. That will demonstrate how far digital compression has come. It will also show you how...lets be charitable and say "encoding-speed-focused" iTunes was when iPods were huge.
The information given about midi and mod files is pretty incorrect, actually.
Midi files are first and foremost about sequencing music. They basically tell a computer when a note, and what note, is being played, and by what instrument. Later, a standard called general midi almost every computer a standardized set of instruments, and a minimum amount of playable notes at a time-24 to be exact-that allowed, but wasn't made _specifically_ for sharing music between computers and keyboards.
Mod files are essentially a whole different ballpark. They're not sequenced like midi-they're usually made in piece of software called a tracker (such as FastTracker, OpenMPT, etc.). (Though, tbh, I don't know all the subtle differences between mod files and midi.)
I totally forgot about mod files. I remember finding some mod files on a CD borrowed from someone that won the CD on a radio show. It had some demos on it and a dos player for mod files. It was mesmerizing. At the time i was playing Duke Nukem and Descent and the CRT monitor had that funky new PC hardware smell, probably very toxic. We upgraded the PC to 16MB of RAM from 8MB so we could play a demo of MechWarrior. 2 :)
midi is not even a true audio format.. it's the note's for the computer to play.. you could compress the midi with pkzip quite good.. not sure if he knows what he's talking about..
I don't agree with the argument that today we are less interested in bandwidth. If Spotify could save could safe also 5% on bandwidth, their server costs would reduce enough to be an interesting option to explore.
We are less interested in bandwidth though.
In the 90's when the average user had a 28Kbps (you would get roughly 3 KB/s ) and the premitive mp3 players had something like 32MB of storage you wanted to reduce size as much as you could.
Of course it was much more crucial back in the days.
Today is more an economical motivation than something that was a must
Agree about a more in depth video. The interviewers questions were too simplistic
module files have become quite complex indeed. Many different file formats allowing more and more complex channels.
Look up Fear Of Dark - Motorway for an idea on what is possible
Is that the most interesting background you could find to talk to him?
Is aptx worth it to go out of the way to buy supporting hardware (phone, headphones, bluetooth adapter, etc.)? Of-course assuming good source quality and hi-fi headphones.
Can we get a discussion of the lowest size-to-quality ratio formats, and what their strengths and weaknesses are? I'm trying to decide which format would be the best to covert my 10,000 or so song music library to that will give me greatest compatibility between devices, flexibility of editing, smallest size on disk or devices, and best overall dynamic quality, and I'd love to see somebody very knowledgeable walk me through that. Is that a super selfish video request? Well...but I'd really appreciate it!
Did you find an answer?
3:00 does this mean that the louder you hear the music the bigger is the effect?
I'd like to know what the current research trend is in this area. like audio codec in VoIP. Is there any research trends in this field?
A very unfair description of MOD music which was a genre that was a strong influence on modern music and was not meant to be an emulation of "real audio".
Look up the original sound track of games like Star Control 2.
exactly what was the pen and paper for
Just to give it a proper name instead of "psychoacoustic bits", what he is talking about is "Auditory Masking"
I love how midis have made an ironic comeback in some ways.
More on digital audio please!
I came here to see about digital audio compression, but you're teaching me how ears work?? It's very unneeded information and I already know it, just tell me how you compress the audio files.
Tracker formats like .mod have nothing to do with MIDI what so ever. Nothing at all.
MIDI files represent a score that plays from a pre-determined set of instruments, MOD has a score that plays from a set of samples. They both are score notations rather than digitised audio so are practically the same in that respect and in the terms of how Dr. Mitchell referred to them. If you're going to nitpick, at least be accurate.
Yeah, and music score on paper in front of a conductor is basically the same as well.. You know.. the same as in a Word document and a Emacs document are also interchangeable..
Do you really think the Dr. was talking about the small minutiae of the file format or talking broadly, to a lay audience about the generalities of the way the music is stored? I'll give you a hint, this was a rhetorical question.
Diggnuts It appears you're a bit on the slow side. So I'll try one last time to elucidate it for you. MIDI and MOD are to vector graphics as mp3 is to bitmaps. So you understand yet you obtuse git?
Especially insulting to ignore the fact tracker music was almost a genre unto itself. This video is not up to par with the usual standard of Computerphile.
Check out the audio comparison between 16kbit/s MP3 vs. Opus to hear how much audio compression has advanced: en.wikipedia.org/wiki/Opus_(audio_format)#Quality_comparison_and_low-latency_performance
I once downloaded a 30 minute Beethoven midi. It was great.
@00:17 We hit the pinnacle right there. Everything after was just fluff.
One of the most iritating things of the early 90s internet was pages that would play midi files.
They were GREAT.
Also marquee text, visit counters, frames and popups for no reason.
Aaaah, the 90's...
lol when you said Steam Engine as an example of innovation, I really thought you were talking about Source 2 or something xDxD
And everything is a question?
Loved the video, would be awesome to have a more in depth one as well!
It's amazing how MP3 (with the amazing LAME codec) can rival much newer formats just because of better psy models, kinda like x264 rivals H.265 codecs for the same reasons.
Still, this particular video was _LAME_ for different reasons...
Cubase lol brings back memory when I playing guitar back then
Well, that was awful, even at 2x speed. What content compression are you using? Where's all the fine details? Are you dumbing down for the sheeple that won't watch it in the first place?
That was a pretty bad description of how hearing works.
Well its not that accurate, but the things that are important to audioencoding he explained pretty well
lucky the title of the video wasn't called "how hearing works" then
cut him some slack.
@matsv201: not really. I was paranoid it would get overly specific, ex: talking about things specific to MP3 and Vorbis while totally ignoring things like ADPCM (IOW: what most talk about 'audio compression' amounts to...).
... it didn't even get this far ...
FWIW: I could apply those descriptions of psychoacoustic modeling and similar to an encoder I have for a customized ADPCM variant... (which is technically almost completely different than something like MP3; and technically if-anything borrows more directly from FLAC).
(ADD: though, like other ADPCM's, used fixed-form bit-packing rather than Huffman; mostly because of speed and also because the format needs to be strictly CBR for the use-case anyways; so some additional space-saving may be done externally via Deflate or similar...).
why?
Whats the Weissman score?
over 9000
This guy is great id love to see him more on this channel ,thx!!!
Products appear to have been placed.
computerphile should do an audiophile series
Well.. mod fiiles is really as good as you make them. Yea sure, the original mod files was 8 bit, some later was 16.
But because its one isolated sound, it really don´t have that much of a benefit having 16 bit.
Same thing with MIDI, because it's really just a set of instructions rather than an audio compression technique. Case in point: modern movie MIDI soundtracks.
Well.. the quality of the midi is of cause of the data it use to play it and quality of the sample in the hardware. This is quite obvious in the Super Nintendo having rather high quality sample in the ROM of the console. (it can also address extra sample on board the game ROM., so it works like a hybrid between a MIDI and MOD.
Looking at the highest quality late module files. They are pretty much CD quality with files in 100-250k range.
Even have some where they extracted the vocals and run the lyrics and chorus as different samples repeating and piching them up and down... still making it sound pretty much identical to the original song.
Some of the early/Mid 90-tys popular songs was actually made that way to begin with. E-type used normal tracker tools for most of his early stuff
Nothing about a transformation to the frequency domain?
"Fundamental Tennents" - that's the brand of lager that makes you say Tennents (or tenants) instead of tenets?
Joking about this slip aside, this is, sadly, nowhere near up to your usual standard.
yeah, skin...
Getting that Silicon Valley vibe from this
Computerphile should explain the middle-out compression algorithm!
This video tries to do too much in a single video and ends up really shallow and imprecise in how it covers the material as a result. Also the guy seems to be my age, but doesn't give the impression he was very involved with computers before ~2000.
Soo cooool...
I've worked for years developing audio software. This explanation is absolutely horrible.
Basic TENETS - not "tenants"!!
Tom's Diner!
Fraunhofer rulez!
tf is that product placement?
Nice.
well that was pointless
Oh boy, first! Gotta think something clever to say...
Hi mom!
You're not actually first, your mom must be very disappointed in you. As am I.
yep
First
Jake Goykia *Fourth