Uhm, from the reading I did real quick (replaygain.org, BTW), replaygain won't fix recording dynamic. Its useful for equalization of volume levels across an album, but does not change track dynamic, which is what we were talking about. To fix that properly, you'd need access to the master (you could play around with trying to decompress the 2-track file, but it wouldn't come out well most likely).
You're confusing conversion and calculation. ProTools is a multitrack program; the idea behind allowing that much bitdepth on the main buses is to provide a ton of headroom for mix and processing. 60 bits is probably overkill, but I know they use at least 48 bits on the main bus (remember, the program may have 256 tracks, each with a 6dB boost, leading to a 9 bit increase over single track - beyond that, plugins and such may lead to even higher gains, which is why they provide so much headroom).
D/A converters can't physically have an effective range better than 20 bits at room temperature (without getting into more exotic architectures like sigma-delta converters), but that doesn't mean that processing shouldn't work on higher bit-depth bussing.
Foldover is just a non-technical name for aliasing; same exact thing.
As to your spectral analysis; it depends a lot on the downsampling process. If you recorded at 88.2 and downsampled to 44.1 with no further processing, you'll get aliasing from any components between 22kHz and 44kHz. However, almost all audio programs are smart enough to run a filter during the downsampling; still, I try to run a filter to remove possible aliasing prior to any downsampling, just in case.
Next; anything telling you its recording at better than 20 bits is probably lying to you. The reason for this is thermal noise; the motion of the molecules inside your semiconductors is a physical limit, and at room temp you're limited to around 20 bits by the electrical noise of that motion. As long as your editor has proper calculation and bussing, it won't screw things up if you import at 16 bit.
I like vinyl, but not for the sound. I just like records. They feel right, you know?
I'm a fascist on terminology only because I'm formally trained in signal processing, and have to listen to half-assed home recordists try to sound smart weekly. But its important to get terms right if you want to understand the more technical information out there, especially if you start getting into things like A/D architectures, DNL, INL, sampling jitter, windowing, and all the other fun stuff that kept me occupied for a year or two in college.
I don't do a lot with MaxMSP, but my understanding is that most multitracking programs, or at least ProTools, do their buses as 24 bit fixed point. Processing is usually done as at least 32 fixed, often 32 or 64 float.
Foldover is a non-technical name for what any formally trained signal processing person will call aliasing. I have no idea what you're referring to, but the correct term for foldover is, in fact, aliasing - this is why pre-filters are called "anti-aliasing" filters. I believe you're thinking of aliasing the way video people use it, which in the sound world is called quantization noise - the error generated by having discrete steps in amplitude as opposed to a continuous amplitude scale.
I recorded a track for an indie band once, and just for fun made a mix of it compressed down to almost no dynamic; played it back for one of the bandmembers and he said something about it being "like having a red hot poker shoved through my ear".
I have no idea what your link is trying to say. That said - I don't know of any systems that track at greater than 24fixed. Once its tracked, mix precision is nearly always at least 32 bit and most good systems use at least 48 bits. My understanding, based on talking to a couple of Digidesign's guys a couple years ago, is that Pro Tools internal mix buses are fixed point bit depth.
As to the apples and oranges, it was intended to enhance the comparison. You want to compare a recently mastered jazz recording to a 60s jazz recording, the trend will be there as well, it simply won't be as pronounced.
You have your terminology totally wrong, first off. You state that noone needs a frequency range greater than 40kHz; what you meant to state was that noone needs a *sampling rate* greater than 40kHz. The 20kHz you state next is the frequency range of human hearing (usually a little bit lower for adults, most adults will top out in the 15-19kHz range).
Engineers need to work at high bitdepth and sampling rate, not at high frequency range, although FR is a direct consequence of sampling rate.
Next; there's no such thing as a floating point sampling rate. You're thinking bitdepth, and using a floating point bitdepth is uncommon. Most current digital editing systems (i.e. ProTools) record 24bit fixed point audio during tracking, and maintain some higher level of precision during mix; IIRC, ProTools 5 had 60 bit main buses, but I could be quite off on that.
Next, physics.
Yes, theoretically you only need 2xBW (bandwidth), but anyone who actually works on this shit will tell you that they want more. This is because in order to avoid aliasing artifacts, you need to filter everything above BW. Unfortunately, brick wall filters are not implementable in realtime (and not really implementable in a stored data system either, but that's another story again). So you've got a non brick wall filter, which means you need some frequency range above your max desired signal frequency, but below your 1/2Fs frequency. This range is where your filter is transitioning from passband to stopband.
Next, we have beating artifacts. This occurs when you have a sound at a frequency very close to your 1/2Fs frequency; while frequency will be properly reproduced, you'll get amplitude modulation artifacts. Because of beating, typical industrial sampled data systems sample at a minimum of 5xBW; 10 or 20x BW is preferred. Since we're looking at a 20kHz BW, 192kHz (DVD-A) should do quite nicely.
I'm with you on the lack of dynamic range in modern music though; load a Britney Spears track into an editor, then load a classic jazz track (I recommend Miles Davis' "So What") and compare the envelopes. Scary.
Not at all a myth if you ever want to do real engineering, for one thing.
I know a lot of people who work in the auto and aerospace industries. We have a name for the ones who didn't graduate college; "technician".
I love my technicians, but the ones who have been doing it for 10 years make about the same amount of money as me, and I've been out of school (B.S.EE) for about a year.
If the ad was made by someone else, presumably they have already paid the artist the fee the artist asked for in order to allow use of their song in the ad. In that case, its still questionable in legality and would probably depend on whether or not the ad's producer felt like playing the heavy.
My point went to creating your own 'ad' featuring some song; that's illegal.
Never used any of Mentor's FPGA tools, I've mostly had their capture/layout toolchain inflicted on me. My comment regarded those tools. If I ever experience Leonardo or PRTL, hopefully they're better designed.
I vaguely remember Modelsim from way back when (school) - I don't recall liking it all that much, though it was years ago.
Mentor's interface is ugly and totally counterintuitive to modern interfaces. It runs slower, in my experience, than its counterparts. Library management is ugly, and I don't particularly like their workflow. This is why I don't like their toolset.
Never tried PADS. I have tried Orcad, which you would know if you had actually read my post; Cadence owns Orcad now, didn't you know? I've done designs in Orcad/Cadence (Capture/Layout), DA/BS, and the full Altium Protel DXP suite. They all have pretty massive flaws. DA has an interface that is totally, totally counterintuitive and needs improvement; the only reason they won't change it is all the board hacks who've been using it forever and would scream bloody murder if it changed.
In terms of design capture, I seem to work fastest in DXP and slowest in DA, with Capture coming in somewhere in the middle. As to board layout, I can't honestly compare my work speed, since my BSRE layouts were all small, and the layouts I've been doing in DXP are comparatively quite large.
That said, I don't like *any* of them, I just happen to dislike Mentor more. I know more than a few people who agree with me on that.
Honestly, I just wanted it anon because who knows if I caught a glance at something internal; I'm pretty sure its all publicly released info that I'd seen, but just in case. That said, after reading that, I haven't seen anything that isn't in that article.
Well, since I blew it - I work for Motorola's automotive group, not any of the phone/802.11/networking groups, so what I've heard has been rumors only - we're not even in the same physical city as the phone guys.
(note: I work at Motorola, which is why this is posted anonymously.)
I've also been hearing rumors of a Motorola VOIP over WiFi, and eventually a dual mode GSM/VOIP-WIFI phone. I'd expect to see one in the next year or two.
For the most part, I'm with you; the one exception I would make is the foreign languages. When I went to school, we were required in (6th? 7th, maybe) to take 9 weeks each of Spanish, French, German, and Latin, the 4 languages our school system offered. The point was not to teach anything useful; the point was to get students interested enough in a language to decide to take a full year the next year. Good plan for those students who didn't really know, and not exactly a huge waste of time (total of one class over a year) for the ones who didn't care.
In a similar vein, my high school required a total of one year of phys ed over the 4 years of high school. Yeah, we get some lazy fat fucks coming out, but trying to make the lazy people do phys ed does jack shit for anyone.
Not sure if you meant have elementary school be more like later schools, where you have seperate teachers for seperate classes; that, I think, would be a bad idea. Kids that young, it helps to just have the one teacher; I wouldn't start moving them between before 4th grade, 5th or 6th is probably better.
I was lucky enough to have most of my required history courses be a lot more about essay writing than about memorizing dates; the one 'dates' course I had, I actually wound up walking out about 1/3 of the way through the year (after cursing out the teacher for being "a useless son of a bitch who knows less history than my dog", IIRC) and proceeded to ace the AP just to get in his face. Got full credit for the class too, since he had a rule correlating AP results to grades in the course.
My definition of CS excludes IT, actually. Config on a W2k proxy server utilizes the fruits of CS, but that's like saying driving a stick shift is employing mechanical engineering because you have to use the clutch and the shift lever. Using a CS graduate to configure a fucking proxy is a waste of an education; it's like asking an electrical engineer to set up your VCR.
CS - Coding is on the boundary; architecture and algorithms are the core.
And the plumber makes more money than the CS PHD in *what* fucking world? CS PHDs tend to make at least 80k for a 40 hour work week, at least in my neck of the corporate woods. That's entry. A journeyman plumber working 40 hours a week doesn't make anywhere near 80k. Yes, the PHD put in more schooling; on the other hand, the journeyman probably doesn't own his own company, and will make significantly less as a result. If he does own his own company, he's got upkeep on tools and shit like that.
Nope. But then again, I spend some of my time working with European and Japanese engineers, and they mostly speak English, or at least enough for me to be aware that they still have pretty good engineers.
I also suspect that if you look at any of the top 10 countries, there will be strong concerns about the state of education; the thing is, education is one of those things that will never be perfect, that can always be better. In other words, there will always be room to complain about it.:)
I don't have enough interest to purchase a repeater; in addition to the added audio streaming, this is nice and small.
Also, I don't always believe footnotes. I was hoping for a bit more hard info on "Apple just won't guarantee it for support reasons, but it probably ought to work" vs. "No, it won't work."
Slashdot Mirror
Uhm, from the reading I did real quick (replaygain.org, BTW), replaygain won't fix recording dynamic. Its useful for equalization of volume levels across an album, but does not change track dynamic, which is what we were talking about. To fix that properly, you'd need access to the master (you could play around with trying to decompress the 2-track file, but it wouldn't come out well most likely).
You're confusing conversion and calculation. ProTools is a multitrack program; the idea behind allowing that much bitdepth on the main buses is to provide a ton of headroom for mix and processing. 60 bits is probably overkill, but I know they use at least 48 bits on the main bus (remember, the program may have 256 tracks, each with a 6dB boost, leading to a 9 bit increase over single track - beyond that, plugins and such may lead to even higher gains, which is why they provide so much headroom).
D/A converters can't physically have an effective range better than 20 bits at room temperature (without getting into more exotic architectures like sigma-delta converters), but that doesn't mean that processing shouldn't work on higher bit-depth bussing.
Foldover is just a non-technical name for aliasing; same exact thing.
As to your spectral analysis; it depends a lot on the downsampling process. If you recorded at 88.2 and downsampled to 44.1 with no further processing, you'll get aliasing from any components between 22kHz and 44kHz. However, almost all audio programs are smart enough to run a filter during the downsampling; still, I try to run a filter to remove possible aliasing prior to any downsampling, just in case.
Next; anything telling you its recording at better than 20 bits is probably lying to you. The reason for this is thermal noise; the motion of the molecules inside your semiconductors is a physical limit, and at room temp you're limited to around 20 bits by the electrical noise of that motion. As long as your editor has proper calculation and bussing, it won't screw things up if you import at 16 bit.
I like vinyl, but not for the sound. I just like records. They feel right, you know?
Hah, true, I did.
I'm a fascist on terminology only because I'm formally trained in signal processing, and have to listen to half-assed home recordists try to sound smart weekly. But its important to get terms right if you want to understand the more technical information out there, especially if you start getting into things like A/D architectures, DNL, INL, sampling jitter, windowing, and all the other fun stuff that kept me occupied for a year or two in college.
I don't do a lot with MaxMSP, but my understanding is that most multitracking programs, or at least ProTools, do their buses as 24 bit fixed point. Processing is usually done as at least 32 fixed, often 32 or 64 float.
Foldover is a non-technical name for what any formally trained signal processing person will call aliasing. I have no idea what you're referring to, but the correct term for foldover is, in fact, aliasing - this is why pre-filters are called "anti-aliasing" filters. I believe you're thinking of aliasing the way video people use it, which in the sound world is called quantization noise - the error generated by having discrete steps in amplitude as opposed to a continuous amplitude scale.
I recorded a track for an indie band once, and just for fun made a mix of it compressed down to almost no dynamic; played it back for one of the bandmembers and he said something about it being "like having a red hot poker shoved through my ear".
I have no idea what your link is trying to say. That said - I don't know of any systems that track at greater than 24fixed. Once its tracked, mix precision is nearly always at least 32 bit and most good systems use at least 48 bits. My understanding, based on talking to a couple of Digidesign's guys a couple years ago, is that Pro Tools internal mix buses are fixed point bit depth.
As to the apples and oranges, it was intended to enhance the comparison. You want to compare a recently mastered jazz recording to a 60s jazz recording, the trend will be there as well, it simply won't be as pronounced.
You have your terminology totally wrong, first off. You state that noone needs a frequency range greater than 40kHz; what you meant to state was that noone needs a *sampling rate* greater than 40kHz. The 20kHz you state next is the frequency range of human hearing (usually a little bit lower for adults, most adults will top out in the 15-19kHz range).
Engineers need to work at high bitdepth and sampling rate, not at high frequency range, although FR is a direct consequence of sampling rate.
Next; there's no such thing as a floating point sampling rate. You're thinking bitdepth, and using a floating point bitdepth is uncommon. Most current digital editing systems (i.e. ProTools) record 24bit fixed point audio during tracking, and maintain some higher level of precision during mix; IIRC, ProTools 5 had 60 bit main buses, but I could be quite off on that.
Next, physics.
Yes, theoretically you only need 2xBW (bandwidth), but anyone who actually works on this shit will tell you that they want more. This is because in order to avoid aliasing artifacts, you need to filter everything above BW. Unfortunately, brick wall filters are not implementable in realtime (and not really implementable in a stored data system either, but that's another story again). So you've got a non brick wall filter, which means you need some frequency range above your max desired signal frequency, but below your 1/2Fs frequency. This range is where your filter is transitioning from passband to stopband.
Next, we have beating artifacts. This occurs when you have a sound at a frequency very close to your 1/2Fs frequency; while frequency will be properly reproduced, you'll get amplitude modulation artifacts. Because of beating, typical industrial sampled data systems sample at a minimum of 5xBW; 10 or 20x BW is preferred. Since we're looking at a 20kHz BW, 192kHz (DVD-A) should do quite nicely.
I'm with you on the lack of dynamic range in modern music though; load a Britney Spears track into an editor, then load a classic jazz track (I recommend Miles Davis' "So What") and compare the envelopes. Scary.
Not at all a myth if you ever want to do real engineering, for one thing.
I know a lot of people who work in the auto and aerospace industries. We have a name for the ones who didn't graduate college; "technician".
I love my technicians, but the ones who have been doing it for 10 years make about the same amount of money as me, and I've been out of school (B.S.EE) for about a year.
Wasn't the 486 the first x86 processor to clock the core at a different speed from the buses? I'd argue that to be a significant change.
I still remember the debates... DX33 or DX2-50?
If the ad was made by someone else, presumably they have already paid the artist the fee the artist asked for in order to allow use of their song in the ad. In that case, its still questionable in legality and would probably depend on whether or not the ad's producer felt like playing the heavy.
My point went to creating your own 'ad' featuring some song; that's illegal.
No, because you have no legal right to use someone else's recording as music in an advertisement.
Never used any of Mentor's FPGA tools, I've mostly had their capture/layout toolchain inflicted on me. My comment regarded those tools. If I ever experience Leonardo or PRTL, hopefully they're better designed.
I vaguely remember Modelsim from way back when (school) - I don't recall liking it all that much, though it was years ago.
Want insight?
Mentor's interface is ugly and totally counterintuitive to modern interfaces. It runs slower, in my experience, than its counterparts. Library management is ugly, and I don't particularly like their workflow. This is why I don't like their toolset.
Happier, jackass?
Never tried PADS. I have tried Orcad, which you would know if you had actually read my post; Cadence owns Orcad now, didn't you know? I've done designs in Orcad/Cadence (Capture/Layout), DA/BS, and the full Altium Protel DXP suite. They all have pretty massive flaws. DA has an interface that is totally, totally counterintuitive and needs improvement; the only reason they won't change it is all the board hacks who've been using it forever and would scream bloody murder if it changed.
In terms of design capture, I seem to work fastest in DXP and slowest in DA, with Capture coming in somewhere in the middle. As to board layout, I can't honestly compare my work speed, since my BSRE layouts were all small, and the layouts I've been doing in DXP are comparatively quite large.
That said, I don't like *any* of them, I just happen to dislike Mentor more. I know more than a few people who agree with me on that.
For that matter, maybe he can convince them to make tools worth running.
God I hate Mentor. I can't really say the Cadence or Altium toolsets are any better, but I still hate Mentor.
Honestly, I just wanted it anon because who knows if I caught a glance at something internal; I'm pretty sure its all publicly released info that I'd seen, but just in case. That said, after reading that, I haven't seen anything that isn't in that article.
You might want to check my own reply to my post, where I noted that I did, in fact, forget to check that inviting little box.
Haha. Posted anonymously indeed.
Whoops.
Well, since I blew it - I work for Motorola's automotive group, not any of the phone/802.11/networking groups, so what I've heard has been rumors only - we're not even in the same physical city as the phone guys.
(note: I work at Motorola, which is why this is posted anonymously.)
I've also been hearing rumors of a Motorola VOIP over WiFi, and eventually a dual mode GSM/VOIP-WIFI phone. I'd expect to see one in the next year or two.
Always nice to see the reminder that NASA can do great fucking engineering when the mission is properly separated from politics.
For the most part, I'm with you; the one exception I would make is the foreign languages. When I went to school, we were required in (6th? 7th, maybe) to take 9 weeks each of Spanish, French, German, and Latin, the 4 languages our school system offered. The point was not to teach anything useful; the point was to get students interested enough in a language to decide to take a full year the next year. Good plan for those students who didn't really know, and not exactly a huge waste of time (total of one class over a year) for the ones who didn't care.
In a similar vein, my high school required a total of one year of phys ed over the 4 years of high school. Yeah, we get some lazy fat fucks coming out, but trying to make the lazy people do phys ed does jack shit for anyone.
Not sure if you meant have elementary school be more like later schools, where you have seperate teachers for seperate classes; that, I think, would be a bad idea. Kids that young, it helps to just have the one teacher; I wouldn't start moving them between before 4th grade, 5th or 6th is probably better.
I was lucky enough to have most of my required history courses be a lot more about essay writing than about memorizing dates; the one 'dates' course I had, I actually wound up walking out about 1/3 of the way through the year (after cursing out the teacher for being "a useless son of a bitch who knows less history than my dog", IIRC) and proceeded to ace the AP just to get in his face. Got full credit for the class too, since he had a rule correlating AP results to grades in the course.
My definition of CS excludes IT, actually. Config on a W2k proxy server utilizes the fruits of CS, but that's like saying driving a stick shift is employing mechanical engineering because you have to use the clutch and the shift lever. Using a CS graduate to configure a fucking proxy is a waste of an education; it's like asking an electrical engineer to set up your VCR.
CS - Coding is on the boundary; architecture and algorithms are the core.
And the plumber makes more money than the CS PHD in *what* fucking world? CS PHDs tend to make at least 80k for a 40 hour work week, at least in my neck of the corporate woods. That's entry. A journeyman plumber working 40 hours a week doesn't make anywhere near 80k. Yes, the PHD put in more schooling; on the other hand, the journeyman probably doesn't own his own company, and will make significantly less as a result. If he does own his own company, he's got upkeep on tools and shit like that.
Nope. But then again, I spend some of my time working with European and Japanese engineers, and they mostly speak English, or at least enough for me to be aware that they still have pretty good engineers.
:)
I also suspect that if you look at any of the top 10 countries, there will be strong concerns about the state of education; the thing is, education is one of those things that will never be perfect, that can always be better. In other words, there will always be room to complain about it.
Apparently, my model does not.
I don't have enough interest to purchase a repeater; in addition to the added audio streaming, this is nice and small.
Also, I don't always believe footnotes. I was hoping for a bit more hard info on "Apple just won't guarantee it for support reasons, but it probably ought to work" vs. "No, it won't work."