I have some strong doubts that the current generation of MP3 hardware decoders can easily be converted to Ogg Vorbis decoders. Ogg Vorbis requires far more memory than MP3 decoding. In addition, the Ogg Vorbis format apparently has difficulty running in 32 bit fixed point math. Most of the current MP3 decoders run on 16, 24, or 32-bit fixed point chips (the ARM7 on the iPod is 32 bits).
The upshot is that an Ogg Vorbis decoder would require more expensive hardware, with more memory, and either a floating point processor, or more free cycles to perform extended precision math on a fixed-point processor. If the cost of this extra hardware is more than the $.75 licensing fee for hardware units, then Ogg Vorbis becomes a much less attractive option for hardware MP3 decoder manufacturers.
Re:pitch correction is nothing new
on
Pitch Perfect Karaoke
·
· Score: 3, Informative
One interesting quote from the article:
MIT's Vercoe, who lauded Taito for finding a way to bring the sound synthesis technology to market, said pitch correction with Csound had been demonstrated long ago. It could even convert a spoken voice into melody, in real time.
"It's in the technology. You just have to switch it on," he said.
This makes me think that Vercoe did NOT ship the pitch correction "switched on." My suspicion is that this is due to IVL's patents. IVL is a Vancouver, BC company that has been marketing formant-preserving pitch shifters for quite some time now. IVL's technology is licenced by Digitech and TC Electronics, among other companies. One of IVL's patents makes claims on the idea of pitch shifters that automatically harmonize and correct pitch.
Personally, I think that IVL's claims in this area are somewhat tenuous. They have taken a pitch-shifting algorithm that they did not come up with (the algorithm in question was originally from a Computer Music Journal article in the late 1980's), and added a few features that ARE obvious to those "skilled in the art." Still, it sounds like Vercoe didn't want to take any chances. It seems like Vercoe provided a pitch output from the pitch shifter, and let Taito do what they want with it. Nice way to sidestep any legal issues.
Antares uses the same basic algorithm as IVL. Both companies have their own proprietary pitch detection algorithms.
It is interesting to see that Extended Csound is still up and running. I thought that it was pretty much dead in the water since 1999 or so. Hmmm...
I haven't seen anyone mention Supercollider yet. From what I have seen, Supercollider is the most powerful computer music language out there, barring C, C++, and assembly. Supercollider is a real-time, object-oriented computer music system, with a syntax similar to Smalltalk. It is text-based, which is really nice when constructing complicated sounds (unlike Csound, it has the type of control flow you would expect from a modern programming language). The sounds I have heard from Supercollider are beautiful; after working in Csound, it is amazing to hear such complex sounds being generated in real time, using a small fraction of the CPU. The current unit generator list is VERY powerful. It is easy to implement subtractive sythesis, modulation synthesis, granular techniques, frequency domain synthesis - pretty much everything except some physical models, which require finer control over the output vector size than is currently available in Supercollider.
Right now, Supercollider is Macintosh-only. However, the author of Supercollider is working on an OSX version, which he feels could be the basis of a Linux port.
Csound is still pretty cool. However, it is rather primitive, as far as synthesis languages go. There are no for() loops, nor any other way of looping other than if...goto. This makes constructing complicated instruments a rather tedious cut 'n' paste effort.
Under the hood, Csound has some inefficient unit generators. A lot of this has to do with the fact that many people learn how to program through working with Csound (I know that my first C coding went into Csound unit generators). Some of the unit generators do things that are really ugly - like having (value/anotherValue) for every sample, where you could have easily precomputed 1/anotherValue at initialization time. I have the feeling that this inner ugliness is why the various ports of Csound to real-time applications always seem to run slower than other software synths.
Another problem with the Csound source code: Many of the people who have coded the more complicated Csound UGs seem to be allergic to writing comments. You have huge, arcane unit generators, with NO comments whatsoever.
As far as the Csound book, I was disappointed in it. A lot of the chapters were written by members of the Csound list, who volunteered their time (I have some stuff on the CD ROM). This is great, but the book is often pitched as a good source of audio DSP techniques. Quite frankly, many of the algorithms in the book are sub par, and demonstrate a lack of comprehension of the literature from which they borrow. Nothing is wrong with publishing these algorithms - hey, if it makes sound, it works. My beef is with this book being used as a textbook, as I think it will result in a lot of bad algorithms being taught as state of the art. Many of the authors are not "top experts in the field" - they are enthusiastic newbies, who would be fired if they tried to put their inefficient, unstable algorithms into a commercial product.
As far as open source programs, check out Miller Puckette's PD. This program was the basis for MAX/MSP, although MSP adds a number of useful extensions. PD is open source, and is designed by a true expert in computer music and real time DSP.
Does anyone know what type of DSP chip is in the PEG-NR70? Depending on the power of the chip, this might be a nice thing to hack into for music synthesis, DSP effects processing of audio, etc.
If the main processor is a Motorola, perhaps the DSP is Motorola also. If the DSP is a 56K derivative, well, that would be most excellent, as this is somewhat of an industry standard in the audio industry (Motorola 56K-compatible processors are used by Eventide, Digidesign, TC Electronics, the XBOX, etc.).
http://www.gweep.net/~shifty/ is the homepage of a project to hook a Palm Pilot up to an ADI DSP dev kit for music DSP. Maybe the new Sony would allow you to do the same thing without the external DSP.
I would think the main use would be to learn how to program parts of the PS2, without spending $18K for a full-on dev kit. The tricky part about PS2 programming is learning the microcode for the vector units, graphics synthesiszer, etc. If you are a game developer, it is worth spending $199 just for the PDFs of the manuals for the EE, VU0, VU1, and the GS.
Yeah, go ahead and mod it as funny. Meanwhile, I can go out on any given weekend, buy a used turntable for $20 at a garage sale, and get almost any record that came out before the mid-80s for $.50.
And vinyl is still being made. Generally speaking, it is the smaller labels that still put out records. Of course, it is the smaller labels that actually put out GOOD music these days.
Just looked at the Analog Devices, and the list price of the 2181 EZKIT is $295. However, ALL of the EZKITs seem to be in that price range. If you are going to get an EZKIT, take a look at the following:
ADSP-2191 EZKIT. Code compatible with the 2181, but faster core. Also has 24 bits of memory address, as opposed to 14 for the 218x core. This is useful for longer delay lines, for reverbs and such.
ADSP-21065L EZKIT. 32 bit SHARC, can operate in integer or floating point modes. Probably can't run 2181 code without some work. However, there are some EXCELLENT audio application notes available; check out http://www.analog.com/library/applicationNotes/AdA udio/21065L_Audio_Tutorial.pdf. No crappy Schroeder reverbs here; we're talking Dattoro/Griesinger reverbs, with full source code!
ADSP-21161 EZKIT. Similar to 21065, but with SIMD capabilities. Audio application notes at http://www.analog.com/library/applicationNotes/dsp/32_Processors/21161_Audio_Tutorial.pdf, although the code doesn't seem to be SIMD-optimized.
Slashdot Mirror
The upshot is that an Ogg Vorbis decoder would require more expensive hardware, with more memory, and either a floating point processor, or more free cycles to perform extended precision math on a fixed-point processor. If the cost of this extra hardware is more than the $.75 licensing fee for hardware units, then Ogg Vorbis becomes a much less attractive option for hardware MP3 decoder manufacturers.
MIT's Vercoe, who lauded Taito for finding a way to bring the sound synthesis technology to market, said pitch correction with Csound had been demonstrated long ago. It could even convert a spoken voice into melody, in real time.
"It's in the technology. You just have to switch it on," he said.
This makes me think that Vercoe did NOT ship the pitch correction "switched on." My suspicion is that this is due to IVL's patents. IVL is a Vancouver, BC company that has been marketing formant-preserving pitch shifters for quite some time now. IVL's technology is licenced by Digitech and TC Electronics, among other companies. One of IVL's patents makes claims on the idea of pitch shifters that automatically harmonize and correct pitch.
Personally, I think that IVL's claims in this area are somewhat tenuous. They have taken a pitch-shifting algorithm that they did not come up with (the algorithm in question was originally from a Computer Music Journal article in the late 1980's), and added a few features that ARE obvious to those "skilled in the art." Still, it sounds like Vercoe didn't want to take any chances. It seems like Vercoe provided a pitch output from the pitch shifter, and let Taito do what they want with it. Nice way to sidestep any legal issues.
Antares uses the same basic algorithm as IVL. Both companies have their own proprietary pitch detection algorithms.
It is interesting to see that Extended Csound is still up and running. I thought that it was pretty much dead in the water since 1999 or so. Hmmm...
Right now, Supercollider is Macintosh-only. However, the author of Supercollider is working on an OSX version, which he feels could be the basis of a Linux port.
Under the hood, Csound has some inefficient unit generators. A lot of this has to do with the fact that many people learn how to program through working with Csound (I know that my first C coding went into Csound unit generators). Some of the unit generators do things that are really ugly - like having (value/anotherValue) for every sample, where you could have easily precomputed 1/anotherValue at initialization time. I have the feeling that this inner ugliness is why the various ports of Csound to real-time applications always seem to run slower than other software synths.
Another problem with the Csound source code: Many of the people who have coded the more complicated Csound UGs seem to be allergic to writing comments. You have huge, arcane unit generators, with NO comments whatsoever.
As far as the Csound book, I was disappointed in it. A lot of the chapters were written by members of the Csound list, who volunteered their time (I have some stuff on the CD ROM). This is great, but the book is often pitched as a good source of audio DSP techniques. Quite frankly, many of the algorithms in the book are sub par, and demonstrate a lack of comprehension of the literature from which they borrow. Nothing is wrong with publishing these algorithms - hey, if it makes sound, it works. My beef is with this book being used as a textbook, as I think it will result in a lot of bad algorithms being taught as state of the art. Many of the authors are not "top experts in the field" - they are enthusiastic newbies, who would be fired if they tried to put their inefficient, unstable algorithms into a commercial product.
As far as open source programs, check out Miller Puckette's PD. This program was the basis for MAX/MSP, although MSP adds a number of useful extensions. PD is open source, and is designed by a true expert in computer music and real time DSP.
If the main processor is a Motorola, perhaps the DSP is Motorola also. If the DSP is a 56K derivative, well, that would be most excellent, as this is somewhat of an industry standard in the audio industry (Motorola 56K-compatible processors are used by Eventide, Digidesign, TC Electronics, the XBOX, etc.).
http://www.gweep.net/~shifty/ is the homepage of a project to hook a Palm Pilot up to an ADI DSP dev kit for music DSP. Maybe the new Sony would allow you to do the same thing without the external DSP.
I would think the main use would be to learn how to program parts of the PS2, without spending $18K for a full-on dev kit. The tricky part about PS2 programming is learning the microcode for the vector units, graphics synthesiszer, etc. If you are a game developer, it is worth spending $199 just for the PDFs of the manuals for the EE, VU0, VU1, and the GS.
Yeah, go ahead and mod it as funny. Meanwhile, I can go out on any given weekend, buy a used turntable for $20 at a garage sale, and get almost any record that came out before the mid-80s for $.50. And vinyl is still being made. Generally speaking, it is the smaller labels that still put out records. Of course, it is the smaller labels that actually put out GOOD music these days.
ADSP-2191 EZKIT. Code compatible with the 2181, but faster core. Also has 24 bits of memory address, as opposed to 14 for the 218x core. This is useful for longer delay lines, for reverbs and such.
ADSP-21065L EZKIT. 32 bit SHARC, can operate in integer or floating point modes. Probably can't run 2181 code without some work. However, there are some EXCELLENT audio application notes available; check out http://www.analog.com/library/applicationNotes/AdA udio/21065L_Audio_Tutorial.pdf. No crappy Schroeder reverbs here; we're talking Dattoro/Griesinger reverbs, with full source code!
ADSP-21161 EZKIT. Similar to 21065, but with SIMD capabilities. Audio application notes at http://www.analog.com/library/applicationNotes/dsp /32_Processors/21161_Audio_Tutorial.pdf, although the code doesn't seem to be SIMD-optimized.
Sean Costello