[motm] file uploaded, Additive Synthesis, Strings, Fourier, etc
2003-06-14 by elhardt@att.net
OK, so I'm a little backlogged at responding to some old messages. This in in regards to all the talk about additive synthesis and whatnot over a month ago. Since a lot of this stuff is right up my alley (and I just uploaded a related MP3), here it goes. Paul Haneberg writes: >>I am wondering if anyone has any suggestions for a piece of software capable of doing analysis of a single note played by a particular instrument, for example a wav file of a sax. I am looking for something that can give me a time varying spectra so that I can see the individual envelopes of each harmonic as well as determine the individual harmonic amplitudes.<< Some standard sound editting programs like Sound Forge and Wavelab can give you 3D time varying spectra plots, although sloppy and hard to derive useful info from a lot of the time. Otherwise, follow Pauls advice and search the net. CoolEdit Pro can give you a realtime animated harmonic plot of instruments as they play or you can manually slide the pointer along a waveform and see the changes in harmonic amplitudes. Their FFT filter is great for filtering out individual harmonics and listenting to them by themselves (a technique I use). >>Also helpful would be if anyone knows of a source (book or website) for this sort of information in a quantitative form, not just some general looking graphs.<< I've found almost nothing myself. There is supposedly a well known set of harmonic graphs for most instruments, but those are only averages, and you need a different one for every note, and all the links to them are dead. For strings I've been analysing samples. Of course you either need to store different harmonics for every note to preserve the formants of the instrument, or come up with a multi-band formant filter. If you go the formant filter route you start to defeat the need for additive synthesis in the first place. >>It would be equally interesting if an analysis tool could easily show you the frequency response of an instrument body, if it could be done without sticking transducers all over the place and then applying a variety of excitation sources.<< Unfortunately the acoustic world is far more complex than it should be. Resonances and dips in the frequency spectrum can move or disappear depending on the note being played. It's a mess that only my future synth engine will handle correctly. However you can pick a range of notes and try to nail them down and hope the rest still sound pretty good. Looking at the harmonics of a violin note sliding up a couple of octaves up a string can help you see harmonics get sucked into dips and pop up at major resonances. Roger Rossen writes: >>Additive synthesis has existed for quite awhile now- especially in the academic sector - 'modulars' or software based ones for that matter have been around for quite awhile.Its a LOT of wanking around with minimal effect - my general proof: no major synth manuf. are pursuing it anymore.<< It's all about user interface. Fairlight had it right. Just draw the time varying envelope of each harmonic with a lightpen. Simple and Fast! 20 years ago a wrote a program on the Apple ][ that did it that way (trackball instead of lightpen). It worked great. On the K5000, dialing in one harmonic at a time, and doing that four times, then connecting those to different points on a multi- segment envelope to morph between them, all from a little LCD window is the most miserable hell a person can go through, and not very flexible either. Additive is coming back because of the computer. There are two knew softsynths that are coming out just for additive. Paul Haneberg writes: >To achieve that goal, it would be musically useful to be able to generate a Fourier Series with complete control over the amplitude of each harmonic over time, which you certainly cannot do at present with the MOTM.<< If you have a K5000, why duplicate that effort. You can just use your K5000 as an additive oscillator to feed into your MOTM. Ken Tkacs writes: >>For certain tasks, such as Additive Synthesis, digital is pretty necessary. To do anything interesting, you need something like 128 oscillators, 128 amplifiers, and 128 envelope generators... then you need a way to control it all, and about three weeks to turn knobs<< 128 oscillators? The K5000 can only do 64 at a time, and I think the fairlight was 32. My MP3 demo only uses 18 for test purposes, and it's only a little bit less bright than the real sound. Interesting thing is I can filter some sound files so that instruments are down to between 1 and 4 harmonics and you can still tell what instrument it is. For non-natural harmonic stuff, like kettle drums for instance, you're better off forgetting additive altogether. Tobias Enhus writes: >>Additive synthesis is a very valid technique to investigate.<< Exactly. It's the only way to mimic some acoustic instruments almost exactly. >>There's an additive technique in particular, that I've used a lot in both Csound and Kyma (with great results) and hope to recreate with MOTM modules, additive harmonic filtering (Paul Lansky). You feed a sound to a bank of filters with high resonance tuned and enveloped to your specs. This technique captures more noise and enharmonic material which is very important ingredients of a natural sound.<< One of those new additive softsynths claims to have a 128 band noise filterbank. Sounds similar. For most acoustic instruments, that's not necessary. But I've found with the human voice, there is a lot of breath noise going on between the harmonics. I've pieced together a file without that noise and the voice sounds synthetic. With the noise, it comes to life. The following file is uploaded into the motm yahoogroups file section: StringSynthesisElhardt.mp3 Some of you have heard the violin tone before, but this MP3 gives you a Wendy Carlos - Secrets of Synthesis (voice and all) style explaination as to how it was done. Plus you get a corny advertisement (I was just amusing myself) for my synthesis technique showing the awful results of current technology for doing string sounds, and the latest Nord modular samples using multiple delay lines (comb filtering), phaser, parametric EQs, wavewrappers, etc, to get pretty good strings sounds now, since "Microdetailed" synthesis won't be available until I write a synth engine to handle all aspects of it. Thanks to the Boss VT-1 Voice Transformer for helping me out with the voice overs. -Elhardt