Thought you analogue guys might like this story on wow flutter removal from tape and film.
Very clever process. The bias signal is used as the "clock" to detect the speed changes, then the wow & flutter are removed via software.
The NPR story has some nice audio examples of before and after.
http://www.npr.org/templates/story/story.php?storyId=7489316
Very clever process. The bias signal is used as the "clock" to detect the speed changes, then the wow & flutter are removed via software.
The NPR story has some nice audio examples of before and after.
http://www.npr.org/templates/story/story.php?storyId=7489316
Yeah, I've got a bunch of old cassette tapes left to go digital, too.
Recently I aquired a tape drive that is as similar to the original recording unit as I could find.
But it reveals "wow", which means it's the tapes...
So what's required?
- a DC coupled tape head amplifier,
- a DC coupled A/D converter,
- reasonable resolution (>=20bit) to catch the signal while offset by the "wobbly DC" bias content,
- sample frequency e.g. four times the original bandwith (as wow/flutter also means accelleration of the tape),
- an algorithm to extract the DC bias signal ("tracking" low pass filter) and feed a sort-of PID control that regulates the pitch of the AC content;
A microphone amplifier without capacitors in the signal path could double as a head amplifier.
A DIY A/D converter could be built without input decoupling, too (e.g. the Burr Brown PCM290x series USB codecs already are 16bit/48kHz). A commercial ADC could probably be modified to suit, too.
Extracting the DC content is described as being difficult by the interview partner in the mentioned audio feed, as some mechanical influences reach into the audio band (e.g. wow/flutter higher than, say, 20Hz). It sounds more like an assisted smart process done by a sound engineer through listening and trial-and-error. But influences below, say, 15Hz could be isolated automagically by a brickwall low pass filter, still.
Then comes the secret. By resembling the original machine (as it's called in the interview), they probably mean isolating uncorrelated modulations of the infrasonic content, then modeling a compensation for each of those that are relevant, then superposition them to a complementary (de)modulation signal. The amplitude of the reference signal then kind of represents the modulation index of a "demodulator" stage, which would (in this context) be mostly FM (but AM can probably still be used for compensation of magnetization-loss-vibrato).
Sounds like a lot of fun with Max/MSP (the audio engine), or Nyquist (the plugin scripting language). Anyone? 😎
Cheers,
Sebastian.
Recently I aquired a tape drive that is as similar to the original recording unit as I could find.
But it reveals "wow", which means it's the tapes...
So what's required?
- a DC coupled tape head amplifier,
- a DC coupled A/D converter,
- reasonable resolution (>=20bit) to catch the signal while offset by the "wobbly DC" bias content,
- sample frequency e.g. four times the original bandwith (as wow/flutter also means accelleration of the tape),
- an algorithm to extract the DC bias signal ("tracking" low pass filter) and feed a sort-of PID control that regulates the pitch of the AC content;
A microphone amplifier without capacitors in the signal path could double as a head amplifier.
A DIY A/D converter could be built without input decoupling, too (e.g. the Burr Brown PCM290x series USB codecs already are 16bit/48kHz). A commercial ADC could probably be modified to suit, too.
Extracting the DC content is described as being difficult by the interview partner in the mentioned audio feed, as some mechanical influences reach into the audio band (e.g. wow/flutter higher than, say, 20Hz). It sounds more like an assisted smart process done by a sound engineer through listening and trial-and-error. But influences below, say, 15Hz could be isolated automagically by a brickwall low pass filter, still.
Then comes the secret. By resembling the original machine (as it's called in the interview), they probably mean isolating uncorrelated modulations of the infrasonic content, then modeling a compensation for each of those that are relevant, then superposition them to a complementary (de)modulation signal. The amplitude of the reference signal then kind of represents the modulation index of a "demodulator" stage, which would (in this context) be mostly FM (but AM can probably still be used for compensation of magnetization-loss-vibrato).
Sounds like a lot of fun with Max/MSP (the audio engine), or Nyquist (the plugin scripting language). Anyone? 😎
Cheers,
Sebastian.
Tape bias is AC, not DC. Usually 40K or above. That's what you've got to pick up off the tape to be processed.
Then you use that HF bias as the "clock" to correct the speed variations. More complicated than that in practice, certainly, but that's the basic idea.
Then you use that HF bias as the "clock" to correct the speed variations. More complicated than that in practice, certainly, but that's the basic idea.
Yep, meanwhile I've read up on it a little. 😉
Same applies, but we have to use a high pass filter instead.
The ADC faces a problem, though: bandwith.
Maybe it's better to do the filtering in the analog domain and use separate ADCs for the original and the extracted bias signal. Bandwith limitation could be circumvented this way (through convolution), but the requirement would still be in the order of 96kHz or even 192kHz.
Cheers.
Same applies, but we have to use a high pass filter instead.
The ADC faces a problem, though: bandwith.
Maybe it's better to do the filtering in the analog domain and use separate ADCs for the original and the extracted bias signal. Bandwith limitation could be circumvented this way (through convolution), but the requirement would still be in the order of 96kHz or even 192kHz.
Cheers.
I stumbled across a relevant program today, called Capstan and thought to update this thread here as well.
Check out the offering here:
celemony_ :: Capstan
Check out the offering here:
celemony_ :: Capstan
- Status
- Not open for further replies.