Jeff

I've been struggling for some time with an issue of live sampling of a very quiet Chinese instrument guqin. I wanted to see if anyone can shed any light on this struggle.

I use record~/buffer~/play~ to do live sampling. Often because the initial level of the guqin is quiet, (even with my sound interface mic gain a a decent level) I end up with a sample that, when played back without normalization is fairly quiet. I use the normalize message into buffer~ to bring up the level, but have been frustrated with the inconsistency of this process of normalization to successfully boost the signal every time (in effect, sometimes it doesn't normalize, but stays the same. I have searched for a random sample that might have a high amplitude and could keep the other quitter samples in the buffer~ from scaling but consistently found nothing).

I also noticed, that with the same mic levels, a delay line I use in the same patch (tapin~/tapout~) has a much better record/playback level than the record~/buffer~/play~ sampling.

I have been exploring using a compressor to boost the signal below a threshold with the hopes I could record into the record~ object with a louder signal. But if I can understand why tapin~/tapout~ seems to have much better sample/playback levels built into it, and do the same with record/buffer/play, then I wouldn't need use a compressor.

If anyone might be able to offer insight or suggest some directions to explore, I'd be eternally grateful!

I use record~/buffer~/play~ to do live sampling. Often because the initial level of the guqin is quiet, (even with my sound interface mic gain a a decent level) I end up with a sample that, when played back without normalization is fairly quiet. I use the normalize message into buffer~ to bring up the level, but have been frustrated with the inconsistency of this process of normalization to successfully boost the signal every time (in effect, sometimes it doesn't normalize, but stays the same. I have searched for a random sample that might have a high amplitude and could keep the other quitter samples in the buffer~ from scaling but consistently found nothing). 

I also noticed, that with the same mic levels, a delay line I use in the same patch (tapin~/tapout~) has a much better record/playback level than the record~/buffer~/play~ sampling. 

If anyone might be able to offer insight or suggest some directions to explore, I'd be eternally grateful!


help-understanding-different-recording-levels-with-record-vs-tapintapout

My guess is that the normalization takes some time to occur so you may be getting the un-normalized signal sometimes. I'm at a loss to explain a difference in gain if it's really there. Have you compared the dry recorded signal vs the same signal taken from a tapout~?

In general, it's best to address gain problems like these at the hardware level. (Switching to a condenser mic, for instance)

My guess is that the normalization takes some time to occur so you may be getting the un-normalized signal sometimes.  I'm at a loss to explain a difference in gain if it's really there.  Have you compared the dry recorded signal vs the same signal taken from a tapout~?

In general, it's best to address gain problems like these at the hardware level.  (Switching to a condenser mic, for instance)


(You could also use normalize~ on your input, but it's going to amplify the noise floor as well)

Hi and thanks for the feedback. With the buffer normalization, if the normalization takes place, it usually does so in milliseconds. If not, it doesn't matter how many times I bang the normalize msg, the samples will not scale.

I haven't compared the buffer~/play~ and tapout~ signals, that's a good idea.

The strait adc~ to dac~ levels for the instrument are good, using a hypercardioid condenser lavalier. The tapin~/tapout~ are more or less at the same levels, whereas record/buffer/play is not.

a recorded signal is always as loud as when recorded; a difference between record/play and tapin/tapout should not appear unless there is some bug responsible for it.

@roman: the output could be different depending on the fractional part of the delay and the interpolation method. If the time in ms translates to a fractional delay of 22.5 samples, you'll get the weighted sum of sample 22 and 23 (assuming linear interpolation, which I think is still the default for tapin~) In the case of an impulse input (0,0,0,1,0,0,0) you'd get something like (0,0,0,0.5,0.5,0,0,0) out.

Not a huge deal for audio (though the interpolation method can have a subtle effect on the frequency content--this accounts for why not all notes have the same timbre in a tapin~ based comb filter), but could be important when you're delaying sample-accurate control signals. (If the click represents a clock signal and your code is explicitly looking for 1 and expecting that value to only last for 1 sample)

I'm very doubtful this accounts for OP's problem, however.

Not a huge deal for audio (though the interpolation method can have a subtle effect on the frequency content--this accounts for why not all notes have the same timbre in a tapin~ based comb filter), but could be important when you're delaying sample-accurate control signals.  (If the click represents a clock signal and your code is explicitly looking for 1 and expecting that value to only last for 1 sample)

Help understanding different recording levels with record~ vs. tapin~/tapout~