32 bit float - deamplifying peaks results in flattened peaks

I’ve read that when recording in Audacity’s 32bit float sample format, peaks that extend beyond 0dB can be deamplified without any cliping.

I’ve been attempting this, but when I deamplify, the peaks are flat. The recording is fresh, newly captured. (I haven’t exported the recording or anything that could have changed it from 32bit float).

I’m on a mid-2011 iMac. I can’t find any specs that reveal what its sound card’s abilities are but the MIDI preferences include 16bit, 24bit, and 32bit float. (I assume it is capable of these settings or they wouldn’t be present or they’d be greyed out.)

So, I have the iMac’s MIDI’s preferences set to: source (line in), 48kHz, 2ch-32bit Float, Hardware Rate Converter set at “Disable”.

Audacity’s (ver, probably via the dmg installer) preferences are set to: Stereo, 48kHz, 32 bit float.

Audacity’s dropdown settings are set to: Core Audio, Built-in Input, 2 (Stereo) Recording Channels.

The audio coming in via the iMac’s analog line-in is from an analog cassette deck (I’m digitizing cassettes). Usually I get the levels right so there’s no need to deamplify, but, for those times when I judge incorrectly and come back 45 minutes later to find peaks are too high it seems like the ability to deamplify to 0dB would be very useful! It just doesn’t seem to work on my setup for some reason.

Maybe I need to try changing the iMac’s MIDI “Hardware Rate Converter” from Disable to Enable? (I don’t know what that setting does - or even if the iMac’s MIDI controls are a factor at all).

While I am experimenting here I thought I’d ask on this forum as well, since it could be something obvious (to someone else)!

Thanks for any ideas.

I’ve read that when recording in Audacity’s 32bit float sample format, peaks that extend beyond 0dB can be deamplified without any cliping.

Not exactly. Inside Audacity, any effect or filter can be applied without permanent sound damage. Outside of Audacity, the old rules apply. Anything over 0 gets whacked off.

So you still have to pay critical attention how you record the work.


32-bit float is unheard of for sound cards. It may deliver the audio data in 32-bit float format, but the conversion from analog to digital will be 24-bit integer at best, of which 22 bits at best will be useful data (Even for professional quality sound cards, the last couple of bits are just noise)

By the way, this is also why self-contained, home-style USB microphones always record low volume. The instant you go over volume and clip, you’re dead.


Remember which show was recorded loud and use that to set the volume for all the works. Cassettes have very limited volume range before running into their own noise problems, so this isn’t a dreadful way to set the system. The system volume range is almost 100% certain to be wider and better quality than the tape, even if you miss perfect accuracy slightly.

I’ve been known to use the Mac Stereo Line-In for paid work and I recently bought an older design MBP because it still had Stereo Line-In. There’s nothing wrong with it.


So are you suggesting that the peaks above 0dB are being flattened by the sound card before they arrive at Audacity?

If so, under what circumstances is anyone able to take advantage of Audacity’s 32 bit float’s ability to store beyond-peak information?

Thanks everyone for pouring in with advice to quickly btw!

This is one of the features I like about Audacity’s 32bit floating point processing for remastering poorly mastered music files off commercial CD’s. I can apply extreme EQ curves to boost bass, add reverb, noise reduction and adjust pitch which most of the time will show the red clipping bars. I just reduce amplification and no more red clipping indicators and no flatness in waveform peaks. To bring back volume I use a limiter set to 6db increase. Works really good.

However, when I attempt to do this in Garageband '11 with it set to 24bit 44kHz capture/output, applying compression, EQ, etc. to these CD files, I have to adjust volume down or apply it default Normalize preference (which I turn off) or else get permanently clipped/flattened waveforms opening the saved aiff file in Audacity. Garageband has great live editing tools but very little analysis of what it’s doing to the original waveform regardless of source. Garageband’s 24bit high quality capture, processing and exporting engine is more difficult to control due to this lack of interface analysis.

That confirms my experience using a 2000 Pismo Macbook using “Coaster 1.1.3” to record high dynamic range Electronic styled music off my Time Warner Cable digital music channels back in 2003. Coaster gave a very accurate representation of the clipping indicators from the line in stereo source so I could make gain adjusts to stay under the red bars. I burned these high energy (meaning VERY LOUD) songs on a CD-r and I play them in my car and they sound amazingly good. Big beat subwoofer bass and crisp detail. Wave Stats indicates -5db/-17RMS. No clipping.
2003 -5db-17RMSBIGASSBASS.png

Just for fun on what is possible recording through Mac sound cards and line in source see the Spectrum Plot of the subsonic wave below 50Hz. Just for giggles and to confirm the soundwave physics of ported subs no matter how expensive (sealed subs are required) I played this portion of the song shown in the plot on an expensive Sony system at Best Buy and you can’t hear anything below 50Hz. I can on my sealed Walmart subwoofer boxes using Craig factory outlet component 10" subs in the trunk of my sedan I bought from Madisound online. I believe “Coaster” back then recorded at CD standards. No 32bit floating point. I can’t be sure though.

Ah, I think I am “getting it” now. “Outside of Audacity” includes the sound card itself, before the sound is passed to Audacity. So if input levels are too high in the sound card, the tops are “whacked off” by the sound card before it reaches Audacity (which makes sense now that steve said that sound cards don’t exist with 32 bit floating point).

Basically, I really just need to keep the computer’s levels down so that nothing gets clipped.*

Audacity’s ability to not lose the tops is merely something that works to ones advantage when one is in post-production, adding effects or whatnot (to a file that has been saved in the 32 bit floating point “other compressed format”) that may push a proper-level recording above 0dB.

*Koz, you say “the system volume range is almost 100% certain to be wider and better quality than the tape” - so, even with 16 bit recordings (of analog tapes), the quietest passages are still as good as the original analog - even if the levels of the input have had to have been set down to, say, 75% in order to prevent the louder passages from clipping? The cassettes I’ve been digitizing are interviews, and typically the interviewee is near the mic while the interviewer is far away and therefore very quiet. Sometimes the conversation is relatively sedate - everyone remains calm - and the levels can be set at 100% without any peaks shooting out above. But sometimes the interviewee gets worked up and starts shouting or whatever, so I have to transfer that entire conversation at, say, 75%. And doing it at 75% makes me concerned about the quiet parts - I’ve worried that at some point in the future, some future editor will want to amplify the quiet voice and they will be very upset that the digitization was only at 75% instead of 100%. In my mind that lower level capture would mean the quiet bits hadn’t been digitized at their best possible fidelity. But as you can tell, I am not an audio engineer so my notions are rather uninformed. In my noob-ness I’ve even considered making a second pass of every cassette at 100% just for the quiet parts - so that the quiet parts will be available (to future editors) at a louder volume, and that capture can be mined for the quiet parts while the 75% pass can be mined for the louder parts.

Had to make a correction. It was the mac app “Coaster” version 1.1.3. I made the corrections in my previous posts.

Yes, though with some hardware setups clipping could occur before the signal even gets to the sound card. For example, if you were recording something via a mixing console and massively overloaded the mixer input, then the mixer channel preamp would clip the signal.

While your working with the audio (editing and processing).
Tim Lookingbill gave a good example - when using an effect (such as the Equalization effect), the signal could (depending on settings) be boosted higher than 0 dB. If the audio data is an integer format (16 or 24 bit) then the audio would be permanently damaged. Integer formats cannot go above 0 dB, but if it’s 32-bit float, then the waveform above 0 dB is still there undamaged, so you can simply attenuate (amplify it by a negative amount) to bring it back into the “valid” range below 0 dB.

Without that 32-bit thing, applying effects and production would be a nightmare.

OK, We’re going to equalize to boost the crispness of a reading. First step, reduce the volume of the work to where you think it needs to be so it doesn’t clip after you apply the filter.


Watch me and get off the bus one stop before I do.

I started using a new process for AudioBook reading and mastering. The first step is a tool that changes the volume of the reading to meet audiobook standards. Then run Limiter to bang the tips of the blue waves down so Peak standard is met (overload, basically). It’s not unusual for the peaks to go over 0 after the volume step. Soft Limiter is a tool that rescues the normally illegal peaks and gently pushes them into compliance.

If your noise is OK, you’re done and nobody can tell we did anything to it.

32 floating is just the handiest thing.

However, no good deed ever goes unpunished, so you have to pay the piper when you export your work. 44100, 16-bit, Stereo (the outside world) is less accurate than Audacity internal, so a dither signal is added to the work to keep the downconverting errors from lining up and becoming audible.

And Do Not try to export the work while some of the sound is above 0. The destination sound formats will whack them off permanently.


Thanks everyone this was extremely informative!!!

You might find if it does exist some type of software that maintains constant volume levels like cable TV set top boxes provide when watching movies that have scenes of loud explosions followed by quieter scenes of actors talking. Maybe Steve and Koz know of a plugin. Maybe check your system prefs for some type of “maintain volume level consistency”.

Before boosting those calm quiet parts recorded to cassette with a mic feed, you might check if this raises the noise floor especially tape hiss by first listening and raising the volume on your system. This is the amazing thing about digital editing (amplify or limiter effect) vs analog gain (through capture software i.e. Audacity/Coaster) in that digital does a remarkable job of replicating the sound nuances when increasing analog volume. I do this volume increase test using the OS volume slider vs amplify or limiter listening on headphones a lot in Audacity on my MacMini. Sometimes depending how good my EQ curve sculpts the sound digital volume increase sounds a whole lot better than raising the OS system volume slider on the unedited version.

When I used Coaster to record music off the cable TV line out I could raise and lower the gain to bring up the quiet parts live as I listened on headphones on the Powerbook. I really watched clipping indicators intensely. That Moorcheba example I posted previously has noticeable volume increases I can hear on the CD-r in my car. It wasn’t really that much. If you can listen to your cassette live as it is being recorded in Audacity checking gain/clipping levels it may or may not be difficult adjusting gain in the quiet parts. I remember with Coaster it was quite sensitive to being clipped making live adjustments this way. Had to keep my finger on that slider watching for red clip indicators. It was tedious though.

Try recording in Garageband using high quality 24bit setting in preferences/turn off Normalize and all effects in both Instruments & Master Track and do a live volume increase. Don’t know if you can do this live in Audacity.