16 bit import file / un-dithered export?

Hi! I’m looking to import & do some simple edits on a bunch of 16-bit audio files in Audacity. Two questions:
As these files are already in 16-bit format should I set prefs in Audacity to 16-bit (instead of the default 32-bit) and, if so, do I just set dither to ‘none’ when I’m done? Thanks!

Leave the 32-bit default and set dither to “none”.

Great, thanks! Just a thought though, if these audio files I’m working with are pre-formatted as 16-bit (ie: set) but I import them into Audacity & leave the default setting at 32-bit floating am I not upping the resolution to 32-bit at the importing stage? If this be the case then would I not need to ‘dither’ the file back to 16-bit on export? That’s just the way I perceive it this end but apologies if I’m not being logical here - I’m new to all this!

PS: Let me try to explain what I’m trying to do exactly. I have a bunch of songs that were recorded to a standalone multitrack digital music recorder. The recorder is quite old now (but still works) and was manufactured when 16-bit was the highest standard resolution - there was no dithering involved at the final stereo mixdown stage. You’d just mix the tracks down to a single stereo file and either burn said song directly to the onboard CD burner or there is/was the option to transfer to a PC (USB 1.0 = SLOW!) I have a load of tracks recorded this way that I’m looking to collate into some kind of album (my ‘Greatest (non!) Hits’ if you will!) however … they need tidying up. As they stand the recordings are a bit tatty, especially the starts and ends. I’m just wanting to: sort out the gaps; edit out false starts; tidy up some of the endings; maybe the odd fade in/fade out if required; that kind of thing THEN export these ‘cleaned up’ files and burn the entire collection as some sort of D.I.Y LP compilation directly onto CD to distribute to friends/family (maybe even stick up on the web somewhere as a showcase of my songs?) That’s the plan anyway!

SO … getting back to my original post here, by all means correct me If my thinking is wrong here but, my train of thought is thus: ‘If these files are already in a non-dithered 16-bit format to begin with and I plan for them to remain in 16-bit upon export is there any real reason to ‘up the resolution’ as it were (32-bit) when ‘tidying up’ said files in Audacity and, if so, then surely I would need to introduce dithering to export them back into 16-bit format?’ I hope this makes some kind of sense!

If you import a 16-bit integer file into Audacity, and leave the default setting to 32-bit float, then the audio is converted from 16-bit integer to 32-bit float on import. This conversion is 100% lossless and no dither is required or applied.

Audacity works internally at 32-bit float precision.

If a track format is 32-bit float, then when processed by an effect, the processing is done at 32-bit float (extreme precision) and the data returned to the track as 32-bit data.

If a track format is 16-bit integer, then processing still occurs at 32-bit float, but the data is then converted back to 16-bit when returned to the track. This is subject to the dither settings in Preferences, so either the returned data is dithered, or (if dither is off) the sample values are rounded to 16-bit integer.

Thus, keeping the default sample format as 32-bit float is preferable as it ensures that conversion loss does not occur more than once (at the Export step).

If the input file is 16-bit integer and Audacity preferences set for 32-bit float and dither = none, and if the track is edited but not “processed” (no operations that change sample values), then there should be no loss of sound quality because:

  • The conversion from 16-bit integer to 32-bit float is 100% lossless.
  • The 32-bit float sample values can be converted exactly to their original 16-bit values.

Brilliant (what a reply!) Thanks so much for that! So, just to make sure I’ve got this right in my head … As a practical working example:

  1. Firstly I make sure Audacity is set back to its default setting of 32-bit. (I had previously set it in Prefs to ‘16-bit’) then …
  2. I import my 16-bit audio file into Audacity and ‘tidy up’ the file. (For example: Chop off any unwanted silence before the track starts/ends.)
  3. When I’m done I then export said file. I choose ‘16-bit wav’ but I DON’T dither.

Is this right?

Exception to rule: I would choose to activate the dither option were I to have performed any say ‘fades’ or done any ‘normalising’.

Spot on.

Just to add; if you forget to turn dither off, it’s not the end of the world - “shaped” dither noise is virtually inaudible except during silence, and silence could be trimmed or gated later if found bothersome.

BRILL! Thanks again. One last thing … thinking on, I am probably going to need a few fades here & there on this batch of audio files (plus the jury’s still out where ‘Normalising’ is concerned, out until I can sit down & find the time to analyse/hear the files together & investigate further. Thing is, I do actually have the option of doing, for example, fades & any normalising on the actual recorder itself BEFORE bringing them into Audacity. Hmm …? I’m trying to decide whether I should do this kind of ‘processing’ on the source files because I’m aware that, were I to start processing the files with Audacity I WOULD need to set dithering to ‘ON’ right?

PS: Is Audacity ‘good enough’ for sequencing a commercially releasable music album would you know? I’ve never released a self-recorded album before and need all the help/advice I can get!
PPS: Perhaps biting off more than I can chew here but I am now (tentatively!) looking into a service called CD Baby who distribute independent artist’s music … when I’ve sorted out this aspect of the production (post-production?) that is! PHEW! This ‘little project’ is beginning to grow arms & legs. A BIG thanks again for all the help in this area so far - I’ll get there!

As I hinted at in my previous post, dithering is a very subtle effect, and mostly inaudible. Where it is most noticeable is when the audio becomes extremely quiet (below about -70 dB), for example at the end of a slow fade-out. If you make a long, slow fade-out, and listen to the tail end with the volume turned up LOUD:

Without dither: The sound starts to become a bit “crunchy” due to quantization distortion, then “crackly” as the sound judders between “sound” and “no sound”. At -90 dB, the sound turns into a series of clicks.

With dither: A slight hiss becomes evident at about -70 dB, but the “sound” remains clear and not distorted as it gradually sinks below the hiss. The fade-out will retain a natural sound well below -100 dB until it is finally inaudible.

If you perform the fade-out on the recorder, assuming that the audio data is 16-bit integer, then either the recorder applies dither or it doesn’t (we don’t know, but I’d guess it doesn’t), so you will either get dither noise, or quantization noise (probably quantization noise). I’d recommend that you do the fades (if necessary) in Audacity - at least you then have the choice about dither.

Yes, absolutely. I’ve used Audacity for post-mastering corrections prior to recordings being commercially released. The 32-bit float format that Audacity uses by default is “extreme” in terms of precision / quality.

CASE SOLVED … ALMOST! Just one very last question (and then I promise the penny will have finally dropped and I shall crawl back under the rock I emerged from!)

When I first asked my question here I was thinking that I would simply be chopping the start & ends of my sound files and that was that HOWEVER (!) now I’ve got down to t’the nitty gritty’ of it all physically importing then trying to work with these files (16-bit remember) I have realised that each individual song file I have tinkered with so far needs a little more attention than I first thought ie: usually requires at least one fade PLUS I have also decided that I’m going to Normalise the lot of 'em too due to the fact that I now realise I want them all to have a consistent level (“PHEW!”) SO, my FINAL question on this topic (promise!) is thus …

If I’m importing 16-bit resolution files into this (32-bit floating) Audacity program AND I’m applying blanket Normalisation (plus one fade on average) to each of the sound files I import should I:

A) Leave the the Dither option set to ‘NONE’ as I have been doing so far or, conversely …
B) Perform Dithering on these 16-bit files?

If the answer is the latter (B) then should I be using: ‘Shaped’; ‘Triangle’ or, erm, the other one (which I can’t remember the name of right now!)?

Like I say, once I have this cracked I can look forward to properly getting on with the job in hand (when I finally know what I’m doing that is!) Apologies for my continued questioning here it’s just that I’ve never used Audacity before and, as you’ve probably gathered by now, I’m quite stupid too! Anyway, thanks (again!)

I bet it’s not :smiley:

Be careful here. Having the same “peak level” does not mean that the will sound the same “loudness”.
“Peak level” is, as one might expect, the level of the highest (or lowest if it’s pointing down) peak. The Normalize effect amplifies the selected audio so the the biggest amplitude (biggest peak) just touches the level that you normalize to.

“Loudness” is more complicated. How loud something sounds is subjective, and includes such matters as frequency content, whether it’s a long continuous sound or a short sound surrounded by quieter sound, whether it’s a pure tone or “noise” and other factors. There’s a good article on Wikipedia about loudness if you’re interested: https://en.wikipedia.org/wiki/Loudness

As a rough rule of thumb, the “RMS level” (a kind of “average” level, shown in the Audacity waveform as the light blue inner part of the waveform) is a better approximation of “loudness” than the peak level.

…put your settings back to the default: 32-bit float, shaped dither? Yes you should.

Ahh! Got that wrong then (I had a 50/50 chance of getting it right!) What am I talking about? Well, I’m (just) two files in (thankfully!) and on both sound files I had made the executive decision NOT to dither! Ahh well, should be easy enough to rectify! Anyway, I’d just like to take this ‘final’ (LOL!) opportunity to thank you all for your help here - it’s been much appreciated!

Well described :slight_smile:

Here’s an article that you may find interesting and informative: https://en.wikipedia.org/wiki/Loudness_war