What's the best WAV format to export?

I was working on a new project in which I imported a few audio files that were 32-bit, 48,000 Hz. I figured the export should match that of the audio files, so I exported the tune as 32-bit, 48,000Hz. WOW… was the output horrible and scratchy (playback was in Amorok). It seems like I’m limited to 44,100, 16-bit for a successful “clean sounding” export. This can’t be right… right?

My philosophy in this comes from third grade math, or so: any teacher will fail you if you start rounding before the end result.

From what I understand, when you summarize a number of signals it’s important to have a big dynamic range, aka headroom, to avoid getting 0dB clipping, regardless of music genre. I assume that’s why Ardour uses 32bit float as the default, and I see no reason to change it unless you’re short of space.

And similar reasoning on sample rate; the more accurate each individual track is the more accurate the summarized waveform will be. So my recommendation is 48kHz or higher.

And libsamplerate is among the best resamplers out there so I don’t think you lose that much in the conversion down to 16bit/44kHz

But, to return to the original question, not being able to play 32bit/48kHz is either a bug in Amarok, it’s not linked against libsamplerate or you need to configure it to do a proper downsampling. On this 3.2GHz P4 with an intergrated ATI AC’97 chip I can’t use the best samplerate setting (Best Sinc Interpolator) in Aqualung, I have to use the Medium Sinc setting to play 32/48.

I was just reading an article on Tweakheadz that mentioned the very same thing; That is… if you can record at 44.1/48kHz @ 24-bit, then do it. Although the human ear probably won’t make a distinction, its better when re-sampling down to 44.1. sigh Perhaps the next round… as I have recorded everything so far at a sample rate of 44.1 @ 16-bit (other than that little test I did to kick off this thread).

One thing to bear in mind, though, is that tracking at 16 bits is a /lot/ harder work than at 24, because you can’t afford enough headroom in order to avoid clipping.

If you try to leave 10-12dB of headroom at 16 bits, you’re unlikely to get a hot enough signal, so you’ll lose a lot of low-level detail, leaving everything a bit dull-sounding, and if you try to record any hotter, you run a risk of ruining the take because of digital clipping.

That’s an endless debate isn’t it?

Another angle is how good the dithering algorithms are when you downsample from 24 to 16 bits a poor dither-er could completely eradicate any benefits of capturing at the higher sample rate. I can almost guarantee if you did a blindfold test with “A” being a 24 bit recording and “B” being a 16bit recording with a bit of presence added or through a BBE or phase aligning effect people would invariably pick “B” as the better recording.

The greatest benefit of 24bit would be dynamic range…useful for Chorale, Jazz, or Classical pretty much a non-issue for Rock based music. My opinion would be to use 16bit 44.1khz and stick with whatever your original A/D converters gave you through to the end of the project if the destination is CD or mp3.

5.1 DVDAudio may warrant a higher resolution approach but music seems to be getting smaller instead of larger…

Just my $.02


the wordlength (bit) gives you more dynamic range and less noise if set to 24 bit. That’s important if you do a lot of compression etc.
The helpfulness of higher sampling rates than 44,1 khz is still being discussed, but it seems like a higher sample rate (e.g. 48khz) puts the bad artefacts of filters and converters into inaudible regions on the frequency range.

A lot of studios still record and work with 44,1kHz, but in every case with 24bits (and in the DAW with even more like 32bit or float). So I’d say 44,1/24 is save (I use this), if you can afford the disc space and the throughput, higher sample rates won’t hurt.

There’a a lot of literature about it, I recommend Bob Katz about this topic :wink:



yes, you must feed audio-cds with 44,1/16 files. But remember that there are benefits if you record and work with 24bit or higher files. Resampling down to 44,1/16 should be the very last step (before cd-burning) of the whole production.


Err… Nvrmnd. Looks like WIKI answered my question. Since CDs can only be played at a sample rate of 44.1kHz @ 16bit (and most home audio systems can only play that sample rate), then there’s no use investigating other sample rates. I suppose 48kHz @ 16bit is becoming more acceptable for digital audio files and DVD formats.

Many years later dropping a word here… this is what I do:

Exchange between programs during production: 32bit float at constant sample rate. No dithering. If 32bit float not possible, 24bit at constant sample rate (e.g. true for stem exchange between my DAW and NLE for me).

Final export to 16/44.1 for CD, additionally 24/48 for video sound. Which is 48khz in almost all cases. Since I use 48khz in recording also, it’s a very convenient workflow.

Some people say one should use 48khz in AD/DA and 96khz in processing, which is not possible in Ardour to my knowledge.

Is here a simple answer? For me it is: Try not to change the format in production, so 32bit float and project sample rate. Convert the format in the very last step to your outlet (most likely to be 16/44.1 for CD and 24/48 for video). I always do this conversion in Ardour because I don’t trust other tools.