downsampling from 24bits/48KHz. to 16bits/44.1KHz.

On 2005-04-29 21:22, thockin wrote:
Spindrift: what you're missing is the D->A converter's impact.

On 2005-04-29 21:49, Spindrift wrote:

So I don't know about any technologies where audio at very high frequencies is dependent on the waveform.

But one octave is a doubling of frequency. That means that a sound at 10k only can have one harmonic. So according to a Fourrier transformation you don't get much of a square out of one added harmonic.

Even down at 5k you have to settle for the one odd harmonic to shape your square.

I know that we don't really hear the overtones that would follow to shape the square.

But I think the increased resolution of the sound, regardless of what frequency you can recreate, can be percieved.

CD shure is at least slightly too low as a standard with 44.1.

I agree that bitdepth is more important but when you are up to 24 or 64 bit depth the next step to make a increase in the resolution must be to go up to 88.2 or 96.

The consequences of sharp filtering include time-smearing of the audio, possible short (millisecond) echos, which are caused by amplitude response ripples in the pass-band frequency response (20Hz to 20kHz) - even ripples as small as 0.1dB can cause problems. By moving the filter cut-off frequency to 48kHz (for 96kHz sampling), designers can use gentler filters and it is also easier to engineer filters with less ripple in the pass band. The result: purer tonality and more open sound in the 20Hz to 20kHz band. Researchers such as J. Andrew Moorer of Sonic Solutions and Mike Story of dCS have demonstrated other improvements from the higher sampling rate. Moorer pointed out that post-production processing, such as filtering, equalisation, and compression, will result in much less distortion in the audible band, as the errors are spread over twice the bandwidth — and half of that bandwidth is above 20kHz. I've already noticed that the equalisers in Sonic Solutions sound better at 96kHz sampling and, similarly, Daniel Weiss's double-sampling equaliser is very pure sounding. Conclusion: perform all your calculations at a higher rate, then down-sample only at the end."

I've never found a band-limited analogue system (transducer, pre-amplifier, etc.) that sounds as good as the equivalent wideband system. When I roll off the response of an op-amp to 100kHz bandwidth, it sounds worse than 200kHz!

On 2005-04-30 16:10, Spindrift wrote:
What I'm really trying to point out is that sample frequency is not only affecting what frequencies you can represent and bit depth is not only affecting how wide dynamic range you can achive.

Just try it and downsample a sound to 11k sample rate.
It will have an effect on how the bass and mid sounds as well as removing any treble above 5.5k and not sound like you only applied a low pass filter.

Converting a sound to 8 bit will not only reduce the dynamic range of the sound but make it more gritty as well.

We are still talking about resolution, much like when working with images.
A higher resolution will result in a more defined image, up to a certain limit where our eyes cannot tell the difference.

And the listening tests I have looked at confirms that many people can hear a difference.

According to Bob Katz for example:
"My experiences with 96kHz/24-bit sampled recording have been exceptional. The sound is more open, transparent, and dynamic than previous 44.1kHz recordings, with a purer mid range and more apparent depth and space closer to the analogue source."

Thats my experience when trying synths using 96k as well, and if it's explainable mathematically or not does not really interest me.