What did we do ?
This test was first published on our site 1997. Back then there were only one or two other samplerate converters of any quality at all. Things have changed. More samplerate converters seem to come close to Barbabatch's these days. A new test was required. This update of the test uses a different conversion. We test the conversion from 96 kHz-24 bits to 44.1 kHz-24 Bits.

The original file is a sine wave sweep that goes from 1000 Hz to 48000 Hz in 2 seconds. Converted down to 44.1 kHz, there should be no audio where the original sweep passed 22050 Hz. If there is audio in the 44.1 kHz after that point, it is aliasing. Aliasing is unwanted nonsense audio that spoils into your clean required audio as a result of not filtering away frequencies that cannot be reproduced in the new samplerate. A good converter filters in order to prevent aliasing. The shape of the filter that takes care of deleting anything above the highest possible frequency in the destination samplerate can de judged by looking at the waveform of the 44.1 kHz file, which is what we do on the right ->

Criteria for this test
The anti-aliasing filter slope should be steep for two reasons: If the slope is not steep it normally starts attenuating too early and you loose too much high frequencies in the conversion. If the slope extends beyond 22050 Hz too much, you get aliasing. (See also a closer look at ProTools Tweakhead)
The filter should attenuate very severely. It should push away the 22050-48000 Hz content by at least 96 dB in 16 bit situations, and at least 130 dB in the 24 Bit domain.
The audio should remain clean. You should not hear any sidebands traveling alongside or backwards through the sweep tone. Distortion, calculation jitter, rounding errors and aliasing will all result in these sidebands.

Is this important for audiophiles only ?
No. The differences are not subtle at all. Even when you produce low bandwidth audio, samplerate conversion quality is equally important. For example: a bad samplerate converter will make the 's' and 't' consonants in speech blast out of a telephone at 8 kHz samplerate.

This test is not conclusive
There are more factors that can be tested. We could have shown you impulse responses of the samplerate converter that demonstrate better what happens to percussive sounds. We could have zoomed in on general noise issues related to calculation precision. We could have added more converters (and we may when we get requests for them).

So what's the catch ?
CPU. Processing at the precision of Barbabatch's samplerate converter takes a lot more calculations. Tweaking the algorithm in such a way that it is not slower than the next option, takes lots of engineering. Logic is quicker than BarbaBatch. BarbaBatch performs the conversions to the right in about 7 % of the length of the sound file on an Apple G5 1.8 MHz. There are lesser qualities available in BarbaBatch if you need more speed.