Tag: 16 bit audio sample

24 bits vs. 16 bits are there audible differences?

The experiment

The author of the experiment plans to determine if there are audible differences between 24 and 16 bit audio trying to use an objectivist approach.

For this, it selects 3 audio fragments at 96/24, 2 of them edited by the renowned 2L label and a third belonging to the well-known Goldberg Variations performed by Kimiko Ishizaka. Particularly difficult passages are selected for their musical content or dynamic characteristics.

16-bit versions are generated from the 3 original samples, maintaining the sampling rate of 96 khz. Simple dithering is used in the process. In addition, special measures are taken (addition of noise at -140dB, offsets, use of flac compression with more or less efficiency) to avoid that the characteristics of the files themselves could reveal their nature.

The resulting audio files maintain a DR of between 12 and 14, with no evidence of clipping or compression of the dynamics.

The 6 files are arranged so that both versions of each of them are sequentially reproduced, but it is not known a priori if the 16 or 24 bit version sounds first.

In total 140 people participate, who must:

– Determine when one or the other version of each fragment sounds.
– Indicate the degree of certainty with which they have responded.
– Explain if they have used an ABX type comparison tool or similar.
– Provide a series of data, let’s say demographic, including whether they are musicians, have professional experience or not in audio editing and if they regularly analyze equipment.
– Describe the audio equipment used in the tests and indicate their cost.

The test runs for 2 months. Participants are encouraged to take their time.

Participants

Most of the test participants:

– They are men.
– Between 41 and 50 years old.
– They are not musicians (although 22% say they are).
– They have no experience in music production (although 24% do claim to have it).
– They use equipment whose cost is between $ 1,000 and $ 3,000.

The results

– The hit rate is practically 50%. Or what is the same, pure fluke.
– Participants who claimed to be more sure of their answers were not more correct than the rest.
– The group of “musicians” has failed more than the average.
– The group of “audio engineers” has made it a tad better than the average, but without exceeding 55% of hits in any fragment.
– The group of audio equipment analysts has also done worse than the average, but the size of the sub-sample is considered low (8 participants).
– Participants with expensive equipment (44 people with equipment over $ 6,000) did no better than average.
– The use of headphones has not been an advantage.

The conclusions

I translate more or less literally:

In a real study carried out over 2 months in which 140 people have participated and in which high-quality sound samples obtained from digital audio at 96/24 have been used, no evidence has been found that Digital audio with a 24-bit depth could be appreciably differentiated from its corresponding 16-bit versions obtained through Adobe Audition 3 using a basic algorithm (1/2 bit flat triangular dither).