Influence of FLAC Compression Level on Sound Quality
In our previous series on computer audio (TAS Issues 218–221), we reported a sonic degradation when WAV files were compressed into the FLAC format. These results were disputed by those who claimed this was impossible since once the files were converted back to WAV, the files were bit identical. We made the same measurements on the digital files ourselves and obtained the same results. Yet we stood by our subjective results then and continue to do so now. The problem is that measuring digital files themselves has no bearing on the ambient jitter coming from a variety of sources generated within a computer, and it does not reveal audible sonic differences. Only when the digital information is transformed to analog by the DAC do these sonic differences become manifest. So we decided to ask whether we could make meaningful measurements using the subjective and objective methodology we described in Part 1 of this new series. The degradation that occurs when playing FLAC files over a range of different compression levels as assessed with these two methods is shown in Figure 1.
A serious degree of sonic degradation is heard when WAV files are compressed using FLAC. It appears to follow a regular, negative hyperbolic pattern, whether quantified subjectively or by the more objective technique of height measurement. These results agree with our previous finding that FLAC compression of a WAV file (at level 5 in dBPA) degrades playback sound quality (SQ) from a typical PC server (see TAS, Issue 220, Feb. 2012). Under the circumstances of this experiment, we found that the height method was significantly more sensitive than our subjective method. We also show the actual percentage compression at each setting within dBPA. Counterintuitively, it can be seen that there is no quantitative proportionality whatsoever between the degree of compression and sound degradation. We have also examined the effect of the so-called “U” or uncompressed setting on sound quality found on recent versions of dBPA. This setting is not the same as the “0” setting, and it produces a file about 1% larger than the original WAV file (data not shown). The sound quality of this file is still not equal to the parent WAV, but the degradation is close to the detection limits of either of our two methods, something on the order of about 5 points on our subjective scale and within 2 inches on our height scale. As for why this degradation occurs when using FLAC files, we can only speculate that real time format decompression somehow creates an extra drain on computer CPU resources that occurs concurrently with the digital- to-analog conversion step, presumably mediated by an increase in system jitter. Given the apparent audibility of pico-second and even femto-second jitter, perhaps it should come as no surprise that the ear can detect the effect of lossless decompression on SQ. The magnitude of this degradation might well vary with the number of other background processes running simultaneously on a given computer. The solution to this problem is to convert compressed files to WAV, store to the hard drive, and use these files for critical listening. This is what we have done for all of our high resolution downloads1. On the basis of these results, we strongly encourage all commercial download sites to offer the customer the purchase option of a completely un-manipulated WAV file or, for purposes of metadata convenience with some playback programs, minimally compressed FLAC files using the U setting found in dBPA, despite commercial resistance to following this advice.
To BB or Not to BB, That is The Question—A Do-It-Yourself Tweak
In the course of using the NuForce Ref 9 V3 SE’s, which are fairly small and light weight Class D amplifiers, we found it difficult to keep the amps in a stable position when using heavy cabling while balancing the amps on various cones. We resorted to placing heavy rocks or paver bricks on top of the amps to keep them immobilized. We recalled that VPI once made a device with a laminated iron core, enclosed by wood, called the “VPI Brick” that was supposed to improve the sound when placed on top of electronics. This effect was explained by the ability of the iron to attract magnetic flux lines and eddy currents away from sensitive circuits within the device. Many years ago, one of the authors did an experiment in which up to 7 VPI Bricks were placed on top of a then state-of-the-art Accuphase DAC, one at a time. A clearly audible improvement was heard with each addition of a brick up until the seventh one, where the audible effect was small enough to be questionable.
Recently while in a local big box store one of the authors serendipitously spotted packages of zinc-coated steel BBs in various-sized plastic containers made by the Daisy Corporation. The largest package contained 6000 BBs, weighed nearly 5 pounds, and sold for $9. The light bulb went on and we eventually bought several packages to test for sonic effects as well as for stabilizing our equipment.
Much to our pleasant surprise, the BBs seemed to significantly improve the sound. As a control we compared this effect against our rocks to determine if the effect was solely due to mass. The rocks did absolutely nothing for the sound. We then started adding multiple containers of BBs, one-by- one, to each piece of equipment in our signal chain (PS Audio Power Plant Premier line conditioner, DAC, BSG qøl processor, and amps) all to marked sonic benefit. We even slid a box under the power supply of our computer server. We ran another control in these experiments in which we replaced several packages of the steel BBs with ones containing copper BBs with the same number, diameter, and container size. In contrast to the zinc-coated steel BBs, the copper BBs degraded the sound regardless of placement. We believe this is likely due to an inverse Faraday shield effect of the copper BBs acting to reflect internally-generated electromagnetic fields back into the equipment chassis.
In order to determine the maximum BB effect, we added individual packages of BBs one at a time. For this test BBs were repackaged into small plastic tackle boxes (available from the sports section of big-box department stores). The containers varied in size from 4" x 7" x 11/8" to 3" x 5" x 1". After optimally loading up all of the other devices in the signal path (PS Audio Power Plant Premier, the BSG qøl processor, and the NuForce amps), we examined the cumulative BB effect on just the PS Audio PWD DAC. We incrementally added BBs at positions 1 to 7 on top of the DAC as shown in Figure 2A, and at positions 8 and 9 beneath (Figure. 2B). Since all the electronics were supported by Nordost BC Sort Kones, care was taken to shim the lower containers so that they were in close proximity but not touching the underside of the DAC, which can be seen in Figure. 2B.
For this experiment we chose to use track 5 from the Misa Criolla recording (ripped from the original CD and up-converted using iZotope Adv. v2 to 192 kHz/32-bit) since the position of the sopranos in the choir displayed a greater range of heights than the Chabrier or any other recording with which we were familiar. The results of this experiment are shown in Table 1. By using the Misa Criolla cut as our standard for comparison, we were easily able to identify height differences with each added pack of BBs. The height position of the sopranos above the midrange driver of the B&W 802 speakers is noted in column 2 and is reported both as the sum of the distance above the midrange up to the room height plus the distance folded forward toward the listening position along the ceiling (the center of the speaker midrange to ceiling height of 44" and the forward projection distance are also shown in brackets). The equivalent subjective SQ point difference is shown in column 3. Column 4 highlights some first impression comments about specific elements of SQ change that caught our attention during the test. A variety of other characteristic SQ changes, including harmonic, positional, and focus characteristics of specific instruments and voices were also monitored and noted during testing.
In this test sequence, the most marked improvement was observed with the first few BB boxes added to the DAC. As more boxes were added, the SQ improvement became more subtle. If it had not been for our familiarity with the music, it would have been harder to notice the incremental changes between the 7th, 8th, and 9th boxes, but even here there was a subtle improvement to be heard. On this piece of equipment the BBs never conferred a negative sound quality. The pitch and tonality of the music and voices stayed constant, whereas the focus, clarity, ambience, sweetness, and emotional involvement continued to improve with the addition of more BB boxes. At the same time a subtle harsh quality was decreased, which we could only recognize by its absence. Superficially, this effect on harshness might appear to be small. However, from the standpoint of being able to emotionally connect with the music, its effect in these experiments and others we have done is pernicious and more disturbing to the brain than one might at first realize. Overall, the total SQ improvement of adding all the BBs was clearly in the “very large” category and, according to the cumulative point scores shown in column 3 of Table 1, is in the same range as the difference in sound between a CD and its corresponding original high-resolution download or SACD. We know of no other tweak that produces a greater sonic return on one’s investment. From an emotional point of view, we believe this tweak benefits the sound of digital to nothing less than a remarkable degree. Whether this is large enough to convince analog proponents remains to be seen. All we can say is that the effect on DAC performance produces authentic goosebumps of pleasure.