Sound quality compared between local and streaming

I conducted an experiment today to compare the sonic quality of tracks played using A+ from different sources. My setup is a late-model iMac running Mojave with a 3.6GHz quad core Intel Core I7 CPU, 32 GB RAM, a 500 GB SSD and my media is on an 8 TB LaCie Thunderbolt HDD. The DAC is a Beyerdynamic Impacto driving TP1 headphones. I also have an older Mac mini as a media server for my home theater, and I’ll try to test that setup another time. Because my wife has sensitive hearing, however, most of my critical listening is done with headphones.

I selected several tracks for my test spanning a wide variety of music types - including tracks by Bach and Tchaikovsky, Miles Davis, Weather Report and Wayne Shorter, Frank Sinatra and Willie Nelson, The Beatles, The Carpenters, David Bowie, Hozier, Weezer and Daft Punk. The only requirement was that all tracks be available on Tidal and Qobuz as well as in my library. The Qobuz and local tracks were identical in resolution - 24 bit / 44.1kHz, 96kHz and 192kHz. I also have Weather Report in DSD64 for comparison. All of the Tidal tracks were 16 bit / 44.1kHz, with a few of them also in MQA. I didn’t do random blinded testing, as the results were not subtle. The differences in sound quality were striking and in a way, surprising.

MQA vs Lossless: The first big surprise was that the lossless tracks sounded considerably better than the same tracks in MQA. The MQA versions sounded flat, lacking in depth, bandwidth and soundstage. However, that was not my previous experience with Tidal when I listened a couple of years ago. One major difference then versus now, however, was that when I listened before at work, my DAC was an Audioquest Dragonfly Black 1.5 driven by an iPhone 6s+, connected to a pair of Audioengine A2+ speakers. The Tidal iOS app is no match for Audirvana Plus and my iMac, and as good as the Audioengine speakers are, they’re no match for the Beyerdynamic TP1s. On the other hand, The Dragonfly has hardware MQA decoding, which probably makes all the difference. I suspect the software MQA decoding algorithm in Audirvana is no match for the hardware decoding of the Dragonfly, but Audirvana’s handling of CD-quality lossless music is unmatched.

Qobuz vs Tidal: This really isn’t a fair contest, given that the Qobuz tracks were all Hi-Res and the Tidal tracks were CD lossless. Granted, CD quality can be quite good IFF the mastering is done well, which is seldom the case, particularly with popular music. Even then, although the dynamic range of the CD is greater than 90db, it’s static. The human ear is dynamic and so our effective dynamic range is much greater. Further, it’s impossible to reproduce the cannon blasts at the end of Tchaikovsky’s 1812 Overture without losing the softest passages in the quantization noise of the smallest bits. Something has to give, and it’s usually at both ends of the dynamic range. Using 24 bits allows the full dyadic rage of the music to be reproduced if your sound system is capable of reproducing it. A 44.1 kHz sample rate is adequate to reproduce the full bandwidth of what the human ear is capable of hearing, but only with ideal filters. Oversampling and high-quality digital filters are more than adequate to the task, but a lot of popular CDs have been pre-filtered to oblivion, and the mixing utterly destroys any sense of a real soundstage. Of course there’s no advantage to Hi-Res if the mixing and mastering isn’t up to the task. But I digress. The bottom line is that the Qobuz tracks were significantly more open than the Tidal Tracks. I could definitely hear subtle vocal overtones and percussion sounds with Qobuz streaming that were lost on Tidal.

Library vs Qobuz: The most surprising finding was that the tracks from my local library sounded significantly better than those streamed from Qobuz, and the difference wasn’t subtle. As good as the Qobuz tracks were, they still sounded digital. The local tracks sounded as good as what I’ve experienced from the best analog recordings on reel-to-reel tape and vinyl, and they came close to rivaling a live performance. The dynamic range was full with vocal overtones and subtle percussion clearly audible and free of artifact. Individual voices and musical instruments could be clearly separated from each other and localized on the sound stage with pinpoint precision. The soundstage was broad and realistic. This was the reason I bought Audirvana Plus in the first place.

DSD vs PCM: With all of my previous software, which included Pure Music, Amarra Symphony and BitPerfect, I was convinced that DCD tracks sounded much more natural than PCM tracks, but Audirvana has closed the gap. I would be hard pressed to choose between DSD tracks and PCM versions of the same tracks in blind A/B tests.

Summary: Local tracks sound better than those streamed from Qobuz, which sound better than those streamed from Tidal. On my setup, MQA-encoded tracks actually sound worse than lossless 16-44.1kHz tracks. There is no difference between the acoustic quality of DSD tracks versus PCM. High-Res tracks played from my local library, regardless of sample rate, sounded as good as the best analog recordings, with excellent reproduction of vocal overtones and subtle percussion, clear separation of individual voices and instruments, and a broad, localized soundstage. Neither streaming service could match those qualities, even when the sample rate and bit depth were identical to those of my local tracks.

The Bottom Line: For the time being, I’ll still be buying rather than renting my music.

Hi,
I really like to read such reports and I think many of us
do such testing at home to find out whether or not the own
stereo setup is fine or could be improved. The latter is always
the case for me….
Thanks a lot StevInNYC for your efforts and for sharing the results with us.
And I say this because it matches my experiences totally. :slight_smile:)
So from my (very subjective !) view I can confirm your testing results.
But as I said before, looking for improvement and many discussions with my dealer and searching the internet, I replaced my MacBook as my library host and bought a pure network streamer (Aries Mini) for streaming and the included SSD for my music library.
Soundwise this was a big step forward for me, the improvement of the streaming quality was significant! And now, in my opinion, a computer (neither Mac nor PC) is the second best solution compared to a pure streamer.
I have a Qobuz hires subscription and what I hear is that library music is not better than streamed music, but what I hear often is also, that music from my SSD/library seems to be simply louder, not better.
And finally, I do not know whether it is good or bad news, my A+ is obsolete now, because the LightningDS of Aries fits my requirements very well.

Let’s continue testing, comparing, improving….

Regards
Franz

Disagreed. They are at least the same when the tracks are from the same release (the same mastering), and so many 44.1K tracks on Qobuz are newer mastered versions that almost always sound better than what I have here in my drive.

Agreed. I must say something is wrong with Audirvana’s MQA decoding or MQA itself is a horrible sounding format.

Regarding Audirvana’s streaming versus playing of local tracks, at first I thought I must be imagining a difference. After all, why should a streamed version of a lossless, high-resolution music file sound any different from a version stored locally. The files should be identical. Yet I repeated the experiment several times and obtained the same result. That I could tell the difference between the two in and of itself was strange. That I could correctly identify which was which 100% of the time pretty much ruled out chance.

The most striking difference was a loss of depth. The sound stage was significantly flatter with streaming than with the local version. With the local versions, I can pick out the precise locations of every instrument in the band. Listening to local files with Audirvana is similar to listening to vinyl in that regard, and I mean vintage vinyl played on a really good turntable - not one with a USB port. Listening to streamed music from Qobuz just isn’t the same. The music is still more precise than with 16-bit, 44.1kHz CDs, but the differentiation of sounds is more muddy and localization is poor. I couldn’t understand how there could be such a blatant difference between sources when the files, the software and the equipment were identical.

However, recently I read something that explained why there is a difference, and why it’s not subtle. Quite simply, Audirvana avoids buffering. Nearly all other software for processing streamed music uses buffering, in which several seconds of music are downloaded in advance to provide a buffer against steaming irregularities. It’s hard to imagine not doing so, but it invariably involves diverting some cycles of processing power away from processing the music and sending it to the DAC. Audirvana is designed around minimizing interruptions to the signal path and so buffering is avoided. Instead, the bandwidth is slightly reduced until the music can be delivered without interruption. I believe that what Qobuz does and the way Audirvana handles it is that they stream sum and difference channels rather than left and right, much as is done with FM radio. Audirvana will then reduce the bandwidth of the difference channel first.The effect is to reduce spatial information content in preference to amplitude content.

However, with a fast, multicore processor, I think I’d rather buffer the music stream rather than deliberately degrade the music. It’s not the overall bandwidth that matters, as I have more than enough, but rather variability in bandwidth and the need to accommodate the longest delays in the signal path, even if they are infrequent. At the listener should have the option of choosing how much or how little to buffer. Then again, one does have the option to download the music before listening, which amounts to 100% buffering.