A pet peeve of mine. On headphones the noise floor is the gurgling of blood and my tinnitus, but in the house I can't go as loud most of the time and the noise floor is higher. In the car all bets are off for most classical music. Why has there never been any work into coding a variable compression that can be determined at mastering time?
The real world problem with 16-bits isn't what you guys are thinking. Its the quality at which music is commonly downsampled and bit-depth reduced in DAWs that often makes it sound poor compared to the the original hi-res digital recording. The methods Scott Wucer thought up using a very large FFT of a whole piece of music may be a much better approach than the algorithms commonly used. However, even that can't fix the issue with many dacs wherein a dac itself may sound better at higher sample rates. Hence, the example Hans Polack did with different versions of processing all applied to an 88.2kHz dac sample rate may have sounded the same because dacs sound the same when operated at a particular sample rate. That's something Benchmark seemed to notice when they designed DAC-3 to operate the dac chip at only one sample rate. Some things to maybe think about, is all.
Other than audiophool specific NOS setups no DAC operates at an 88.2 sample rate, so no idea what you mean there.
As for bit depth reduction there must be examples of this going wrong?
As for bit depth reduction there must be examples of this going wrong?
Hello George.
I hear progressive differences in tracks 9>16 as follows..
The 16.44 file sounds veiled with the voice booth resonance sounding boomy and wrongly/intrusive boomy, lessening of expected natural sibilants and clarity in the announcers voice and a distortion in the voice.
Moving to 16.96 gives better sibilants and upper clarity in the voice but the intrusive boominess persists, differently but still there.
24.44 has cleaner voice but still curtailed highs in the voice, boominess still but different again, different 'enveloping' to the two previous files.
Moving up in sample rate clarity improves, distortion improves and unnatural enveloping reduces with overall sound still improving at 24.192.
Once in 24bit res and at 96k the sound is pretty good and it is the subjective envelope that improves with higher sample rates.
I reckon it is this digital unnatural enveloping that is objected to by analog enthusiasts.
IMO you don't need supersonic hearing to notice the enveloping differences and it does not need much change in signal phase changes to quite radically alter actual acoustic waveform even though FFT says no change in signal power.
These changes in wavefront slew rates and peaks amplitudes due to altered dispersion are interpreted strongly as change in envelope by our hearing but a bad system will mask these fine changes in envelope by introducing it's own dominating errors.
George how old are you and how is your hearing.....and your system ?.
Max.
Last edited:
George,
I think it is possible to discern music below the ambient noise level. Noise is random and music is organized. Just tried with a dB meter. Due to the almost shut down of this country, ambient noise is just around 34 dB right now in my living room, not far from the centre of The Hague. Eerie silence, but good for a small acid test.
I can detect music by turning up the level to a point where it does not register on the dB meter, yet does become audible.
Oh, oh, I thought we could close the 16 bits book, but it seems that I have opened the almost closed lid from the can of worms again and we start all over again.
Sorry for that.
Hans
Sorry for that.
Hans
How did you acquire these impressions? By changing track with the remote and had a few listening sessions true the tracks?
//
88.2 is a standard audio sample rate. Dac operation can be described in terms playing back that audio sample rate. Dac operation could also be described in other terms, but I thought everyone should be able to understand the number I used.
Its probably not the truncation, that's simple enough. Could be issues with dither. Some people hear it, even if its noise shaped. I once could, haven't checked recently. The NS-10s might make it easier than the Sound Lab speakers do, although the latter sound great.
In terms of examples of going wrong, the dither in AK4137 is added at the lowest remaining bit, not the bit below that was truncated off. Don't know of specific errors like that in DAWs, but there may be some. The effect in AK4137 is quite audible to me so I always disable the dither.
Last edited:
Yet you then quote the benchmark as if it does something magically different. For the majority of DACs if you play 44.1, 88.2 or 176.4 the DAC mimbles along at the same rate. Are you claiming some DACs sound better at multiples of 48kHz?
ESS dacs do (only) if operated in asynchronous mode. For synchronous 128fs mode they are supposed to be operated at a specific multiple of the frame clock rate, although they still work at multiples of that. Don't know if they account for the majority of dacs, but doubt it. AKM dacs have an internal MCLK divider to allow for different sample rates. So do ESS dacs, but theirs doesn't sound very good if used for dividing MCLK for synchronous operation, its really for power saving in mobile use.
No.
Last edited:
Stretching things a bit without some sort of numbers on how most domestic audio DACs are done. But worth throwing a question out in case the likes of Demian know how this is usually implemented.
I am impressed. Good for you!
I am the worst of all here (and my system-no matter how much Low Fi is-is much more capable of than my ears)
https://www.diyaudio.com/forums/members/gpapag.html
Actual hearing examination results of your hearing.
Vacu, please try track 22 and report results (I have verified my DR limit with other material too but this track is IMO good for general assessment) .
What is the lowest level that the display of your dB meter registers?
George
Thanks for that, it helps put your original post into perspective. Was the question not clear?
Very narrow view, the FFT contains the envelope information via the Hilbert transform which is simply a manipulation of FFT's.
This is your acknowledgement that it still needs to be dealt with. Thanks for your input.
When it's all said and done, speakers' interaction with the room still cause far more deviation from the input signal integrity than anything else in audio chain by many folds. Yet, there are those who worry about DACs and cables...
In most of the chips I have looked at (of recent vintage) the internal master clock runs at a constant multiple of the X1 sample rate (44.1 or 48). They all have some internal DSP that implements the reconstruction filter and adapts to the actual sample rate. Typically when I see different performance at higher sample rates it can be attributed to measuring across a wider band, not any fundamental difference, at least in recent implementations. While there have been some newer chips the economic driver has been cell phones where power consumption becomes paramount (as does cost).
AKM did tell me that some chips internal timing has been optimized for the 44.1 clock chain. I did measure slightly less HD on 44.1 which prompted the question.
The ESS approach with a sample rate converter at the input and some internal VooDoo "correcting" for nonlinearities. The bigger problem is that the nonlinearities seem to have a memory aspect so its not a simple lookup table.