I think up/and downsampling is definitely a significant factor WRT to degrading sound quality. I deactivate real time oversampling in all music production / mixing etc. software because it changes the signal in a bad way. But I don't think it has anything to do with harmonics or spuria that clock in at - 160 db below signal level. I think what happens in the time domain because of the anti-aliasing filters is mostly responsible for the effect.
You can up/downsample a file in software and try to null it with the (inverted) original, there will something left that is actually audible, way above -160 db...
Got it. I think my earlier fft comparator screenshots caused confusion. They were meant to show that the Izotope which had >1M taps do indeed perform better theoretically than SoX with maybe 500k taps max.
What's interesting is the extreme improvement in sound going from Paul's excellent (both theoretically and subjectively) EQHQv5 Apo with 2k taps to external sampling via SSRC with 16k taps. Increasing filter length/taps is probably the easiest way to get better performance. I would hope for 1M+, mostly because we can and it does matter. Afterwards, we can maybe do some filter optimization, but it requires quite a lot of expertise in signal processing. I remain convinced that the longer filters provide superior performance and bring this project much closer to what one would hope TotalDAC could achieve.
Last edited:
Not sure if anyone remembers this post. Soren, you could be right that simple simulations may not reflect reality in this case.
Last edited:
You need to try the software oversampling.... and then tell us if the caps still matter....
Tbh idk if there's much harshness left to be reduced. At this point I would take attempts at reducing harshness very carefully and only implement a change if there's theoretical justification. The need for smoothness and sweetness can overcompensate and produce dullness instead. If you want perfect smoothness, try Soren's botched 1.20 beta filters.
Sounds like no need to worry about those anymore...
In case it's useful. I found it interesting that Soren chose 353/384khz as the intermediate frequency, any higher and we wouldn't be able to fully bypass FIR1 with our current I2S max sampling frequency.
Last edited:
I have no further worries about the hardware at the moment. Unless someone discovers an easily correctable problem with the dam1021, the design (with added vref caps, otherwise bass appeared way too thin and punchy) seems quite perfect to me now. Of course, adding more caps and improving various power supply might still help, but probably not by a long shot. I think further evidence of the efficacy of various advanced mods is needed, with the software OS implemented (or whatever improved version we might discover of it), for us to know its effectiveness. Maybe 1121's reclock design can make a difference, but there's no way to change it on the 1021 now. Is better optimized and more powerful software oversampling the new frontier?
Is there an easy way to test objectively and subjectively the difference between, say 16k tap and 500k tap in SoX? Also, it's sad, marginally, that my DTS tracks now sound unbearably harsh... Dolby Atmos only supports up to 16/48 - there probably isn't a way to edit the windows audio graph pipeline to add an upsampler at the end. Though Soren's ultra-soft 1.20 beta filters did sound quite good. For the movie-lovers
Is there an easy way to test objectively and subjectively the difference between, say 16k tap and 500k tap in SoX? Also, it's sad, marginally, that my DTS tracks now sound unbearably harsh... Dolby Atmos only supports up to 16/48 - there probably isn't a way to edit the windows audio graph pipeline to add an upsampler at the end. Though Soren's ultra-soft 1.20 beta filters did sound quite good. For the movie-lovers
Last edited:
Yes, it is a marginal tragedy.
Here is what a 10kHz sine @96kHz looks like under newnos. I remember there being speculation that nos wasn't quite nos. Well, it is.
Attachments
Yes. Only some people interpret this as a significant improvement.
Add to this 10M taps and the attraction becomes irresistible
NOS by itself is no better than EQHQv5, at least not by much if any. This is completely different. Also total DAC does use oversampling to compensate for high frequencies. Regardless Soren doesn’t believe in NOS and it does make sense objectively.
Yes. Only some people interpret this as a significant improvement.
Add to this 10M taps and the attraction becomes irresistible
How do we get 10M? That would be an awesome next step.
Also does your graph show that we’re bypassing fir2 as well? Would that be advisable?
I installed iZotope RX7 advanced but the SRC doesn't seem to work as a VST2 plugin... any ideas?... Also no one publishes their internal filter length figure...
And since Foobar2000 only supports 32bit VST2 there's no way it'll work with 64bit iZotope SRC even if the latter works as a plugin form somehow
Sox can only set one target output frequency it seems, not so convenient. Hopefully they implement a x8 in their next version. It's probably safe to assume that resampler-v uses highest quality setting in sox...
And since Foobar2000 only supports 32bit VST2 there's no way it'll work with 64bit iZotope SRC even if the latter works as a plugin form somehow
Sox can only set one target output frequency it seems, not so convenient. Hopefully they implement a x8 in their next version. It's probably safe to assume that resampler-v uses highest quality setting in sox...
Last edited:
I reposted a statement from Paul just last night in which he said the simulations are likely flawed. If we want real measurements it should be on the vref ripple?.. not easy to get to uV precision. Soren said he tested transistor mod but he didn’t like something about having to feed larger currents to keep impedance low. For the caps, I still think larger and lower ESR is better, seems that there is no evidence against it (in terms of absolute performance, frequency balance is a different thing). The best we can do is maybe find out how much location could matter and what capacitance is enough.
Have you tried software oversampling? It’s not on the same level with NOS or Paul’s filters.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.