Tír Na HiFi

Thanks much Gordon!!

sbgk wrote:things I am working on

24 bit version - has issues hence the delay - will probably go back to a sse4/2 assembly version for this
render loop code
squeezelite and portaudio to learn about KS
until I hear an optimised KS version I won't know if it's worth spending any more time on wasapi

........and who says now that Santa doesn't exist?!!! My best Christmas present (please don't tell my wife)!

Thanks Gordon

The third part of a Q&A series with John Swenson is up on Audiostream, this time where he attempts to explain why or how even "bit perfect" software can sound different from one another. Would be interesting to see what sbgk thinks of his ideas:

http://www.audiostream.com/content/qa-j ... fect-sound

internethandle wrote:The third part of a Q&A series with John Swenson is up on Audiostream, this time where he attempts to explain why or how even "bit perfect" software can sound different from one another. Would be interesting to see what sbgk thinks of his ideas:

http://www.audiostream.com/content/qa-j ... fect-sound

he's coming at it from the hardware angle because that's what he understands. Why do I hear differences when using the laptop via headphone out ie no usb involved.

to me what he says about cache misses is wrong, cache misses are prevented by prefetching the data into cache close to the cpu, extra loops caused by optimisation in the compiler don't affect this, at least not in the code I've seen.

With MQn individual instructions were tested for their effect on SQ eg the sub instruction sounded worse than the add or dec instruction, why that is the case is down to the code that runs on the cpu when these instructions are called, something we have no control over. Which is why I think passing the data through the cpu puts a limit on the SQ.

bit-perfect is worthless
try time-perfect

jesuscheung wrote:bit-perfect is worthless
try time-perfect

the problem is the extra threads required to alter the timing affect the sq, so the cure can be worse than the disease eg jplay, xa

just cut the screen on my usb 2 connector from my external drive to the pc and applied a couple of ferrites, makes a real difference to the sq ie reduced sibilance and it's free.

for one my of harddisks, i find turning it upside down (green side facing up), sounds best. lol

after a while, i realize the less number of hdds, also the less number of drives (c, d, e etc), the better is SQ.

sbgk wrote:

jesuscheung wrote:bit-perfect is worthless
try time-perfect

the problem is the extra threads required to alter the timing affect the sq, so the cure can be worse than the disease eg jplay, xa

impossible to dedicate audio processing with PC of just one CPU. you can change affinity of processes. but how to alter affinity of IRQ, interrupts, drivers and kernel?

with 2 CPU PC, win2012 allows you more flexibility. but 2 CPU introduces other hardware issues. also cost is issue.

havnt tried ferrite cores yet but
all these things seem to matter quite a bit
spinning hard drives are a disaster especially if not externally powered
there are issues that cant be addressed within the software,
analogue engineering?

jesuscheung wrote:

sbgk wrote:

jesuscheung wrote:bit-perfect is worthless
try time-perfect

the problem is the extra threads required to alter the timing affect the sq, so the cure can be worse than the disease eg jplay, xa

impossible to dedicate audio processing with PC of just one CPU. you can change affinity of processes. but how to alter affinity of IRQ, interrupts, drivers and kernel?

with 2 CPU PC, win2012 allows you more flexibility. but 2 CPU introduces other hardware issues. also cost is issue.

not sure what the outcome would be but that would be a interesting build