[hatari-devel] Issues with DMA sound

Per Almered per at brainfish.net
Sat Aug 15 12:35:46 CEST 2009


Here are the votes from the Swedish jury: 8-)

Ok, so what I did to test this: I made a 110Hz square sample in Sound
Forge
and loaded it as an instrument in maxYMiser, along with 110Hz (approx)
notes
played on the YM chip at volumes $f, $e, $d etc all the way down to $0.

For the real STE, I had to make the machine play the notes several
times,
and at a low tempo, because the YM chip sometimes took some time to
regain
correct volume and other problems. The STE was sampled using a semi-pro
soundcard (M-Audio Delta 44, surprisingly good for its time and price)
at 44.1KHz, 24 bits.

AAAAAAAnyway. The final data collected from the real STE, Hatari 1.2,
Hatari
1.3 and STeem 3.2 was then normalized so that the note at volume $f from
the
YM chip was at -6dB.

The results:

(S = STE DMA sample, F-1 = YM volumes)
Errors of 0.2dB can be considered neglectable, but as we shall soon
see... 8-)

REAL STE
---------
S: -5,5
F: -6,0
E: -9,6
D: -13,9
C: -16,8
B: -20,3
A: -23,1
9: -26,3
8: -29,1
7: -32,0
6: -34,6
5: -37,6
4: -39,4
3: -43,4
2: -44,9
1: -48,7

HATARI 1.2
-----------
S: -2,5
F: -6,0
E: -9,0
D: -12,0
C: -15,0
B: -18,0
A: -21,0
9: -24,1
8: -27,1
7: -30,1
6: -33,1
5: -36,1
4: -39,1
3: -42,1
2: -45,1
1: -49,7

HATARI 1.3
-----------
S: -14,6
F: -6,0
E: -9,0
D: -12,0
C: -15,0
B: -18,0
A: -21,0
9: -24,0
8: -27,0
7: -30,1
6: -33,1
5: -36,1
4: -39,1
3: -42,1
2: -45,1
1: -49,7

STEEM 3.2 (just as an interesting comparison)
----------
S: -5,5
F: -6,0
E: -9,0
D: -11,8
C: -15,0
B: -18,0
A: -21,0
9: -24,0
8: -26,0
7: -27,9
6: -29,7
5: -32,0
4: -40,0
3: -52,1
2: -60,2
1: -Inf. (that's right, no signal. ZERO!)

Conclusion: PLEASE don't release Hatari v1.3 with this HUGE error in
mixing,
it's more than twice as bad (6dB = half the amplitude) as the error in
v1.2.

All this obviously with the reservation that I've done something wrong
when
scaling the values (blame Sound Forge), I seriously doubt I would have
done
something wrong when creating the sample, sampling and editing the whole
thing.
All data files are available on request, of course.


Per.




More information about the hatari-devel mailing list