Hi All,

luis jure wrote:

> >The output from this contains a single sample -32768 spike about a third of
> >the way in for no apparent reason.
> 
> steve, your orchestra produces a perfectly clean sine sweep for me too
> (winsound 3.53; gabriel maldonado's directcsound 2.70, based on csound 3.52
> i think). so check your version of csound.
> 
> anyway, i don't think it's the best to put:
> 
> ilevl = ampdb(96 + p4)
> 
> according to the manual, ampdb returns a raw value of 32000 for 90dB, thus
> if you take 0dB as your maximum output, a value of 90 would give almost 0dB
> for 16 bits (-0.3 in fact) and not -6 dB as you seem to want.
> 
> by the way, looking at the output, ampdb doesn't seem to be extremely
> precise. or is it?

Hmm! I was under the impression that (perfect) 16 bit audio had a 96 dB
dynamic range, hence the above. Perhaps ampdb is calibrated wrong, or am I
missing something?

Anyway, my glitching problem remains, so here is a simplified version of my
orc and sco which shows the problem more clearly. When I run this I get a half
wave rectified sine, i.e. no negative output at all. Halving the table size
produces a sine sweep with a 1 sample wide -32768 spike on a zero crossing
point about one quarter of the way in, reducing the table size more cures the
problem. Replacing oscil3 with oscili also cures the problem. 

;Orc
sr     = 44100
kr     = 4410
ksmps  = 10
nchnls = 1

instr    1 ; Sine sweep

ilevl    = p4*32768 ; Level
ifreq1   = p5       ; Sweep start
ifreq2   = p6       ; Sweep end
asweep   expseg  ifreq1, p3, ifreq2
aosc     oscil3  ilevl, asweep, 1
out      aosc
endin

;Sco
f1 0 65536 10 1 ; Sine

;   Strt Leng Levl Freq1 Freq2
i1  0.00 1.00 1.00    20 20000
e

Thanks,

Steve Cook.

>> > The output from this contains a single sample -32768 spike about a
>> > third of the way in for no apparent reason.
>> 
>> steve, your orchestra produces a perfectly clean sine sweep for me too
>> (winsound 3.53; gabriel maldonado's directcsound 2.70, based on csound
>> 3.52 i think). so check your version of csound.

> Anyway, my glitching problem remains, so here is a simplified version of
> my orc and sco which shows the problem more clearly. When I run this I get
> a half wave rectified sine, i.e. no negative output at all. Halving the
> table size produces a sine sweep with a 1 sample wide -32768 spike on a
> zero crossing point about one quarter of the way in, reducing the table
> size more cures the problem. Replacing oscil3 with oscili also cures the
> problem.

Well, 1st: You have an amplitude of 32768, and sin(90)*32768 = 32768,
ie one sample to high for a 16-bit output. Thats why you get a spike,
because 32768 = -32768 when using signed values (as in a sample).
Aparently, CSound will clip the value to 32767 when using word-sized output
(16bit) but not for 8-bit output, this might be why some people get spikes,
and some are not.
Instead of setting /ilevl/ to /p4*32768/, try with /p4*32767/. It might
(should) help.


2nd: Oscili uses *linear* interpolation, in wich a sample will not be
louder than one of the points used when interpolating.
Oscil3 uses some other kind of interpolation, where *(i guess)* if you have
a specific sequence of input values, the output *CAN* be higher than the
inputs. (im not shure about this though).
This should *NOT* be the reason in this case though, as you have a very
large
sine as input, and it's the best signal an interpolator could think of..


// Anders

ps. I used my own GCC-compiled 3.52 Amiga-version