Noplanet
Offline
USA

The speech synthesis mode was removed from opl2 compatibility when making opl3. I never knew of this.
So I was curious if anyone has played with opl2's CSM mode or seen any of it used in a game or program or anything.
I'm trying to make 4op speech macros, really hard to find actual documentation on anything of the sort. Just finding
crazy calculus equations from stanford university and learning how CSM works on other machines. I think it could feasibly be faked in Adlib Tracker 2 and made into an instrument bank. If anyone else feels like contributing the idea's basically to try and model sounds as instruments
to match this scheme like the SP0256-AL2:

1 /OY/ BOY
2 /AY/ Sky
3 /EH/ End
4 /KK3/ Comb
5 /PP/ Pow
6 /JH/ Dodge
7 /NN1/ Thin
8 /IH/ Sit
9 /TT2/ To
10 /RR1/ Rural
11 /AX/ Succeed
12 /MM/ Milk
13 /TT1/ Part
14 /DH1/ They
15 /IY/ See
16 /EY/ Beige
17 /DD1/ Could
18 /UW1/ To
19 /AO/ Aught
20 /AA/ Hot
21 /YY2/ Yes
22 /AE/ Hat
23 /HH1/ He
24 /BB1/ Business
25 /TH/ Thin
26 /UH/ Book
27 /UW2/ Food
28 /AW/ Out
29 /DD2/ Do
30 /GG3/ Wig
31 /VV/ Vest
32 /GG1/ Got
33 /SH/ Ship
34 /ZH/ Azure
35 /RR2/ Brain
36 /FF/ Food
37 /KK2/ Sky
38 /KK1/ Can't
39 /ZZ/ Zoo
40 /NG/ Anchor
41 /LL/ Lake
42 /WW/ Wool
43 /XR/ Repair
44 /WH/ Whig
45 /YY1/ Yes
46 /CH/ Church
47 /ER1/ Fir
48 /ER2/ Fir
49 /OW/ Beau
50 /DH2/ They
51 /SS/ Vest
52 /NN2/ No
53 /HH2/ Hoe
54 /OR/ Store
55 /AR/ Alarm
56 /YR/ Clear
57 /GG2/ Guest
58 /EL/ Saddle
59 /BB2/ Business

Warez Waldo
Offline
Planet Zaxxon

OK WHAT. The world of FM just keeps getting better and better!

somewhat_damaged
Offline
MBR

Yep, the SoundBlaster never ceases to amaze me...

bsa
Offline
Ostraya mate!

I was actually thinking about this the other day! Let us know how your experiments go.

This paper seems quite relevant, but I don't have access to the whole thing. The abstract actually gives a pretty good overview of the technique though:
http://www.sciencedirect.com/science/ar … 8485901112


There's also this, which is for vowel sounds only:
https://ccrma.stanford.edu/~rjc/pubs/au … n_via.html

boomlinde
Offline
Sweden

AFAIK the speech synthesis mode is just additive synthesis, but I might be wrong. To model speech, then, you would probably figure out the most significant amplitudes of the frequency components of speech and feed that data into amplitude controi register/s. Here's to hoping it's somewhat more advanced than that, though! I doubt it's anything like Chowning's FM speech.

Aly James
Offline
FRANCE

Might help smile  >> http://chipmusic.org/forums/topic/11028 … sm-sounds/

Noplanet
Offline
USA

Yea, too bad no one ever used the CSM in opl2.
surprisingly making consonant sounds is the hardest part.

this helps a lot for the rest:

Average vowel formants[7]
Vowel (IPA)    Formant f1    Formant f2
i                           240 Hz      2400 Hz
y                           235 Hz    2100 Hz
e                           390 Hz    2300 Hz
ø                           370 Hz    1900 Hz
ɛ                              610 Hz    1900 Hz
œ                           585 Hz    1710 Hz
a                           850 Hz    1610 Hz
æ                           820 Hz    1530 Hz
ɑ                           750 Hz    940 Hz
ɒ                           700 Hz    760 Hz
ʌ                           600 Hz    1170 Hz
ɔ                           500 Hz    700 Hz
ɤ                           460 Hz    1310 Hz
o                           360 Hz    640 Hz
ɯ                           300 Hz    1390 Hz
u                           250 Hz    595 Hz

sandneil
Offline

is this what DR SBAITSO used? https://www.youtube.com/watch?v=sV3pYZZ2jEw

Noplanet
Offline
USA

That's what I thought when I was a kid, turns out it was digital sample based. Using the opl3 sounds way different. Much noisier with the macro automation. The low bit controls for things like feedback and the multiplier seem to be the most limiting factors. we'll see what comes out though. made a few so far, but it's gonna take months to perfect. Give it a try! It's pretty fun finding the weird things you can make. I'll post an advanced macro editing tutorial on youtube soon.

Noplanet
Offline
USA

PWM type effects seems to be helping smile

jonah
Offline
Noplanet wrote:

PWM type effects seems to be helping smile

haha, yeah you probably stumbled onto VOSIM. smile

i think most of us that mess with synthesis come across it, but that dude got to name it. smile

it also seems like how the SID chip did speech synthesis too. I think it was 4bit? so 16 levels? i dunno - i think it's just modulating a "click" very fast.

it's more fun to just play around, but reading about it can point in interesting directions.

this is a pretty nice VOSIM explanation, but the original paper is available online too.
http://www.clavia.com/nordmodular/Modul … VOSIM.html

Noplanet
Offline
USA

yessssss, I got about a quarter way through this list, but had to take a break. it's mind numbing trying to make a distinguishment between mmm and nnn and oooo and aaaaa for hours on end, and then piecing them together into strings. Pretty fun though. Yea, I have SAM for my SID. I just like that the OPL3 sounds a bit more like the speakjets, SID sounds like a really crunched version of the doppler radar voice, though I think sam could be expanded to an 8580 specific mode and really make use of that analog filter.

boomlinde
Offline
Sweden

You seem like you could be interested in this! Seems like resynthesis recordings though.

Did it go well with the consonants so far? It's hard to produce colored noise with the OPL I think. Maybe you can cheat by flipping some values quickly with the CPU to add some additional rumble for browner noise for "k" etc.

Noplanet
Offline
USA

yea, sounds like a sampled vocoder, offence has a pretty good one in here. https://www.youtube.com/watch?v=Cd36A_8jYj4
Yes, regmacroediting quickly is the way. If 4op was easier to make single instruments that'd be the way to go since it worked pretty well, but keeping it simple and small.

Last edited by Noplanet (Jan 14, 2014 3:42 am)