Forums

Hi everyone,

Below is the code used to play the "Let's Go" sound sample at the start of each level in the Game Gear version of Lemmings 2. I've attached the output table (1024 bytes) and the sound sample as a RAR file. The code was written for KASM (written by Krysalis Software, in 1992!) so you may need to edit it to work in the assembler of your choice.

SMPRTE DB 0

INSREG: LD C,#80    ;Initialise sound registers
LD B,3
ISR10: LD A,C
OUT (PSG),A
XOR A
OUT (PSG),A
LD A,C
ADD #20
LD C,A
DJNZ ISR10
RET

PSGSAMP: DI
CALL INSREG
LD HL,LETSGO ;Sample data
LD BC,6346    ;Length
LD A,16    ;Rate
LD (SMPRTE),A
PSGS10: LD A,(HL)
SUB 127
PUSH HL
LD L,A
LD H,0
ADD HL,HL
ADD HL,HL
LD A,(SMPRTE)
LD E,A
PSGS20: DEC E
JR NZ,PSGS20
LD A,L
ADD 255&SAMPTB
LD E,A
LD A,H
ADC SAMPTB/256
LD D,A
PUSH BC
LD A,(DE)
LD B,A
INC DE
LD A,(DE)
LD C,A
INC DE
LD A,(DE)
LD D,A
LD E,C
LD C,PSG
OUT (C),B
OUT (C),E
OUT (C),D
POP BC
POP HL
DEC BC
INC HL
LD A,B
OR C
JP NZ,PSGS10
EI
RET

SAMPTB: INCBIN L2SGSMTB.BIN ; 1024
LETSGO: INCBIN L2LETSGO.SMP ; 6346

Code from twenty years ago, so no guarantees, your mileage may vary, etc. :-)

L2 Sample Data.rar (4.66 KB)

L2 Sample Data

Interesting, it seems somewhat different from how many games do it, possibly for cross platform reasons. The sample data seems to be raw signed 8 bit data, and the lookup table is trying to get some extra resolution out of the 16 volume levels of the PSG by mixing the three channels. This comes at the cost of ROM space, but by the early 90s this was much less of an issue than in the early years of the Master System.

That's an interesting approach and sounds quite decent :)

I experimented with sample playback on the PSG a while ago and went to a different route :
I took the logarithmic curve of the PSG volume register into account and with some DC offset shifting I could create 4bit streams that sound pretty naiss.
http://www.tmeeco.eu/BitShit/SMSPCM.SMS

The lookup table was reverse engineered by hooking up a voltmeter to the headphone output and measuring the DC level for each combination of channel outputs. While this gave the highest resolution, my lack of sample processing software meant that I couldn't take advantage of it.

Cutting back this back to two channels (with two channels combined at the same level) would have yielded results that saved ROM space and might have been good enough for an 8-bit sample - but that required more time that I had.

Do you remember if there were any non-linearities like say highest peaks being flattened ?

From my tests that happened in SMS2 and only because of "wrong" choice of a resistor on the audio out pin. MD1/2 and SMS1 has no such issues, GG I have not touched.

The lookup table seems a bit odd when I graph it. I guess the voltmeter approach may have taken into account real-system effects which you don't get with pure emulation. It is broadly dealing with the non-linearity, but has resolution issues at the high end.

lookup.png (19.7 KB)

Interesting to see it graphed. :-) Certainly the sample playback sounded much better through the speaker and headphones than in an emulator such as Fusion, but it seems at least comparable with the playback on something like Ecco the Dolphin.

Maxim wrote

The lookup table seems a bit odd when I graph it. I guess the voltmeter approach may have taken into account real-system effects which you don't get with pure emulation. It is broadly dealing with the non-linearity, but has resolution issues at the high end.

Author	Message
MattTaylor Joined: 13 Apr 2014 Posts: 25 Location: London, UK	Playing Sound Samples Posted: Mon Apr 14, 2014 12:52 am
	Hi everyone, Below is the code used to play the "Let's Go" sound sample at the start of each level in the Game Gear version of Lemmings 2. I've attached the output table (1024 bytes) and the sound sample as a RAR file. The code was written for KASM (written by Krysalis Software, in 1992!) so you may need to edit it to work in the assembler of your choice. SMPRTE DB 0 INSREG: LD C,#80 ;Initialise sound registers LD B,3 ISR10: LD A,C OUT (PSG),A XOR A OUT (PSG),A LD A,C ADD #20 LD C,A DJNZ ISR10 RET PSGSAMP: DI CALL INSREG LD HL,LETSGO ;Sample data LD BC,6346 ;Length LD A,16 ;Rate LD (SMPRTE),A PSGS10: LD A,(HL) SUB 127 PUSH HL LD L,A LD H,0 ADD HL,HL ADD HL,HL LD A,(SMPRTE) LD E,A PSGS20: DEC E JR NZ,PSGS20 LD A,L ADD 255&SAMPTB LD E,A LD A,H ADC SAMPTB/256 LD D,A PUSH BC LD A,(DE) LD B,A INC DE LD A,(DE) LD C,A INC DE LD A,(DE) LD D,A LD E,C LD C,PSG OUT (C),B OUT (C),E OUT (C),D POP BC POP HL DEC BC INC HL LD A,B OR C JP NZ,PSGS10 EI RET SAMPTB: INCBIN L2SGSMTB.BIN ; 1024 LETSGO: INCBIN L2LETSGO.SMP ; 6346 Code from twenty years ago, so no guarantees, your mileage may vary, etc. :-) L2 Sample Data.rar (4.66 KB) L2 Sample Data

Maxim Site Admin Joined: 19 Oct 1999 Posts: 14744 Location: London	Posted: Mon Apr 14, 2014 1:12 am
	Interesting, it seems somewhat different from how many games do it, possibly for cross platform reasons. The sample data seems to be raw signed 8 bit data, and the lookup table is trying to get some extra resolution out of the 16 volume levels of the PSG by mixing the three channels. This comes at the cost of ROM space, but by the early 90s this was much less of an issue than in the early years of the Master System.

TmEE Joined: 31 Oct 2007 Posts: 853 Location: Estonia, Rapla city	Posted: Mon Apr 14, 2014 8:13 am
	That's an interesting approach and sounds quite decent :) I experimented with sample playback on the PSG a while ago and went to a different route : I took the logarithmic curve of the PSG volume register into account and with some DC offset shifting I could create 4bit streams that sound pretty naiss. http://www.tmeeco.eu/BitShit/SMSPCM.SMS

MattTaylor Joined: 13 Apr 2014 Posts: 25 Location: London, UK	Posted: Mon Apr 14, 2014 8:32 am
	The lookup table was reverse engineered by hooking up a voltmeter to the headphone output and measuring the DC level for each combination of channel outputs. While this gave the highest resolution, my lack of sample processing software meant that I couldn't take advantage of it. Cutting back this back to two channels (with two channels combined at the same level) would have yielded results that saved ROM space and might have been good enough for an 8-bit sample - but that required more time that I had.

TmEE Joined: 31 Oct 2007 Posts: 853 Location: Estonia, Rapla city	Posted: Mon Apr 14, 2014 8:52 am
	Do you remember if there were any non-linearities like say highest peaks being flattened ? From my tests that happened in SMS2 and only because of "wrong" choice of a resistor on the audio out pin. MD1/2 and SMS1 has no such issues, GG I have not touched.

Maxim Site Admin Joined: 19 Oct 1999 Posts: 14744 Location: London	Posted: Mon Apr 14, 2014 12:55 pm
	The lookup table seems a bit odd when I graph it. I guess the voltmeter approach may have taken into account real-system effects which you don't get with pure emulation. It is broadly dealing with the non-linearity, but has resolution issues at the high end. lookup.png (19.7 KB)

MattTaylor Joined: 13 Apr 2014 Posts: 25 Location: London, UK	Posted: Mon Apr 14, 2014 4:31 pm
	Interesting to see it graphed. :-) Certainly the sample playback sounded much better through the speaker and headphones than in an emulator such as Fusion, but it seems at least comparable with the playback on something like Ecco the Dolphin. Maxim wrote The lookup table seems a bit odd when I graph it. I guess the voltmeter approach may have taken into account real-system effects which you don't get with pure emulation. It is broadly dealing with the non-linearity, but has resolution issues at the high end.

Forums

View topic - Playing Sound Samples