Printable View
I'm not denying the Genesis samples are scartchy. I'm simply saying the SNES ones don't sound FINE to me as they are butchered to hell. Honestly I'd rather take the full Genesis samples for the most part. And after hearing the Arcade clip koolkitty linked earlier the scratchiness in the Genesis voices really doesn't seem to be that much worse by comparison.
Your opinion is your opinion. We are repeating ourselves at this point and nitpicking the nitpicks. If you prefer sf2 md voices in any way, okay.
Sega-16 forums should be renamed to this, if only we could make it shorter.Quote:
Originally Posted by Vector
WARO-16 and NTN-16 would work.Quote:
Originally Posted by sheath
Off the cuff, both of those sound like they should be Nintendo forums for Wario and Nintendo fans or something. Odd that.
They both sound bad . . . and I'd argue "distorted" would apply to both as well. Which bothers you more is really up to personal preference or perception (aesthetics).Quote:
Originally Posted by Vector
Which is, again, what makes the comparison different for Super Street Fighter 2, you get rid of that added variable . . . actually 2 added variables (MD samples are prefiltered and muffled, no longer bright/tinny). All that's left is the crappy disorted playback on the MD to detract from it, and that's a software/programming problem. (sound quality is dependent on programming since sample playback is done in software by the Z80 writing to an 8-bit DAC port on the YM2612 . . . that's one more variable on top of all the other factors affecting sample quality -bitrate, format, preprocessing, etc)
As I touched on above, and Kamahl mentioned, the YM2612 has an 8-bit DAC port that can be manually fed by an external CPU (the 6th FM channel gets disabled) . . . there's no hardware support for PCM, so it's all up to programming to load samples and write to the DAC. (additionally, there's also no timer interrupts to handle PCM sample rate taking in hardware for the Z80 -YM interrupt lines aren't connected, though you can poll the YM2612 timers at the expense of CPU overhead or use cycle timed code to time things instead -the latter is the most efficient if done well, and much better than interrupt PCM too, but it has to be done well to actually work . . . and poor timing is present in a lot of MD games due to a lack of good coding) That's not getting into other problems like bank-switching overhead and DMA contention for the Z80 requiring further workarounds to really make things work well. (the set up allows a lot of flexibility if programmed well though, and it was probably the cheapest option for Sega given they wanted SMS compatibility -a few other tweaks like connecting the YM interrupt lines, changing DMA priority, or a couple other things would have helped a lot though . . . not to mention potential hardware workarounds via on-cart hardware -DMA sound chip for hardware PCM playback, memory mappers to work around the bank-switching issue, etc)Quote:
Okay, more can be done in fm on md because it doesn't have to worry about storing a sample like snes for audio all the time sort of like the md is a synthesizer and it can play samples if it can sample but snes only uses samples yes ? But where I'm lost is, md is FM/synth. So how does it play samples, from the ti chip or sms z80 chip ? How does yamaha chip playback both, and the md sample storage is less than snes yes ?
Think if ot kind of like those DIY or Covox style parallel port DAC "sound devices" for PCs. (Disney Sound Source is the same kind of thing)
Games using multiple PCM channels do this by mixing them in software, and this is when bank-switching becomes a problem, at least assuming the game uses more than 32 kB of samples total. (the Z80 switches 32 kB banks in ROM space, and the banking mechanism is slow -a serial shift that takes 11 cycles iirc- and multi-channel games that use more than 32kB would thus be switching banks after every single sample played -every sample byte, that is- adding a ton of overhead . . . that-is unless you mix to a buffer in Z80 RAM and thus read from 1 bank for a while and then mix in the other sample to the buffer -more elaborate buffering also avoids the DMA contention issues, but it seems that wasn't typically implemented)
Oh, and one more small note, you CAN also play samples through the PSG channels as 4-bit nonlinear PCM of sorts (modulating the 4-bit logarithmic volume of the PSG channels) or allowing somewhat better resolution by adding the 3 channels (MOD players on the ST do that). Only 1 game does this AFIK, and uses plain 4-bit PSG formatted samples at around 3 kHz, and this is After Burner II, which actually does 2 sample channels simultaneously with 1 dedicated to music (percussion and orchestral hits) and the other to speech and sfx (guns/missiles/explosions). Actually the speech doesn't sound that terrible given the crappy 4-bit PSG volume formatting and low sample rate (should be less than 1/2 the bitrate of SFII's samples). Actually, I think they might sound about as good (or bad) as SFII SCE's overall, or pretty close. (that says a lot, both for the decent optimization of ABII and the crappy quality of SFII :p -it also sounds better than PC Engine AB II's speech, but I assume that's super low sample rate 4-bit PCM)
After Burner II is interesting for that, and it's also the only MD game to use 6 FM channels along with samples because of that. ;) (then again, plenty of games make good use of PSG channels for music and/or SFX too)
8-bit uncompressed PCM can sound better than the 4-bit (actually 4.5 bit) ADPCM format the SNES uses, and you can do higher than 16 kHz. That said, it's a matter of ROM space to actually allot that to . . . but reserving a chunk of space for a limited selection of high quality samples is what several games opted for. (several of the best sounding games were on the small size ROM-wise -Mega Turrican, Vapor Trail, and Atomic Runner are all only 1 MB . . . SFII and SSFII were both among the largest carts of their times at 3 and 5 MB ironically enough :p -the latter is the largest on the system prior to Pier Solar)Quote:
How when md is 8 bit and snes is 12 bit ? I thought yamaha was fm. Can you explain how it can play samples and is the fm/synth 8 bit output and sample pcm 8 bit or 4 bit to save space like what stef does. Or simply explain the md audio process to me from info out of cart to yamaha chip and what it does to the max quality output of md either just for fm or fm and samples.
The best the MD can do is 26 kHz 8-bit PCM, and that's limited by the speed of the YM2612's DAC port, not the Z80. (TmEE had a demo pushing over 44 kHz using the Z80, but he established that you start missing writes around 26 kHz -Fusion actually caps it at ~19.6 kHz and Gens at ~16 kHz iirc)
-Technically, you can force the SNES's SPC module to manually play 16-bit uncompressed PCM too (and actually do 32 channels for that matter), but that was never done until modern homebrew, and even then it's very limited due to the 64 kB sample RAM constraint. (you might be able to decode other PCM formats using the SPC700 CPU too, but that's not been done yet AFIK)
That said, at any given bitrate or fixed allotment of ROM space, you get a lot more trade-offs. (shorter/fewer high bitrate samples vs longer/more low bitrate ones) But you also have a lot more flexibility in general since there's no issue of loading samples into sound RAM. (just fetch them from ROM) You can also use any PCM format you want (that the Z80 can handle) and decode to 8-bit PCM. (there's demos using 1, 2, and 4-bit CVSD, ADPCM, and DPCM derivatives as well as plain 4-bit PCM, and both 4-bit DPCM and 4-bit PCM were fairly common alongside 8-bit PCM for the original commercial MD games in the 80s/90s) So it definitely gives you more flexibility over the SNES. (and that sound RAM isn't just small, but very painful/slow to update on the fly -unlike the PCM chip in the Sega CD, by comparison, which has pretty decent DMA loading for wave-RAM updates -which is also why streaming PCM for FMV and such isn't that big of a problem, including the 32 kHz 8-bit PCM streams used in Sonic CD and Pier Solar -the CD soundtrack is entirely 32 kHz 8-bit stereo, not CD-DA, to fit more on the single CD)
Stef uses 4-bit DPCM to save ROM space, Tiido uses 22 kHz PCM for maximum quality. (the full sample set in his sound engine used 512 kB by default) For modern homebrew, ROM space constraints aren't THAT much of a problem, but they're still a consideration, especially working within the 4 MB limit for normal non-bankswitched games. (the SSFII mapper supports up to 32 MB though . . . and if you ignore the address space reserved for expansion -used for MCD and 32x- you have 10 MB of space without banking)
MD FM samples at ~53 kHz, PCM is anything up to 26 kHz (old games rarely use more than 16 kHz -more often 4 to 11 kHz), and SNES is anywhere up to 16 kHz iirc. (fixed ADPCM type sample format on SNES, lots on the MD). As for what actual sampled instruments use on the SNES on average, I'm not sure . . . but with 64 kB dedicated to just samples for music, that's only about 7.1 seconds worth of samples at 16 kHz. (with limited use of streaming updates -which relatively few games actually did) Plus, you have to use that for all instruments, vs MD which used PCM only for percussion in music typically. (if at all)Quote:
Okay. What is the typical snes game audio sample hz and bitrate to the typical md audio fm/sample hz and bitrate ?
Tons of variables there, and this is totally off from my actual point about the advantages of sample playback on the MD in general. (I didn't mean to say there weren't disadvantages either -it'd be tough for even a good programmer to do a decent sounding multi-channel MOD player on the Z80, for example)
I'd compare it to the Amiga or PC . . . but floppy disk software+RAM vs ROM carts kind of skews that in general. (and that's a broad topic in itself . . . though looking at OCS Amiga games with limited numbers of disks is more straightforward)
Which version of Kega are you using?Quote:
Just so it's known, I'm listening to all this off hardware, not emulators. I've noticed the frequencies in Kega for instance sound off sometimes. Virtua Racing D sounds better on my hardware than my Kega. I can just tell some games have different keys or something, plus the orange bridge is lighter but more dark on hardware, it's not totally perfect but that is nitpicking as Kega is a great overall emulator.
I know there's still some sound differences, but the most obvious simply come from filtering differences . . . or generally problematic amp circuits on some MD models, and/or aging capacitors. (introducing distortion, muffling, hiss, etc)
TmEE prefers to use his modded model 2 with excellent amp circuitry (the so-called Crystal Clear Audio Mod or CCAM) without the heavy low-pass filtering of stock consoles. This makes some sounds overly bright and harsh, but others exceptionally clear and clean. (all of his music is made with that in mind iirc) It sounds a bit like fusion with filtering disabled IMO.
I'm sure there's more subtle differences in sound between fusion and real hardware (and I know some of the PCM related problems aren't reproduced -namely DMA contention artifacts), but the obvious differences come from real hardware that has problems with it. :p
Thanks, I was using Kega 3.51 for years. I'm understanding all this better now, I see PCM mentioned (analog to digital) but why would the md do digital to analog as Kam mentioned ? Output ? So the SNES has 7 stored seconds let's say to play with sampled sounds basically, but md only less than a second (if samples are used, if not, then it doesn't really need to store anythng because chip plays fm yes ? And if I drag a 44.100 wav on Mega Everdrive, when it plays through Genesis it's 4 to 8 bits yes ? Amazed MD can do that btw. Edit. Also didn't know Sega CD games used 32 hz, thought it would be 44.1 because game + 32 hz only take up on avg 300 to 400 mb on some games. On 200 mb games, they could of used 44.1 cdda.
Fusion doesn't handle the 32X PWM samples correctly. It's written to handle "adequately" the kinds of samples played in 32X games. As such, it doesn't match real hardware, and doesn't usually sound right with my 32X homebrew. The only 32X emulator to handle the PWM correctly is Gens/GS r7 with the patches I posted in the RetroSonic Gens/GS r7 thread.
SegaCD CDDA tracks are standard Redbook - 16-bit stereo at 44.1kHz. The PCM chip has a base sample rate of 32552 Hz for playing PCM samples. It uses 8-bit sign/magnitude format data, with separate 8-bit volume control, and 4-bit panning control (16 steps for left, and 16 steps for right). As such the PCM chip is pretty good at playing MODs, as my demo in the Blast Processing forum shows. The only "complaint" I have is that 64KB of sound ram isn't a whole lot when dealing with uncompressed samples.
Cool info.
There is no analog to digital done in the MD at all, there is quite a lot of digital to analog though . . . all audio and video signals are output as analog signals, but everything is digital internally, so you've got sound and video DACs to convert the digital audio and pixel data to analog sound and video signals. (and there's 2 audio DACs, one for the SN76489 PSG and one for the YM2612)Quote:
Originally Posted by Vector
Now, this usually doesn't matter at all to the programmer, since everything is digital on there end, and all DAC handling is done by dedicated hardware that's not going to be programmable. What matters in this context isn't the DAC itself, but the fact there's mode in the YM2612 that provides a direct parallel 8-bit port to write to the DAC for digital output. (you need an I/O port of some sort connecting to a DAC to make it useful for a CPU to feed data to -a lot of old arcade boards did all sound with a CPU or MCU driving a DAC in that manner, and that's also the way the Tandy CoCo's sound works -and a DAC plugged into a computer parallel port)
PCM (pulse code modulation) itself is just the common uncompressed digital audio format that's convenient to store and play back in software in this case (and for numerous other old consoles and computers -be it DAC ports or volume modulation of PSG channels, the latter being pretty similar to the former if the volume is linear).
With the MD, you've got the YM2612's DAC port using 8-bit unsigned linear amplitude, so samples must be formatted as 8-bit unsigned PCM. (and any other format must be decoded to 8-bit unsigned PCM)
To actually play PCM as such, you need to write a program for the Z80 to feed 8-bit unsigned samples to the YM2612 8-bit DAC port, and that's where cycle timed code or polling interval timers (ie timer A or B in the YM2612) comes into play . . . or for similar cases on other platforms, interrupts signaled by interval timers are also common choices, but that's not an option here. (hence why it's commonly cited that it would have helped a lot for Sega to have connected the Z80 interrupt line to the YM2612 . . . and even for coders using efficient cycle timed PCM playback code, timer interrupts are very useful for timing music players and such -- as it is, you can use the vblank interrupt for 60 or 50 bpm timing in music -60 NTSC, 50 PAL- and a lot of games do it that way; the other option is to poll the YM interval timers)
Wait a minute. I only mentioned the 7 seconds in the specific context of maximum sound quality possible, not typical usage on the SNES. (done in the context of what's possible on the MD) So don't read any further into it than that. (it was a statement about capabilities, not typical SNES game use)Quote:
So the SNES has 7 stored seconds let's say to play with sampled sounds basically, but md only less than a second (if samples are used, if not, then it doesn't really need to store anythng because chip plays fm yes ?
I'm not sure where the "1 second" comes into play on the MD end of things either. The SNES comment was made in the context of the max quality samples being used (72 kbps) and the 64 kB (512 kbit) of sound RAM available (512/72= 7.11s).
On the MD, there's no restriction for sound memory size other than the ROM size limit itself. You don't load samples into RAM before playing them, you stream straight from ROM. (well, that, and the 32 kB bank size constraint -which is only problematic for certain situations) The Z80 does have 8kB of work RAM, but that's not used to store samples, but critical CPU data and code (and for a PCM mixing buffer for better multi-channel sample playback drivers).
How much ROM space you want to allot to samples is up to the developer, but there are demos using basically the entire ROM space for sound, hence how THIS is done: https://www.youtube.com/watch?v=oKsUAywSyEg
Streaming samples to sound RAM is possible on the SNES too, but it's pretty restricted and also painful to program for from what I understand. (Nintendo's docs didn't help all that much with the limitations either)
Oh, also synth stuff (FM and PSG) still takes space too, you need memory to store the instrument patches and sound code, music data (sequences), etc. It just uses a lot less than sampled stuff would use . . . or at least remotely decent quality sampled stuff. (or more space to dedicate to what samples are used)
You wouldn't be doing 4-bit anything in that case, unless you loaded a 4-bit wav file in the first place. ;)Quote:
And if I drag a 44.100 wav on Mega Everdrive, when it plays through Genesis it's 4 to 8 bits yes ? Amazed MD can do that btw.
If you used a standard 44.1 kHz 16-bit (mono) PCM wav file on the MD, the player would have to clip that to 8-bits and output it as such . . . no reason to clip it to less than 8-bits though. An 8-bit PCM wav file could be played straight, or require some decoding due to needing 8-bit unsigned samples. (if the wav file was stored as signed PCM, it wouldn't work without additional decoding -or you'd get a garbled mess if you fed the YM DAC like that ;))
It also isn't going to handle 44.1 kHz properly since the DAC will get overloaded as I mentioned. The DAC port the Z80 writes to is limited to ~26 kHz before you start dropping samples (missed writes), so 44.1 kHz samples are going to sound off because of that, but anything up to 26 kHz is fine so long as the player code itself can handle it. (ie 22.05 kHz 8-bit mono wav files will sound perfect, and should sound very similar to they way a PC would play them -barring any differences in filtering, or interpolation and such if the PC's audio driver has that done automatically)
Oh forgot to post this with the Matt Furniss (Krysalis) music/percussion examples listed before:
https://www.youtube.com/watch?v=4sXgFmZ-tM0
(well, that and the polygons :p )
Or, more commonly for early MD games, 4 kHz 8-bit samples were used . . . and honestly, that's going to be worse for a lot of stuff than 8 kHz 4-bit.Quote:
Originally Posted by Kamahl
That depends on preprocessing too though. You could low-pass filter the hell out of a sample to reduce those sort of artifacts to fit better with the limited bandwidth of something like 4 or 8 kHz PCM and/or filter the hell out of it after the fact too (like the Amiga can toggle in software). That's the main reason you don't have the same graininess in Super Street Fighter II on the MD as SCE. (the garbled playback is still a problem, but actual samples tend to be muffled but not scratchy as such)Quote:
Uneven playback means the Z80 code sucks and delays writes to the DAC, so there's kind of a "wavy" effect done on the sound, making it garbly (barely noticeable at 16KHz, very noticeable at 8KHz)
Of course, whether super heavy filtering is aesthetically preferable depends both on the listener and the sound in question. (it's better to have more artifacting if it allows the sound to be better recreated in general -including music samples that would be better off "bright" and scratchy than overly muffled . . . though there's some middleground too)
For the same reason, relatively low-pitch sounds in general will sound better at low sample rates than high pitched sounds. So deep male voices will sound better than high pitched female voices, or screeches, screams, etc. (hence why the speech in Space Harrier II and Altered Beast sounds pretty decent in spite of being 4 kHz 8-bit PCM)
If he hasn't, he should, to understand the context of the PC Engine's sound at the very least. (since it's scratchy compared to the SNES too, but cleaner and clearer than any of the others in general -the artifacts that are there are probably partially from the ADPCM arcade sources too, since it's not typical PCM artifacts)Quote:
Originally Posted by TrekkiesUnite118
Some of the artifacting comes from the arcade sources (or at least it really sounds like it -given the Turbo beta on MD, SCE, and CE on PC Engine), but then there's the additional garbling in SCE . . . and general added graininess (tinny) of being downsampled to 4 kHz.Quote:
Originally Posted by TrekkiesUnite118
If you've got a MD with an overclock (and halt switch), try testing the speech and halting the 68k while the speech sample plays. What you here there should be pretty much the raw PCM quality without playback errors. (Capcom's sound driver is a typical 68k driver -68k for sequencing and Z80 for PCM handling, but it manages to screw up Z80 timing with 68k accesses it seems . . . must be really excessive time with the 68k on the Z80 bus -Z80 is halted while 68k accesses it- and I don't think that's a problem with SMPS 68k or even pre-SMPS 68k -or the 68k drivers used for Vapor Trail/Atomic Runner or the Krysalis based games -or the 68k drivers used on the 32x for that matter) That and 68k drivers typically have FEWER problems with handling PCM . . . that's the whole point of slaving the Z80 for PCM handling. (Mega Turrican probably has the only non-homebrew MD sound driver to have great sounding sample playback while using a fully Z80 based sound engine . . . well, actually, it seems like the 68k handled PSG, but Z80 does PCM and FM -if you halt the Z80, PSG continued normally . . . seems to be doing software PSG envelopes too)
Fusion also seems to upsample and interpolate PWM done at particularly low rates, at least given what happens with Tempo's intro music. (several others have low sample rate stuff that sounds pretty close in fusion -like Metal Head's horrible quality speech- so I assume that's scaled up already -which would avoid PWM squeal issues and would allow other sounds at higher sample rates too)Quote:
Originally Posted by Chilly Willy
https://www.youtube.com/watch?v=KFg8f5YdX2o (I'm assuming that's Fusion -sounds very similar in any case, and not like real hardware, as in Joe's video)
Yes sheath, I heard the audios on page days ago and stated tg16 is best of 3. (Said that many times).
Kitty, the reason I brought it up is because I keep seeing PCM typed, but "Pulse-code modulation (PCM) is a method used to digitally represent, sampled analog signals. It is the standard form of digital audio in computers" but your saying "There is no analog to digital done in the MD at all". So I'm confused.
Good read otherwise. "22.05 kHz 8-bit mono wav files will sound perfect," Okay I will convert a 44.1 wav to that and put on everdrive.
Yeah sequencer data is small , like 4 to 64 kbs compared to samples which might range from kb to mbs (at least on music samplers).
You're confusing things. PCM is used to represent analog signals in digital form. That is, you sample (store at regular intervals) a series of values (digital), which will be used later to regenerate the sampled waveform (analog).
The MD can't record PCM at all, it just plays it back, it does the "digital to analog conversion" (DAC).
Okay.
One other thing that just occurred to me:
The YM2612 DAC misses writes beyond ~26 kHz in the MD, and the YM2612 is clocked at 7.67 MHz (same as the CPU -so 7.6 MHz in PAL), and the chip's sampling frequency is 1/144th the external clock rate, so 53.27 kHz (same for OPL on PC sound cards -hence 49.7 kHz for 7.16 MHz chips). The 26 kHz DAC port speed is thus very close to 1/2 the YM sample rate (26.63 kHz), so I wonder if that's more than just coincidence.