Yep. As far as I remember they double buffered the Z80 RAM to get this quality. So double buffering is a technique that allows software both read and write to memory. While the Z80 (secondary CPU) was playing the samples in order, the M68000 (main CPU) was filling the other buffer. When the time comes, you swap the buffers, rinse and repeat.

The Zilog Z80 uses Amiga MOD file!.

kzfaq.info/get/bejne/pNl6Y86FnKvaiIU.html

@@TheZudah It's a Namco XGM MOD expansion!.

Freegin John and TT were programming ninjas! Gamehut videos never disappoint!

Do you happen to have or know where I can download the MOD files from the game itself to see if my Amiga 1200 will play them back? :)

VitinhoCarneiro Never knew the Amiga had the game?? the MOD worked though. Cool that all these years I knew that music had an Amiga sound to it on the Sega Genesis.

898doffo I don't think this game ever released on the Amiga.

Anyone know what would be the quality limit if you use more than one channel of sampled sound at once? I remember reading 22Khz somewhere.

Depends how good your modplayer code was, really. The MD's PCM channel appears to be hardware limited to roughly 26khz (about half the chip's overall effective output rate, seems the PCM can only load a new sample every other tick), which is about midrange vs most other hardware of the time - e.g. Mac at 22khz, Soundblaster at 22~24khz, Amiga at ~29khz, SNES at 32khz (and Atari STe at 50khz max but rather more comfortable at 25khz), and whilst not CD quality still sounds perfectly fine for something coming out of your TV so long as the samples are fairly clean. Replay-code wise, it's probably best to compare against other software-mixing modplayers of the time, especially on 7 to 8mhz class 68000-based machines. The Amiga itself is a tricky one because its main output was through very nearly selfcontained hardware - the CPU just had to update each channel in terms of sample start position in memory, length, playback rate and volume/panning, then didn't need to be bothered with it in any way, there wasn't even any RAM access contention because of interleaved access. However it did have some trick software like Octamed that did various things to add additional channels beyond the 4 hardware ones. I don't know the exact typical rates that could be achieved on the 7.1mhz models, but they can be looked up. However ISTR something like you could do 5 or 6 channels at full rate or very close to it, after which it started to drop off pretty hard, maybe even only 15-16khz with 8 channels? Certainly it seemed to be that generally only 7 would be used, 6 were lower frequency (three pairs, essentially, corresponding to hardware channels), and the 7th was pure hardware for high frequency sounds. Still, that suggests software mixing could be done at fairly high frequencies for 4-channel playback. The other less obvious issue but one that can be heard with the Toy Story modules (esp. the ending theme) is a loss of dynamic range / bit depth. Yer Amiga hardware actually has four clean 8-bit samples, which are prescaled through analogue attenuators, added together through analogue mixer / amps, and given a pass through an analogue low-pass filter to get rid of any nasty hissy quantisation/sampling noise... the overall result can be equivalent to having 14 bits or more overall. Mixing in software, often the result isn't anywhere near as clean as various corners get cut with sticking the samples together, most particularly mixing everything down to a single 8-bit value without any filtering and with fairly janky amplitude scaling (essentially doing fused multiply-adds in an age where the term hadn't yet been thought of, so making it happen at speed on a very much integer-focused chip required reducing the mathematical precision). Taking the hardware channels and other such complication out of the picture, I'd expect similar baseline performance from any other similarly powerful machine. For example the STe can replay 4-channel mods quite nicely at 25khz mixing rate, and on a good day manage 50khz with various tricks calling on other hardware like the Blitter to act as an ersatz DMA burst transfer device for data other than images. However as it only has *two* hardware PCM channels (with adjustable volume at least), even this requires a degree of software mixing, so the ultimate fidelity isn't quite as good, particularly at 25k (which is a fixed figure - it can actually only DMA at one of four different set rates, binary divisions of 50k, thus some extra distortion and system load comes in from either having to precalculate / timestretch any samples that aren't an exact multiple into the buffer, whereas the miggy has fairly fine-grained hardware sampling rates). And whilst it also has hardware post-filtering on each of those two channels, that doesn't operate on the individual samples at the point they're being summed. It is, however, maybe what we'd call second best, leaving PCs and later Macs out of the picture. Taking an aside for a moment, a slower PC with a Soundblaster Pro card or an older Mac which was capable of stereo but still at a fixed ~22.2khz rate would count as third-best on this scale, for the same reasons... their hardware is essentially of the same type but at progressively (slightly) lower rates, and with lesser post-filtering. The SB could at least be set to various arbitrary output rates, but not super-fine-grained and any paired samples would still need some degree of stretching if they weren't replaying at the same rate as each other (just like using Octamed, really). SNES is a tricky one because it has its own entirely independent sound unit, which basically runs as software by default, and whilst it has 8 independent channels that can each run at up to 32khz with fairly deep bit depth digital mixing, levelling/panning and filtering (and reverb!), the samples themselves are stuck as 4-bit ADPCM with a rather rough encoding scheme and the module only has 64kb memory for all of samples, sequences, and replay routines, so we hit the problem of a very good quality output stage, compromised by crappy input to that part. However the 8-channel / 32khz bit alone kinda disqualifies it from this discussion, as if all the machines had the same quality samples coming in, it would win easily with near-zero CPU use. Back to the main track, Megadrive I expect would be the next-best thing, but not far off the bottom of the list, saved only by its DAC channel being a proper 8-bit output into which samples could be latched without any further messing around, at an entirely arbitrary rate (upto a max of 26khz). So if you had ready-baked sounds that were perfectly good when played that way, and didn't want to mix them at all, the output would be top notch, as demonstrated here with the better quality percussion - only one sound plays at a time, at presumably maximum digital amplitude (so best possible resolution), and at exactly the same rate that it was sampled. No stretching, no low-precision multiplication, etc. Post filtering of course varied per model (bit of a sore point for some people). However that starts to fall away when you want to mix 2 or more sounds together, especially in completely variable ways instead of in any direct 2:1 or other integer relationship in terms of volume or speed, because we're right back to what had to happen for the STe/etc, and with Octamed. Except now there might be four different sounds trying to cram into the same channel, and even if you approach it in a sophisticated way using 16-bit ints or some kind of floating point, it still eventually has to be truncated to a single somewhat scratchy 8-bit value which hasn't had the benefit of any proper filtering or the like along the way. So the waveform gets a bit muddied up rather like pulling 8 channels out of the STe or 16 from the Amiga. And if the maths have been simplified to get a better output rate, that's even worse. Rate-wise, it's probably not so bad. You're performing 3 mult-adds (probably safe to assume that the loudest channel is taken as the thing that all the others add to, with their scaling normalised to the difference between their notional amplitudes, and it's kept digitally at 100% then attenuated down to its set value in hardware... some pre-calculation before writing the actual song data to ROM would help simplify things eg by always having the loudest thing in channel 0) alongside 4 pieces of rate scaling, and maybe a final divide or truncate. Essentially the same as getting 5 channels (4.5?) out of the STe (4 channels there would only need 2 additions) or 6 (...5?) out of the Amiga, so a playback rate in the 20s of khz should be doable if nothing else was going on. Keeping it going alongside a game might be more challenging, but as it was only used alongside title and credits screens (in the Toy Story case), it's not anything that we'd need to worry about. Figuring how bad it may be in more challenging cases requires looking at PSG based machinery - and the example actually shown off in this video can be a good guide. Other more comprehensively exercised examples are the non-e ST's and things like the Spectrum, MSX, Tandy PC etc. They can be manipulated into producing sample output with about 4 to 8 bits of effective resolution, nominally up to very high frequencies, but the amount of software calculation required to come up with the appropriate PSG values tends to mean relatively low sampling rates, with there generally being a tradeoff in terms of bit depth vs frequency (4 or 5 bits, depending on the actual chip, is essentially plain PCM, because it can be pulled out of a single channel, or the master volume register; anything greater means using combinations of channels, possibly with the main volume too, and relying on some maths and some look-up tables, and maybe a hell of a lot of precalculation before writing the ROM... or in other words there's a quality vs how much effort you can be bothered to put in tradeoff). On top of which there's all the other stuff about combining multiple samples into a single limited-precision channel, the calculation required to produce the combined value that then has to be translated into PSG commands, utter lack of filtering at any stage, etc. Good results were achievable, albeit with a lot of work and mainly but not always with static or fairly simple screens. A game on a PSG system that had a good "digital" soundtrack alongside any kind of serious action was a particular triumph - even now I'm having trouble thinking of any that weren't essentially early 80s arcade clones like Pacman With Fancier Sprites And Music and so-on. Even then you might really be looking at sample rates in the low or mid teens khz and effective resolution of maybe 6 bits in the main. What actually was rather more common than a full MOD-style soundtrack on that hardware was to use samples in the same way as a lot of the examples in this video - to provide fixed-frequency, 1-channel "digidrums".

I mean, the original Amiga computers, at least most of them, had only 4 sound channels total. Granted they were PCM, but if you wanted music with a lot of depth it required effort to do so. Limitations breed creativity, of course.

European game developers in general back then produced some wonky games with amazing music. See: Monty On The Run for C64. Looks terrible, plays meh, wow that soundtrack.

domdomdomdomdomdomdomdomdom

kuma

@@khhnator Ma-ma-ma-ma-ma-maruyama.

Strider II/Returns too. The game is garbage but the sampled speeches amazing.

You got another try!

i agree. older games than ps1 arent really my thing when it goes to actual playing, but i sure as hell enjoy the music and the videos about it!

+andyukmonkey The 8-bit DAC is fed with the sum of two 7-bit "virtual channels". This sort of thing was very common on the GBA to produce sample-based music.

za909returns That's the sort of thing you wouldn't expect in a commercial Megadrive game but rather a tech demo or something.

A MOD audio?.

@@user-ql2re2es9y Yes just me suffering things straight-up disappearing then...

Is that 2 samples tho? It sounds like 3 (2 vocal + drum, you can hear the higher vocal cut off sooner than the lower)

Samuel Shock *blind kid trying to play football*

GameHut (main developer) explain this feature on youtube.

Blast Processing in action XD

Should be Randy Newman/Allister Brimble. Newman composed for the actual movie,while Brimble converted the music.

I think the key is using the chip and the samples as an "instrument" in their own right and getting the best out of it by playing with what it can do in its own unique way, rather than trying to emulate an existing physical one and it being all too obvious where the chip then falls short of that. Like, the guitar sample is just the starting point for what you then do with it in the composition, rather than an end point in itself.

Yes quite as good this time

+Okami McCormick™ on the Mega Drive? That's a tough call...probably a tie between Thunder Force IV and Rocket Knight Adventures.

@Lucas Molessani Since nobody asked, I have a tough time choosing between Revenge of Shinobi, Herzog Zwei, and Sword of Vermilion. Also Gauntlet IV.

@Lucas Molessani yeah Mega turrican has a huge Snare sample, but the rest of the FM patches are pretty weak in my opinion

Agreed on Sonic 3. That's why I, unpopular opinion, think Sonic 3 has one of my least favorite Sonic OSTs ever made, only behind Chronicles and Boom RoL of course. I'm sorry, but I just think the entire game was overrated.

Sonic 3 did at least set the PCM channel to left or right on occasion, most Mega Drive games didn't even go that far so at least that's something. The obvious example of this is Hydrocity Zone Act 2.

@@stephandolby Sonic 3 is actually my favourite soundtrack of the series, especially the first six zones and Sonic 3 alone's exclusive themes.

What does DualPCM do then? I assume uses the regular channel with volume manipulation of either PSG or FM channels to add in missing LSBs and improve the dynamic range?

@@solarflare9078 there's no way you compared 3 soundtrack to sonic boom rise of lyric for me it's a really good soundtrack and I don't know why the PCM samples are what determines the quality of a game OST even though sonic 2 had some pretty low quality samples compared to 1 and 3 samples are better and more creative than 2 and 1 which are just snare drums so yeah can't see how that game OST is bad to you but I guess 🤷‍♂️

+Nick Thuren Check out his cover of "Anvil of Crom".

+SpiralPegasus So good I almost punched a camel.

Funnily I don't think they actually sounded any different in the PC version... the audio driver was probably fine, they were just really poor samples for some reason!

me too

I believe the sample channel works like any other FM channel on the 2612, meaning that it can play left, right, or both at once. As long as the sample channel is playing one sample at a time, it can alternate this way.

Sonic the Next Level has sampled music during actual gameplay (meaning in addition to the game running + some sound effects).

Yeah but that was based off Tweaker's own music engine.

Yeah, but Traveller's Tales still had quite a good handle on the Mega Drive to get that stuff, and other things besides, to work on the same game. They even managed sprite scaling.

+Purple Trolling? Anyway it's because it was the earliest attempt like this on Mega Drive and it's already pretty cool! But, as it has already been pointed out, the Mega Drive can handle streamed music during gameplay, alongside sound fx, as Sonic the Next Level does. And Markey Jester, the author of said demo, recently upgraded his sound engine... :)

S1 Next level does little if no actual sample mixing (I think the latter), unlike Toy Story. It's not the same to mix channels than playing long hand-picked samples in sucession. The reason "only the title screen and credits musics are sampled like this" is because mixing instruments is CPU intensive, and these screens happen to require less processing time (unlike a platforming level or so).

It seems to be a cover of the track "Battle of the Mounds" from the Conan the Barbarian OST, which cover, unfortunately for us, seems to still be unreleased. You can hear the snippet from Conan in question by following this link: kzfaq.info/get/bejne/aciYaZagtrHIpac.html and going to 1:56 This info was posted somewhere below but I'm paraphrasing and condensing from multiple sources and posting here for visibility

Now I really wanna hear that music, holy crap

@@zacharybrown3010 It's more likely a cover of the intro theme to Alternate Reality: The City and Alternate Reality: The Dungeon, which was itself an homage to the Conan movie soundtrack. It was released on a ton of microcomputers back in the day. The track didn't translate well to the Amiga but the Atari 8-bit home computer and C64 versions are legendary.

@@emmettturner9452 It's very clearly more like the conan track.

Toy Story actually mixes the 4 Amiga MOD channels to the 1 channel on the fly with a driver in the game

@@RWL2012 at what khz and bit?

@@rustymixer2886 around 10K hz or something like that.

@@crimson-foxtwitch2581 ah ok thx

@@rustymixer2886 It’s actually 8 KHz

An homage was used in Alternate Reality: The City and Alternate Reality: The Dungeon main title themes but this is likely an homage to/remix of that. Granted, I haven't heard every possible version of Alternate Reality since it was released on nearly every microcomputer (Atari 8-bit, Atari ST, C64, Amiga, Apple II, DOS, etc).

mega turrican's drums sound really clean

Arf, I expected to find the ref of that track in the comments... Cuz it's indeed amazing.

MastaGambit, Savaged Regime produced that himself. Look at his Videos for Anvil of Crom. The dude's a beast; there are some good modern YM2612 composers out there, but he may be the best. Also check out his renditions of Ken's theme from SFII and Theme of Simon from Super Castlevania IV.

+Nathan Daniels But Masta and I want the music that plays here, during the first 50 sec of the vid (and which doesn't appear in his Anvil of Crom cover). It's so epic and beautiful...

Oops! can't help you there. It IS amazing.....

+Nathan Daniels Let's hope it's one of Savaged Regime's future uploads :)

and I'm sure people downloaded the audio of this video to use in their own MOD tunes :D

@@bsdims idk where to download its separate wav/flac samples?. maybe you need amiga music trackers to rip it like openmpt, and many more.

@@user-ql2re2es9y you can just download this video's audio, plop it into the samples tab in OpenMPT, select a portion, copy to a new sample, crop off and loop here and there, save, have fun (oh yeah, and tune your samples to C- if they're not percussion) :)

JoseBronxRican u

konakonaa Sample playback is actually pretty great for the most part in GEMS (See also: 2:46). In fact, it's the one thing that sound driver does right.

@@solarflare9078 The only real issue with GEMS was the dogshit stock patches and samples that most composers didn't have the knowledge to swap out. Sonic Spinball and most games scored by Drossin use GEMS (being lead dev on the project helps a bit), Ecco the Dolphin (and the sequel?), probably a few other classics you don't think about because they aren't hitting you with the fart bass constantly

@@shukterhousejive Yeah you're right, alot of devs liked using that ''AHa Take On Me'' bass patch for their GEMS osts. Chakan, Barbie and Ballz 3D are some of the worst sounding GEMS you could listen to.

@@shukterhousejive Ecco didn't use GEMS, and neither did Barbie @Adrian. Ecco used Novatrade's in-house driver and Barbie used Sound Images, iirc

😂

Instead of using a separate cpu, the Z80 IS the cpu, costing a lot of cpu time.

The Megadrive effectively has an entire SMS built into it, dedicated to reading the joypads and playing music, pretty much. It can be completely maxed out doing that, so consider what would happen if you tried to play a game at the same time without the rest of the MD bolted on. It really doesn't have that much processing ability and was never meant to have any sample ability at all, hence the offering of the FM module in Japan... Any machine that managed such things was either 68000 based / using some equivalent 16-bit setup, or had an audio coprocessor. Sometimes both...

sparkster7744 16 bit version of Battle of the mounds, from the Conan The Barbarian movie

@@willthomas2310 ...and the main theme to Alternate Reality.

It uses an early version of Sonic 2's sound driver, which is actually a Z80 port of Sonic 1's 68k driver.

TheRedFox I can't give a technical explanation but by using all the psg sound channels you can play pcm samples, or something similar. There were a few Master System games that used the trick but the music had to cut out briefly for the sample to play. There are videos on KZfaq that compile examples of this from Master System games.

Ok, thank you!

Basically if you set a channel to one of the frequency extremes (can't remember if max or min, but it doesn't really matter for now), it will hold the square wave output as a steady "on", and then you can directly control the sample value by changing the channel volume. Think it might even be by design (or a happy accident they decided to leave in) and mentioned in the data sheets. Works similarly for both the SN7 PSG and the more common AY8910/YM2149. It's kinda crude and only really has 4 to 5 bits of resolution (made both worse and better by the volume being logarithmic not linear), and any kind of decent sample rate murders the CPU, but it's better than no PCM at all and can be quite effective in the right circumstances. Plus, and IDK if it's doable or was ever used on the MD, there are additional techniques - particularly for the 2149 - which can improve the apparent bit depth somewhat, by using all three channels and adjusting their volume independently, which can give you an extra bit and a half or so of dynamics, but that puts even more demand on the bus and can't really be used for live mixing (rather than straight replay of specially encoded samples or looping pieces of music) because the order in which you address each channel becomes crucial and really needs to be worked out algorithmically ahead of time on a much more powerful computer...

But the YM2612 was already enough from the start. Lazy and/or inexpert people will always produce shit results even with the most up to date hardware.

@@marcelopires711 the 2608 is still better, though. But as Sega retained the original SMS hardware and threw in a parallel 68000 and DMA system as well they probably thought it unnecessary to retain that and the PSG parts of it. Bit of a shame really, even though it did make the machine cheaper. Six PSG channels and two squares (plus the envelope generator and wider frequency range for half of that) instead of a bare 3+1 wouldn't have gone amiss alongside the FM section, nor having a more automated and natively ADPCM capable sample engine... More channels but less CPU load (and the potential of upto six scratchy PCMs alongside the main 8-bit one), what's not to like? Anyway to answer Jg's original question... What if, indeed. The Z80 should have been entirely capable of feeding instrument and note data to the FM and PSG parts, and PCM would no longer have been as much of a pain and would have sounded much cleaner and more consistent, as to the copro it would have seemed just like another such channel but with a different set of data (all of about seven or eight bytes per instance, or just one or two if repeatedly playing the same thing) going into the registers to initiate the DMA process. After which the sound chip would have been able to bus master during DMA cycles and fetch the much bulkier sample data for itself without any timing jitter. Which is why it sounded so good in the PC98 et al, and why it's part of a fantasy spec of mine for a better attempt at the Atari STe. You get a lot of capability for minimal CPU load, and it can just drop in place of an existing PSG from the POV of the computer bus and software (and fairly minimal board rework and addition of a DAC on the other side of the hardware bridge). It was actually meant to be used in conjunction with relatively low powered CPUs originally, after all, which wouldn't have had the ability to brute force the samples into a simpler MD style DAC whilst still being able to do something else. The only problem would likely have been the cost. Yamaha were probably a bit ambitious thinking it would get paired with noddy CPUs, and IRL it was mainly standalone synths/keyboards, arcade machines and higher end home computers that made use of it or anything particularly similar. It was even too pricey to make it into the MD, so it would never have been a practical option for the SMS even if it had been available a little earlier. That's why the SMS FM option is a much simpler and cheaper OPLL instead of an OPNA...

Lower the sample rate to something like 8KHz, and introduce small cuts of silence into the samples every so often. This will mimic the way unoptimised Z80 drivers input interrupts into the samples during playback.

ChiptunedRaijin Sure, I could do all that, but what about the tinny effect? I tried to test the 8khz rate on a voice clip of Marcus from Fallout, but whenever I do that, the clip sounds muddy instead. I even tried to save it in unsigned 8bit or 4bit. They said something about turning off interpolation, but I can't find any such option in neither Audacity nor FL Studio.

+ServiceGamer9 I've got OpenMPT installed, and I can see something related to samples, but I don't know how to turn off interpolation still. Any help?

+Hal Coley Well, it is a (converted?) 4 channel MOD file, after all (the original file is actually tied in with the credits theme). Funny thing is that the song is listed in this video as being by Randy Newman, while the MOD file is listed as being by Allister Brimble. Not sure who's correct here.

+Scooblee It's composed by Allister Brimble. Randy Newman did the soundtrack to the actual movie. xD

+Scooblee Randy Newman, arr. Allister Brimble. Strange Things for the Start Menu; Friend in Me for the Credits. Feel free to look up Toy Story in the Mod Archive! :D

+littlevince104 "Battle of the Mounds"

And the remix? :/

*opens MODtracker* thanx for the idea fam lol

MPC?

Akai MPC. It’s the classic foundation to making sample-based music.

@@exactspace explain?

@@markpenrice6253 Sure. For instance, you could use the same MPC workflow, where you have samples assigned to each of the 16 pads and play them to a click. Then, you could have each pad output a midi note message during the recorded playback, and capture that with a computer. Then you could do the necessary conversion of the midi note information to whatever you'd need to use in assembly to play back the matching PCM samples you used on your MPC.

it's a stock sample

+za909returns yep, most sound drivers run a loop on the z80 that streams the sample data to the DAC channel, and when the vblank interrupt occurs the code that updates the FM and PSG channels is run. some drivers do those updates with the 68000, which allows the z80 to keep running the sample loop throughout. fully z80 based drivers have to pause the sample loop to update the other channels. there are trade offs either way, partially 68000 based drivers allow for slightly better quality samples, fully z80 based frees up a bit more cpu time on the 68000 for gameplay

Crash I imagine with some very hard work and a thoughtfully designed FM and PSG routine you could still keep the PCM running, but you'd have to manually time runs of the sample updater while doing your FM and PSG stuff... So a sample should be output to the DAC every x clock cycles, and therefore you have to call the same routine that is normally running during the rest of the frame. Sucks that the Z80 only has one interrupt available. This idea sounds like a nightmare to implement, kind of like doing anything on the Atari 2600.

how do you cycle count effectively in a sound driver with 10+ channels? not every channel updates every frame. 90% of frames will have no channel commands issued, then all of a sudden almost all channels will be updated on the same frame. Do you just waste the cpu time required to update every channel on every frame? do you only allow 1 or 2 channels to be updated per frame? there's not really an optimal solution to this without introducing some severe restrictions on the music composer

As long as the time difference between those successive samples output during the sound driver execution isn't that great, it shouldn't be that noticeable. At the very least it should be better than no output at all during that period.

there's another proven option: a big old ram buffer that gets streamed to the dac channel. i don't think it was ever used back in the day, but in modern times multiple people (stef, markeyjester) have made drivers that use a buffer to 1) avoid dma bus contention 2) software mix multple dac samples 3) fuckin just be cool as fuck

Really? It kinda reminds me of Strange Things..... oh wait

sample decoding in realtime, plus maybe using the square-tones to replay samples, just a hunch, I know some ZX-Speccy tunes use samples.

@@bsdims WHOA! I had no idea you could play samples like that! I always thought they were playing square waves so fast they sound like 1-bit samples.

Cuz the SMS didn't have enough power through the rest of the system (half the clock frequency, half the bus width... In fact effectively a third of both because the MD hardware has a conjoined SMS running in parallel... Plus less cartridge space) to handle it. The capability was there and it can be seen doing similar whilst showing a static screen, but it puts so much load on the machine that it can't do very much else at the same time. Also the technique is rather less efficient than even the MD's unsophisticated PCM output. Afterburner probably did things that way as percussion generally doesn't need as much dynamic fidelity (ie, bitdepth) as other samples and it may have wanted to play cleaner sounding PCM SFX at the same time as the music. Plus it would have saved a bit of cart space if the samples could be stored as 4-bit values instead of 8-bit, matching the PSG's output capability, and IIRC the Z80 has built in capability to handle such data encoded into byte based memory thanks to its shared history with the early Intel chips (originally meant to go into calculators, cash registers, etc that had need of such processing) so splitting it out and sending it serially to the sound chip wouldn't be as much of a time consuming pain as with even the 68k...

it's only an 8-bit CPU, so can't process the data as fast as the 68K

It's not the processor so much as the way it's used and the number of tasks it has to handle at once. Feeding samples to the sound chip requires cycling the code in a very regular fashion (eg 10,000 times a second without more than maybe five percent variation cycle to cycle and much greater consistency than that on average) - and if something causes the chip to be occupied for a significant period then that gets interrupted (eg something that distracts it for a full millisecond loses ten samples...) which distorts the output much like AM radio static or a bad speaker cable connection. Doing what is essentially copying 10 or 20kb of data per second between two places in the memory map, and keeping track of where it is and should be vs a steady clock, is well within the capabilities of a Z80, but interlacing that neatly with other routines demanding processor time is the trick, especially if any of them are timer interrupt driven rather than using cycle counting and unrolled loops, because that does exactly what it says on the tin and breaks into the running code wherever it may be, taking a certain amount of time to save the current CPU state into memory and then reload it whenever control is eventually returned. A clever and experienced music programmer who's given enough time to craft their routines with care and isn't expected to do the impossible can make it work like a circus juggler. A newbie or someone having to take on the job when their expertise lies elsewhere, racing a deadline on a low budget or mismanaged project that's trying to be too ambitious will probably fluff it and struggle to the point of just having something that sounds kind of recognisable before time runs out.

I think the guitar and the drums are in one big sample

Actually, 6 bits each.

Probably adding the sample values together then blasting the resulting samples to the Digital Audio Converter.

@@Mizu2023 can be easy trick using an auto chop software like recycle from propellerheads software, back then you can also try with an akai hardware sampler but all sample must converted lofi with the sega sdk.

Cuz snes dont need to skeet, its already sample based unlike md

@@LevskiBADBOY heck, its even got several strides on the Amiga in all ways except for amount of sample/score memory, unless you're one of those clever types who've figured out livestreaming of fresh data to the SPC700 whilst it's still playing... More channels, full coprocessor rather than just a DMA mule, etc...