SN5B Vocal simulation

Post your original music here and receive feedback.
Forum rules
This area is dedicated for original music, please do not post covers here.
User avatar
sterne
Posts: 31
Joined: Sun Oct 11, 2015 12:46 pm

Re: SN5B Vocal simulation

Postby sterne » Thu Mar 17, 2016 10:04 pm

I can understand DPCM speech more clearly, and that takes a tenth of the effort this does.
But... why?

User avatar
BioMechanicalDude
Posts: 309
Joined: Sun May 24, 2015 2:54 pm
Location: Bulgaria

Re: SN5B Vocal simulation

Postby BioMechanicalDude » Fri Mar 18, 2016 11:28 am

I don't mean to burst anyone's bubble here, but all these vocal simulations feel kinda... "eeh" to me. Maybe because it's in Japanese, but you can never tell what it's "saying". Now I know I shouldn't expect this with the 5B, but even with something like VRC7, the speech is practically unrecognizable. I think the best example of Famitracker vocal simulation I've ever heard is when someone made a FDS cover of "Still alive" from Portal.
Greetings! I'm That Bio Mechanical Dude and I like creating various stuff like movies, games and of course chiptunes!
You can check out my YouTube Channel.
You can also follow me on Twitter.

User avatar
Sky Yoshi
Posts: 985
Joined: Tue Nov 03, 2015 8:09 am
Location: Japan Osaka (。-`ω-)
Contact:

Re: SN5B Vocal simulation

Postby Sky Yoshi » Fri Mar 18, 2016 11:47 am

The FTM says
"ルカ ルカ ナイトフィーバー / Ruka Ruka Naito Fi-ba- / Lika Lika Night Fever"
and
"千本桜 夜二紛レ / Sen Bon zakura yoru ni magire / One thousand sakuras seeking in the night"
I think my Translation is pretty bad.
FYI, those words are from the song "ルカ ルカ ナイトフィーバー / Luka Luka Night Fever" and "千本桜 / Senbon zakura"
I'm Sky Yoshi and I'm a Japanese Tracker.
I am back from 5 months of absent!
Check out my Links to see what I do!
Do you wanna be pointless? Click here!

User avatar
w7n
Posts: 241
Joined: Fri May 15, 2015 1:37 am
Location: Nanamori-Chuu
Contact:

Re: SN5B Vocal simulation

Postby w7n » Fri Mar 18, 2016 3:26 pm

sterne wrote:I can understand DPCM speech more clearly, and that takes a tenth of the effort this does.

What's so new about DPCM vocals anyway?
The techniques of the NSFs in this post should most likely be used as tech demostration.

BioMechanicalDude wrote:I don't mean to burst anyone's bubble here, but all these vocal simulations feel kinda... "eeh" to me. Maybe because it's in Japanese, but you can never tell what it's "saying". Now I know I shouldn't expect this with the 5B, but even with something like VRC7, the speech is practically unrecognizable. I think the best example of Famitracker vocal simulation I've ever heard is when someone made a FDS cover of "Still alive" from Portal.


Let me explain some bits and pieces of vocal simulation. I may actually make a post in the General Talk section about this someday.

The simulated vocals in my covers from FCP2015 got somewhat mixed (though mostly positive) reception in the Japanese NES chiptune community. However, most negative comments weren't directed towards the fact that the vocals aren't clear enough (they are clear enough to be deduced for some), rather they were about the vocals being 「ベタ打ち」 -- direct transition with little focus on the nuances in the vocals, for example detunes, pitch bends etc.

The clearness of FDS vocals mainly depends on the sample quality. All that the FDS channel captures are the data in the 64-step waveform, thus clearness and brightness are often lost because what make up clearness and brightness are often something of higher frequency.
Different sounds' clearness in FDS vary. Take Japanese for example: Most vowels sound fine, although 'I' and 'E' sound bad in lower frequencies. Dropping a few variants, the consonants M,N,R,W etc. sound good if you can find good samples. The consonants K,G,Sh,J etc. take some efforts to sample; while the consonants T,P,F,H,Ng are a pain in the **beep** to sample. Sometimes, whether the plosives sound good or not depends on luck(the timing of the exact phases)
Simulating English would be a bigger pain in the **beep**, since English has more vowels and consonants than Japanese, and many English words end with plosives which are hard to simulate with FDS. (Plosive finals sound better when aided by the NOI channel.)

SN5B vocals should be classified as tech demos. VRC7 vocals have been experimented before in FCP2014, but I made my own version as I thought the vocals there aren't clear enough. Still, a VRC7 patch carries even less useful information than an FDS waveform.
But 'what's new with DPCM vocals anyway?'

Theoretically, now that the 64-instrument limit in FT is out of the way, if you have enough efforts to put into waveform sampling and go through thousands of trial-and-errors, you can make FDS vocals as clear and distinguishable as the hardware allows.
So that's it, clearness maxed out, that can't be more ideal right? Well, it's in fact far from ideal.
I would take the song Ojamajo Carnival (Ojamajo Doremi OP) for example: To create the overall cute and childish vibes, the vocals contain a lot of nuances like pitch bends, detunes etc. Now, try using just square waves and copy every single nuance in the vocals. That might even appear harder than directly simulating the vocals in FDS.
Not only that, certain pitch bends may happen in less than a frame's time and thus can't be recreated under 60Hz (well, the FDS FM doesn't work well, I tried that before). And, when a part of the vocals gets emotional, certain sounds would sound different from their usual counterparts and would require extra sampling.
Frankly, I'm yet to cover a song with emotional FDS vocals, and I'm studying these techniques. For an example of emotional FDS vocals, check Wiz.'s cover 'Koujou no Tsuki' bundled with his FaMiC program. (However the clearness of the vocals in that cover is abysmal, so I wouldn't pursue those techniques either)

So the conclusion: The 'best' vocal simulation is not mainly about clearness, but more about the nuances of the vocals. As for clearness, you can't require something more than the hardware allows.
nsf.nesbbs.com under reconstruction, better player might be available in the future.

VRC6 PWM is GAY because it has colours of the rainbow in the NSFPlay keyboard visualizer.