(Images used for review purposes only.)
I want to be a Vocaloid Producer, vol. 6, 1,500 yen, plus tax.
Time to set up a regular format for these blog entries. But first, some comments.
I’ll be honest, these training DVDs aren’t the first thing I automatically run to when I get up out of bed in the morning. There’s a number of other things I’d rather be doing instead, or that I have no choice about. Not including my regular work, I’d gotten my hands on a couple Gameboy games, and that ate up 3-4 weeks of free time. Plus, I’d written up and illustrated the Riemann articles, created two animations for that and then came up with soundtracks, which took another few weeks. Then I accepted a part-time job to type up transcripts for a 3-hour business lecture in English, and a short trial project for doing an English rewrite, which combined represented a total of 60 more hours of work. In with all of this, I’ve been creating patterns in Gimp that I’ve been running on Facebook for the last 2-3 months.
What this means is that I’ve fallen behind on the Vocaloid tutorial videos. On Tuesday, I had a little extra time, so I sat down to finish the MMD tutorial from volume 4. This has been the most time-consuming one so far because I’m creating a walking cycle for Rana, and it takes several minutes to make each small move, then switch between MMD and the video to make sure I have the pose right.
MMD really is very powerful. If you select the right wrist and then pull it to the right, the bone structure is such that the upper body is pulled too, in a way that looks kind of natural. Lifting the right knee causes the upper thigh to rotate properly, and the lower leg and foot to lift the corresponding amount. But. While MMD does have English support for the menu items and buttons, the model components are mostly in untranslatable Japanese. So, I have to switch from Japanese to English to determine what the buttons are for, then back to Japanese to try to hunt for the model items “right hand” or “upper torso” prior to trying to position them.
Eventually, I did get halfway through the volume 4 tutorial, which consisted of raising the left arm and right leg in a marching pose. From here, you select the entire model, copy the current bone positions, select frame 10, and do a flip paste. This creates a key frame, but with the right arm and left leg raised. Select frame 20 and do a regular paste to make a second key frame that returns Rana to her pose in frame 0. Then, play the walk cycle in a loop. Somehow, MMD decided to reject the motions I’d registered for raising Rana’s legs and swing her arms properly. After 1-2 hours of work, I had nothing.
On the other hand, the tutorial goes on to say that Rana’s motion looks like she’s swimming through water and that there’s a lot of clean-up that needs to be done to fix things. The entire tutorial is about 9 minutes long, and I’m only at the 6 minute mark. Sigh. I’ve only briefly looked at the volume 5 MMD tutorial, and haven’t done anything with the volume 6 DVD at all.
One of the things about stepping away from a problem is that almost immediately I started coming up with work-arounds. The tutorial occasionally reminds the viewer to click “Register” for the Bone Manipulation section, but I apparently missed a key point for this. Registering updates the current key frame (0 frame) only for the currently selected joint (right wrist, left ankle, neck), so EVERY time you move something, you have to register it before selecting the next joint. However, all of the positioning I did on Rana is still in effect if I save my work and reopen the file later. Couple this with the use of “V-Sel” (Select all joints) in the Frame Manipulation menu prior to doing the key frame copy-paste step, and maybe my fix would be to select all of the joints and then Register them in bulk. Using V-Sel only selects the registered joint positions. So, I tried using SHIFT-click to select a full range of joints at once, but this works more like CTRL-click in other programs (selecting only one item, not a range). Eventually, I selected all of the joints by SHIFT-clicking each one one at a time, and then clicking Register once for each block. So, yeah, now the walk cycle is broken exactly the same way as in the video.
The next step is to move the key frames for both feet so that they only move during a shorter period than the arms do, and add 2 new key frames so that Rana’s knees bend mid-step. After this, we drag and drop the “Rana theme music”, add a few blank frames to make the march line up with the music tempo, and finally copy-paste the walk cycle for the full length of the song. A possible bug I encountered here is that certain key points for specific joints don’t advance when a blank frame is inserted or deleted, and occasionally the copy-paste feature “loses” some of the selected key bone points. And, it took me a while to figure out how to delete unwanted key points that got pasted in the wrong place (rubber band select, or SHIFT-click select, then use the Frame Manipulation menu Delete button.)
Playback of the walk cycle matches the demo video. The tutorial suggests making small tweaks to have Rana’s mouth open and close to mimic singing, and you can make her eyes blink, if desired, as “self-study”. I didn’t bother with this part.
Once I had the vol. 4 MMD tutorial out of the way, I could move on to vol. 5. Here, the primary tasks are to import accessories and relate them to Rana. In the Model view, you can select specific accessories, such as the microphone or the dance stage, move them around, and change their size or transparency. For this tutorial, you import the Rana model from the vol. 1 DVD, and follow that with a pre-made vol. 5 motion file to give her the marching cycle again. There’s an interesting feature where you attach the mike to a specific joint of the Rana model, and any movement of that joint causes the mike to track appropriately. In this case, it’s Rana’s wrist. Use the pre-made pose off the DVD for Rana’s right hand half-open, and position the mike next to her palm, then individually curl her fingers so she’s kind of holding it right.
The idea is that she’s supposed to be singing in the mike, but the walk cycle has her swinging her arms, so you have to delete all of the key frames for her right arm and hand. When Rana is ready, delete all of the key frames for her main body position, import the dance stage, and move Rana as a unit up onto the stage. Whew. Done with this tutorial.
I’d jumped over the vol. 5 Vocaloid tutorial last time, so I came back to that, too. The focus is on Dynamics. This changes the pitch or energy of Rana’s voice for specific phonemes based on the current assigned note so that it sounds more human-like. As an example, say Rana is singing “la” for a full second. With a flat DYN pattern, the “La” sounds flat. Instead, you can draw a curve for DYN (which goes from 0 to 127) so that the “La” starts low, rises to a peak and then falls back down. This works just like the Pitch control from vol. 4. You can use straight lines or draw freehand with the pencil tool. That’s the entire tutorial.
Moved on to the vol. 6 DVD. See below.
New format for these blog entries
New magazine features:
Rana learns about the different parts of the guitar, and how to play simple chords. However, regular guitars are too large for her hands, so Robo-panda gives her a new accessory – White-kun, a voca-tar. Which she still can’t play. Then there’s the interview with Vocaloid Producer Mikito-P, an overview of the Chika Vocaloid model package, and a short write-up on the pick-up artist, Digital Sound Laboratory.
New DVD Features:
Video by Digital Sound Laboratory, entitled: Radiance of Nature
This is a kicking little piece that starts out with a lot of synth effects for the first minute, then settles into a pop-disco rhythm with a rather typical 80’s anime “looking to the future” lyric. Taken together it’s Jean Michel Jarre meets Macross. The video is just a still of Rana from one of the magazine covers, with a transparent “moving spotlights effect”. The music is good, but there’s nothing to the video, model animation-wise.
Rana’s voice samples:
Another 27 samples, laughing, talking and counting “1, 2 3”, with variations.
(Extras page, with Chika, and Digital Sound Laboratory.)
Sing-Songer Writer has a large library of pre-written song examples, that it calls “arrangements,” for each genre. This includes Jazz, Cafe, Rock, Country, Fusion, etc. Each arrangement is made up of 2 to 5 instruments, and divided into “intro”, “A set”, “B set”, “Final” and “Closing”. The tutorial is to take the demo song and fill in the empty bars with parts of a specific arrangement. Some of the arrangement combinations come out pretty cool sounding, but I think this goes a long way in explaining why modern J-Pop has such a cookie cutter sound to it.
This time, it’s the Velocity parameter, which works the same way as Pitch and Dynamics. Velocity is that MIDI parameter that states how hard (or fast) a keyboard key was pressed. Because Japanese phonemes are generally a consonant-vowel pair (“ko” or “Ju”), [Vel] can be used to add stress to the consonant portion, and have the stress drop off for the vowel portion.
This one is a really good tutorial. I just wish it wasn’t so hard for me to follow. Steps: Import Rana, the guitar, the stage, Rana’s full motion data, and the .wav soundtrack. You now have a complete music video, with Rana singing, dancing and playing the guitar. Your task is to add the camera and lighting keyframes. I spent several hours trying to relate what the tutorial is talking about to the menu options. Eventually it did make sense, though.
I was planning on doing some frame renders to demonstrate just how cool the entire model set looks, but for some reason, MMD just gives me black images. When I tried rendering to an .avi movie file, I was told that some .dll can’t be found in the data directory. Because the required file is actually in the right location, I may have to uninstall and reinstall MMD. I’ll wait to do this until I have to enter the next registration number in a week or so.