subject: Square Enix's Craftsmanship To Eliminate Automatic Frontiers [print this page] CEDEC last day of Square Enix's research and development department "Natural Expression Natural aims to represent" the lecture was entitled.
The first stage of his development director Huzii Eizi. First, I talked about why Natural Expression is important.
Natural Expression and literally, and without representation "natural" but which, Natural as a means of "crude" because it was emphasizing things that, as I have also add a hand crude It seems that what we aim to automatically generate the data.
In other words, Square Enix company, though, the artists and technicians to hire craftsmen, the data captured by motion capture animation animator still went to fix the cross (cloth), they dress out from the kneecap in the simulation Cod, will have to manually correct. In terms of making the game more Natural previous example is not handling that they require more processing craftsmanship. But it is not easy to increase the craftsmen. In addition, CG and parts production means, the cost of equipment and software that I have been on the rise in labor costs down. This basic proposition is that it can not control or take advantage of the power of high-performance hardware continues. Surprisingly little seems to be cutting-edge technologies being developed as part of an attempt to reduce costs through efficiency.
Such in-house automation of Square Enix seems to be some efforts have been made, this time related global illumination, were announced between two types of speech recognition.
GPU computation of global illumination
CEDEC, including coverage of events in the global lighting technology that you have mentioned several times: global illumination, shading to create a soft and natural, in recent years could say that technology is becoming a must. Square Enix, the CG in non-real time, the operation of global illumination using ray tracing that it is accurate GPU-based system to do.
In the example shown at the beginning rendering, photon mapping using 10 million photons (photons from the light sprinkle, method of mapping the surface brightness objects) were presented with images of global illumination. Detarashii is rendered normally takes one hour to full HD resolution GPU6 units (in the description had been so late and Quadro Plex FX5800, GeForce GTX 280 Do not worry I equivalent) with about 30 seconds, rendering Then say. Next, explanation is made about the characteristics of the system.
Parametric surfaces (such as Bezier surfaces or spline surfaces) processing system to make direct rendering, 3D polygon data input and modeling tools made based on a parametric surface, and flows into the parametric surface rendering and seems. In this system, to increase the parallelism that has been adopted and breadth-first ray tracing. In ray tracing, the ray from the viewpoint for each pixel (ray), but will not skip, repeat the reflected ray, which determines the final color when the light reaches. The usual method, and this depth, the breadth of operations in advance of full brightness from each end point is not that type. That depth and breadth but different algorithms are often used as exploring the tree, when ray tracing, skip over each pixel to lay the primary first overall in stages and beyond lay the second from the top reflecting surface breadth-first ray tracing is equivalent to working through every single level.
The Neumann series expansion for ray-tracing equations used for rendering, it is redefined as a series of reflections per order will be processed for each order (we hear). Optimization algorithm is interesting here. Square Enix systems that are coherent ray tracing technique is employed Hachisuka PM: Megumi. This is roughly a collection of scattered ray obtained in the breadth of each pixel, but that method to calculate things the same direction at once. Also, do not know which part but often data are taken Oburibiasu cash layout, and that the data have been placed along a Hilbert curve. Hilbert curve, the algorithm will recursively filled with stroke from the edge plane or space can be redefined one-dimensional space, so something relatively close Kiyasui back and forth into space, the cache Kino Toshi Good as they are useful methods. By skipping the first ray and said that the Hilbert curve is used in order to collect a skip when the photon ray.
Looking at developments, but it does not appear to be entirely complete yet, as already in practical use, (flowing Worse was) and that used in the titles to be released later this year. In addition to creating Purirendamubi, that can also be used for baking and light mapping. Incidentally, NVIDIA has "started Reitore" as was a slow news recently, NVIDIA GPU from the previous approaches to render using Gelato is promoting a project name. MAYA from using it as it is supposed to be rendered directly, and asked if this use "Gelato is actually much faster," he said. Apparently, seems to require a fairly sophisticated global illumination would come to influence and usefulness have not optimized for it. The technology is the future of rendering in real time to realize that, after five years of hardware (the console Flachot did) that it is targeted to be processed in real time.
Automated lip-sync
The following were the study's speech.
This is the lip sync, that is, that it intended to make adjustments to the lip line of the characters spoke the voice and game audio. Recent games, talking earnestly and so many characters, this work has been something that would require considerable effort to imagine.
More words on every single movement, we decided to look and animation of facial expressions. This work has likely experienced CG was done in three animators, 0.6 person-days per second, per person per day means that it could only 1.6 seconds of processing. 5 seconds each day with the team. No wonder it would not want to automate. Rather that generate appropriate animation of the mouth of the voice spoke the words, so simple and difficult. The scheme seems to have tried various methods to apply a universal database to create a single phoneme from a lot of data, voice actor, and that as more data become counterproductive. Phoneme accuracy and create a database in one voice was raised by, the work needed for each voice. And work to create a database with words, almost 100% identification rate, but was practical, too much work and can not. Ultimately, the decision was a practical method of voice processing units.
In making its identification engine itself, and considering the results of existing engines, and ultimately was selected as the engine seems NHK has provided Engineering. Adapt well to consider that the process was sequential in the future, and it has been disclosed and acoustic model adaptation method for data format, and that there was a significant difference with in-house and more . By this system could cost as compared to conventional 1 / 8 becomes, it can achieve better quality. The title will be released the future, that that is applied to this system (FF that and also used series), have a chance to see the cut scenes, it is running such a system behind Let us recall that.
Square Enix is planning to expand its operations outside of these automated, especially "virtual actors" such as planning and portion into the crowd, motion, for each of the Feisharuanime, who have lost a key frame animation so that what one generation can be quite high expectations. If able to introduce a more effective part of craftsmanship, quality improvement can be expected that the game itself. We also hope that further research in the future.
