You are here

It's Not Over Until The Fat Computer Sings

Leader By Paul White
Published May 2003

I'm sure that many people will remember the huge outcry against samplers and synthesizers when traditional musicians felt their job security was being threatened; yet today, artificial sounds are accepted by the majority of composers and are used in many high-profile film scores, both with and without a traditional orchestra. However, I'm not going to follow this topic to the latest round of strikes by musicians on Broadway following a reduction in the minimum size of an orchestra, but rather look to see what other threats to tradition lurk on the horizon. One such technology was previewed by Yamaha at the Frankfurt Musikmesse in the form of Vocaloid, a software technology that allows computers to sing. Admittedly the technology is in its early stages, and anyone with half an ear can spot unnatural-sounding artifacts, but used carefully, it's already good enough to provide backing vocals behind a human lead singer.

Paul White c.2003Vocaloid isn't a harmony-generating technology, though: it's a singing speech resynthesis system, which was codenamed Daisy during development after the rendition of the song that was sung by the computer HAL in the film, 2001. At the demonstration, we saw notes being entered via a sequencer-like piano-roll editor, followed by lyrics that were typed in above the notes. Finally, a few simple tools were used to adjust inflection, tone, vibrato and so on. Within minutes, the computer was singing like a professional!

The first Vocaloid libraries are set to be released before the end of this year, as sequencer plug-ins plus the stand-alone editor. More singer personalities are also being developed by third-party companies. And it can only be a matter of time before other software companies develop their own virtual vocalist technologies if the idea proves popular.

The whole system seems so simple and effective that soon, anyone requiring layered backing vocals will be able to dispense with session singers; and while the voices display artifacts that give away their synthetic origin when exposed, they can sound rather convincing in an appropriate context. It's my guess that somebody will have a hit single within the next two years using the software to generate a lead voice where the side-effects become a marketing gimmick, rather like Cher's notorious Auto-Tuned single 'Believe'.

If you're sceptical, you only have to look at the way virtual pop stars are taking off, particularly in Japan, where 3D rendered female singers have their own fan clubs. All you need to do is team a virtual pop star with a virtual vocalist, and you can cut out the middle man (or woman) completely. If this isn't something likely to worry the traditionalists, I don't know what is, although I guess it will be a long time before bearded guys in white polo necks sit around a PC drinking real ale and joining in with virtual folk songs! When they finally do, however, I guess it will be to sing about how good the old version of Vocaloid was!

While the current technology isn't perfect, I would imagine that within three or four years we'll have synthesized vocals that sound indistinguishable from a human singer in the majority of pop music applications. However, I suspect that by that time, somebody else will have brought out a plug-in that makes human vocalists sound like the early version of Vocaloid, since, like most imperfect audio technologies, it will inevitably become revered as 'cult retro'.

Paul White Editor In Chief

Published May 2003