For years, there have been convincing electronic versions of many conventional instruments. The human voice, though, has alway been elusive.
Yamaha hopes to change that with Vocaloid, a new technology that lets you write vocal lines and have your computer sing them. The results are surprisingly realistic.
Yamaha is working with Zero-G and other companies to create voice libraries that will allow musicians to have a set of virtual singers within their computer.
The virtual voices are developed by recording professional singers singing series of nonsense phrases. These phrases are designed to combine all possible vocal sounds with all the possible transitions between them. By recording all of these possibilities, the sound designers have all the necessary raw materials to create a virtual voice.
The recorded voices are then analyzed using Fast Fourier Transform, and divided into individual elements with separate properties. These elements are stored in a phonetic database. Expressive properties, such as attack, pitch bend and vibrato are also stored in a database.
These various elements are drawn upon at the synthesis stage. A score editor, shown above, is used to enter music and lyrics. This is a standard piano/scroll editor as found in most sequencers. The music can be entered or imported as MIDI, while the lyrics are typed in.
Once the music and words are entered, the score can be synthesized. The engine pulls the elements from its database and combines them to create words. Melodies are created by shifting the fundemental and overtones, while leaving the resonant vowel formants constant. The virtual voice can then be rendred into an audio track.
The results are quite interesting. An example can be heard at the Zero-G site.
Zero-G has announced a licencing agreement to use the Yamaha technology in several products. It has announced three virtual vocalist packages, including male and female soul singers. No virtual Britney has been announced.