Comment Could it make dead singers "sing" again? (Score 1) 344
On visiting Yamaha's site, I read the follwing snippet in the Features section
--snip--
"singing articulations" (collections of voice snippets, such as syllables and snippets of vocal expression variations, like vibrato) needed to reproduce vocals, are collected from custom-produced recordings of professional singers and put into a database after conversion into frequency domains. To synthesize vocal parts, the system retrieves data consisting of voice snippets, applies pitch conversion, then splices and shapes them to form the words of a song as typed by the user. As this processing is done at the frequency-domain level, pitch can be easily changed according to the specified melody, and the voice snippets can be spliced in a way that reproduces smooth-flowing words. For example, "sai" of "saita"is produced by using two snippets "sa" and "ai". Because the timbre of the vowels "a" and "ai" are usually different to each other, if these sounds were simply spliced together they would not sound right to the listener. To solve this problem, smooth processing of the splicing facility within the frequency domain is carried out, resulting in a smoother vocal.
--snip--
From this, I gather that the synthetic "voice" of the program is constructed by linking together samples of vowels etc provided by an actual singer?
What if we could harvest samples from a dead singer (vocal-only tracks from a studio master etc). Couldn't we then make a dead singer "sing" again? New songs? The Elvis comeback single of '04?
I downloaded the first "Leon" sample on Yamaha's site, and it sounded very convincing.
da5id
--snip--
"singing articulations" (collections of voice snippets, such as syllables and snippets of vocal expression variations, like vibrato) needed to reproduce vocals, are collected from custom-produced recordings of professional singers and put into a database after conversion into frequency domains. To synthesize vocal parts, the system retrieves data consisting of voice snippets, applies pitch conversion, then splices and shapes them to form the words of a song as typed by the user. As this processing is done at the frequency-domain level, pitch can be easily changed according to the specified melody, and the voice snippets can be spliced in a way that reproduces smooth-flowing words. For example, "sai" of "saita"is produced by using two snippets "sa" and "ai". Because the timbre of the vowels "a" and "ai" are usually different to each other, if these sounds were simply spliced together they would not sound right to the listener. To solve this problem, smooth processing of the splicing facility within the frequency domain is carried out, resulting in a smoother vocal.
--snip--
From this, I gather that the synthetic "voice" of the program is constructed by linking together samples of vowels etc provided by an actual singer?
What if we could harvest samples from a dead singer (vocal-only tracks from a studio master etc). Couldn't we then make a dead singer "sing" again? New songs? The Elvis comeback single of '04?
I downloaded the first "Leon" sample on Yamaha's site, and it sounded very convincing.
da5id
