The classic description of unit selection, described as a search through a network.
Andrew Hunt and Alan W. Black. “Unit selection in a concatenative speech synthesis system using a large speech database” In Proc. ICASSP, volume 1, pages 373-376, Atlanta, GA, USA, May 1996.
Publisher’s version (normally this would be preferred)
Open Access version (this version actually prints in higher quality than the publisher’s version, which is a scan)
The most important sections of this paper are 1, 2, 4 and 5. Section 3 (Training the cost functions) discusses ways to automatically learn the weights. This would be important, if it it worked well. In practice, it is more common to tune the weights by trial and error.
