Interspeech'2005 - Eurospeech
A new time-domain, non-pitch-synchronous method for timescale modification targeted on broadband speech is proposed. The method is based on the SOLA (synchronous overlap-add) and EM-TSM (envelope-matching time-scale modification) methods, where the sign envelope of the EM-TSM method is replaced by a generalized envelope formed by the highest bits of the samples. (The actual number of bits will depend on word length constraints of the specific hardware.) In addition, a fixed length scheme for calculating cross-correlation is proposed, eliminating the need for normalization after computing each cross-correlation value. With these improvements, the proposed method outperforms EM-TSM both in terms of output quality and computational efficiency.
Bibliographic reference. Sakurai, Atsuhiro (2005): "Generalized envelope matching technique for time-scale modification of speech (GEM-TSM)", In INTERSPEECH-2005, 3309-3312.