Unit selection in a concatenative speech synthesis system using a large speech database

Abstract

One approach to the generation of natural-sounding synthesized speech waveforms is to select and concatenate units from a large speech database. Units (in the current work, phonemes) are selected to produce a natural realisation of a target phoneme sequence predicted from text which is annotated with prosodic and phonetic context information. We propose that the units in a synthesis database can be considered as a state transition network in which the state occupancy cost is the distance between a database unit and a target, and the transition cost is an estimate of the quality of concatenation of two consecutive units. This framework has many similarities to HMM-based speech recognition. A pruned Viterbi search is used to select the best units for synthesis from the database. This approach to waveform synthesis permits training from natural speech: two methods for training from speech are presented which provide weights which produce more natural speech than can be obtained by hand-tuning.

Towards automatic cross-lingual acoustic modelling applied to HMM-based speech synthesis for under-resourced languages

“…Currently, unit selection [7] is still the underlying technique in most commercial systems. Since this requires large resources of well-recorded and labelled speech data to ensure the optimal unit coverage, its use is becoming less common for obtaining cheap and fast language portability, especially in the case of an under-resourced language.…”

Multisyn: Open-domain unit selection for the Festival speech synthesis system

“…A comparison between the Multisyn engine and two of the first unit selection implementations -CHATR (Hunt and Black, 1996) and Festival's "clunits" method (Black and Taylor, 1997)-is useful to clarify how Multisyn differs from other techniques.…”

“…A full tutorial on unit selection speech synthesis is beyond the scope of this paper; we refer the reader to Hunt and Black (1996). However, we will define the terminology to be used in the rest of this paper.…”

“…Unit selection speech synthesis (Black and Campbell, 1995;Hunt and Black, 1996) was proposed as a way to solve some of the problems of unnaturalness introduced by the signal processing techniques needed to produce convincing synthetic speech from a database consisting of a single example of each diphones that occurs in a language. Instead of having one example of each diphone, a number of examples in different contexts are included, and the synthesis process is formulated as a search problem.…”

“…The general search method (Hunt and Black, 1996) has been refined (e.g. Conkie, 1999;Taylor, 2000;Bulyko and Ostendorf, 2001) and complemented by other procedures for specific tasks such as limited domain speech synthesis (Black and Lenzo, 2000).…”