CHATR the Corpus; a 20-year-old archive of Concatenative Speech Synthesis

CHATR the Corpus; a 20-year-old archive of Concatenative Speech Synthesis Nick Campbell Speech Communication Lab, School of Computer Science & Statistics Trinity College Dublin, Ireland nick@tcd.ie Abstract This paper reports the preservation of an old speech synthesis website as a corpus. CHATR was a revolutionary technique developed in the mid nineties for concatenative speech synthesis. The method has since become the standard for high quality speech output by computer although much of the current research is devoted to parametric or hybrid methods that employ smaller amounts of data and can be more easily tunable to individual voices. The system was first reported in 1994 and the website was functional in 1996. The ATR labs where this system was invented no longer exist, but the website has been preserved as a corpus containing 1537 samples of synthesised speech from that period (118 MB in aiff format) in 211 pages under various finely interrelated themes The corpus can be accessed from www.speech-data.jp as well as www.tcd-fastnet.com, where the original code and samples are now being maintained. Keywords: ATR, CHATR, speech synthesis, corpus, preservation, concatenative synthesis, waveform samples, aiff, multilingual 1. Introduction The CHATR speech synthesis system was developed throughout the early nineties in Kyoto, Japan, by re- searchers in Department 2 of the now defunct ATR Inter- preting Telephony Labs (later Interpreting Telecommuni- cations Research Labs) and was announced in 1996 as ”a high-definition speech re-sequencing system” at the joint ASJ/ASA meeting in Hawaii [1], though the basic method was first reported in 1994 at the ESCA/IEEE Mohonk Speech Synthesis workshop [2]. The name was derived from ”Collected Hacks from ATR” and was first suggested by Paul Taylor who was then working on the intonation component. It was not the first concatenative speech syn- thesis system but it was the first to use raw waveform seg- ments directly, without recourse to any signal processing. This step not only greatly simplified the synthesis process but also allowed the use of very high quality recordings (some even in stereo) that exactly reproduced the voice quality and speaking style of the recorded subjects. It re- placed the buzzy artificial sound of parametric synthesis with surprisingly natural-sounding speech. It was suscep- tible to concatenation errors if the waveform coverage in the voice database was incomplete but in that period much progress was made using as little as one hour of recorded speech and the samples in the corpus are all produced from such small databases. In contrast, some commercial users of this system now employ corpora of well-over 100 hours of recordings. 2. The CHATR Corpus The CHATR Corpus has ISLRN 074-692-309-096-0 (reg- istered on Oct 16 2015) and is listed under the URL: http://netsoc.fastnet.ie/chatr/. There are copies under http://www.tcd-fastnet.com and http://www.speech-data.jp. The corpus is freely available as an Open Access resource (Gratis & Libre) for research use and as a historical archive under CC-BY (Attribution) licensing. Figure 1 shows the top page and CHATR logo. 3. Page Layouts This archive maintains the original linking structure …

Content truncated to 500 words