First International Conference on Spoken Language Processing (ICSLP 90)

Kobe, Japan
November 18-22, 1990

Segmental Intelligibility of Synthetic and Natural Speech in Real and Nonsense Words

Rolf Carlson, Björn Granström, Lennart Nord

Department of Speech Communication and Music Acoustics, Royal Institute of Technology (KTH), Stockholm, Sweden

We have been using the preliminary version of the Esprit/SAM test procedure for synthetic speech to evaluate an experimental version of the multilingual text-to-speech system under development at our department. The proposed segmental test battery includes: a) hearing tests of the subjects, b) the familiarisation to the special type of speech synthesizer by an introductory paragraph, c) lists of CV, VC and VCV stimuli according to the phonotactic structure of the individual language. Tests on natural speech have also been performed forming a baseline for the synthesis evaluation and at the same time indicating the subjects' ability to give unambiguous orthographic response to nonsense words. An interesting question in this context is the phonemic awareness of the listeners. The Swedish fricative allophone set is a good example, where difficulties in labelling has to be studied carefully. Results will be presented at the meeting and compared to data reported earlier. We will also present data on the intelligibility of monosyllabic words drawn from the most frequent 10 000 words in Swedish.

Full Paper

Bibliographic reference.  Carlson, Rolf / Granström, Björn / Nord, Lennart (1990): "Segmental intelligibility of synthetic and natural speech in real and nonsense words", In ICSLP-1990, 989-992.