Some comparisons between the articulation rates of LPC and diphone or monophone‐based synthesis by rules

In order to make a quick decision about the quality to be expected from a speech coding or synthesis system, a special articulation test procedure for the German language has been developed. The basic test elements consist of 100 VC or CV transitions which are spoken in a C‐VC‐V or V‐CV‐C environment. These elements were selected to contain all most probable transitions of the German language (about 57%) and to be representative of the probability of the 40 single sounds used. Bandlimited speech, LPC speech, as well as synthetic speech have been tested by this procedure. The synthesizers which have been used are a diphone‐based formant synthesizer (SAMT) developed at our laboratory and a German version of the VOTRAX synthesizer (VS‐6.G2). Articulation rates and confusion matrices give detailed information about the quality of the synthesis and coding systems for the German language. In comparison to similar measurements with the Japanese language, the articulation rate is strongly influenced by vowel confusions. This is due to the fact that there are more than 20 discernible vowels in the German language and only 5 in the Japanese language.