This paper proposes a multilingual set of speech samples serving as the data base for standardizing an artificial voice which will be used to evaluate the talker dependency of digital coding algorithms. To investigate the impartiality of this data base, the fundamental statistical speech characteristics of these samples-long-term average spectrum, instantaneous amplitude distribution, segmental power, fundamental frequency, and voiced/ unvoiced ratio-are analyzed. Detailed dispersions in the samples as well as average values are derived. Taking the dispersions into account, the results agree with previous research. Speech quality dependency on talker and language is investigated when this set of speech samples is coded using three typical digital coding algorithms: PCM, ADPCM, and APC-AB. This set of speech samples reduces the evaluation bias caused by small speech-sample sizes. These multilingual speech samples are stored on CD-ROM and are publicly available.