Degradation of speaker identification for LPC formant-coded speech
J.H. Eggen, Leo Llm Vogten · The Journal of the Acoustical Society of America · 1990
Degradation of speaker characteristics for LPC formant-coded speech was measured with a speaker identification experiment. Spontaneous speech spoken by 14 speakers was analyzed and synthesized with two LPC speech-coding schemes using 2.4 and 12 kbit/s, respectively. PCM-coded speech (120 kbit/s) served as reference speech. Here 20 subjects had to identify the speakers by listening to the utterances. Listeners' familiarity with speakers was scaled before the experiment. The percentage that correctly identified speakers differs signficantly for the three speech types and depends on familiarity. The degradation of speaker identification can also be expressed by the index of loss in voice recognition. This index is defined as the ratio of the percentage correctly identified LPC voices and the percentage correctly identified PCM voices. The index is a function of familiarity only for 2.4 kbit/s LPC speech. If familiarity is controlled, the speaker identification test provides a suitable method for evaluating speech-coding techniques with respect to voice recognition. [Work supported by the Dutch government as part of the national SPIN program “Analysis and synthesis of speech.”]