MIMICz-An Audio Corpus for the Competency Evaluation of Voice Mimicking

The paper introduces a data corpus for voice mimicry analysis. The dataset comprises mimicry samples of 25 celebrities. Samples are recorded with professional mimicry artists in a studio environment. Moreover, the corpus is evaluated using a pilot study. The performance is evaluated using spectral and prosodic features. A DNN-based classifier is used for the scoring mechanism. A perception test initially identifies the best mimicking artist. Later, we investigate whether the DNN-based model predicts the same artist. When the model identifies the mean opinion score(MOS)-identified artist with the highest probability (rank-1), we assume that one hit occurs. The performance evaluation is carried out using top-X criteria. The experiment demonstrates the efficacy of the introduced corpus for further research in mimicking voice analysis.