Speech/music discrimination by detection: Assessment of time series events using ROC graphs

This paper suggests the application of the Receiver Operating Characteristics (ROC) graph to assess the performance of any speech/music discrimination method. ROC graphs are applied in the field of speech/music discrimination to assess the Time Series Events (TSE) method. The discrimination problem is viewed as two detection problems: detection of speech and detection of music. It was found that the optimal feature for detecting speech was silence with a true positive rate of 0.9 and false positive rate of 0.14, whilst the optimal feature for music was non-zero crossing rate NZCR with a true positive rate of 0.71 and false positive rate of 0.08.