Experiments on Landmark Detection

Concatenative speech synthesis depends on accurate segmentation of the pho-nemes in a training corpus. Segmentation is typically performed using HMMs, but recent speech recognition work suggests that the transient acoustic features charac-teristic of manner-class phoneme bounda-ries (landmarks) may be more precisely localized using acoustic classifiers spe-cifically designed for the task of landmark detection. This paper makes an empirical exploration of several features and tech-niques that are capable of improving the performance of the present HMM tools like HTK. On a standard benchmark data set, we achieve new state-of-the-art per-formance, reducing error in marking the landmarks from an average deviation of 28 milliseconds to 15 milliseconds from the manual ones.