Tamil and English speech database for heartbeat estimation

The aim of this research work is to provide an open source database containing speech signals and the corresponding heartbeat rates, so as to further widen the area of research in speech signal processing, especially estimation of heartbeat rate from speech. Tamil and English Speech Database for Heartbeat Estimation consists of 10,040 speech recordings. The speech signals were recorded from 109 persons, 52 females and 57 males with an average age of 25 years and 6 months. The informed consented volunteers were asked to perform three tasks; like answering and reading in rest state; answering and reading after physical exercise and answering after watching video clips. 24-th and 72-nd order Mel-Frequency Cepstral Coefficients and 14-th and 52-nd order Auto Regressive Reflection Coefficients are extracted from the speech signal. Prediction of heartbeat is done by linear regression using support vector machine. The statistical significance of the heartbeat prediction results are improved by 10-fold speaker-independent cross validation scheme. Experimental results show a minimum average estimation error of ± 13.