Real-Time Duration Modification Using Frame Level Epoch Extraction

Knowledge of epochs is extensively used in duration modification. The objective of duration modification is to alter the rate of the speech signal while preserving the pitch period information. Methods such as pitch synchronous overlap and add (PSOLA), linear prediction PSOLA (LP-PSOLA), and epoch-based modification use the information of local pitch period or epochs. The PSOLA method performs manipulation on the speech signal directly, by concatenating the segments of appropriate glottal cycles. Whereas LP-PSOLA and epoch-based modification are performed on the LP residual. All the above methods are applied and evaluated on the entire utterance. There might be some implementation issues arise when they are adopted in real-time, such as the requirement of real-time epoch extraction, partial glottal cycles at the start and end of the frames, and concatenation of the adjacent frames. To overcome this, a real-time duration modification method is implemented using the knowledge of detected epoch locations at frame level. The duration modification is achieved by manipulating the glottal cycles as per the modification factor.