Signal energy-based automatic speech splitter: a tool for developing speech corpus
Suyanto Suyanto · 2007
This paper describes a tool, called as automatic speech splitter (ASS) used in developing speech corpus. Two problems in developing a huge speech corpus are recording process and file renaming. Recording one by one sentence and storing the resulting speech in a file name are time consuming. Solutions to those problems are record n sentences simultaneously with a certain delay between two sentences, split the results into n speeches automatically using ASS, and then store the n speeches into n desired file names. Basic notion of ASS is classifying a speech input into silence and speech segments according to its short-term signal energy rate (SER). ASS successfully split a long speech containing n sentences into exactly n one-sentence-speeches with maximum error of 50 ms, but this error is acceptable since the tolerance is 100 ms.