Townshend Computer Tools DAT-Link - Speech Segmentation

6.15 Speech Segmentation 69
Custom versions of the narecord program can be created in a manner similar to
that describ ed in Section 5.14 for the naplay program.
6.15 Sp eech Segmentation
When the DAT-Link is used to record or transfer sp eech speech signals, it is often
desirable to segment the speechinto words or segments. A front-end program for
the narecord program called narecseg provides this function.
As the speech is acquired by the DAT-Link 's signal pro cessor, the average energy
of each blo ck of samples is also computed and passed to the narecseg program.
These energy computations allow narecseg to heuristically determine where each
speech segment begins and ends. Then, along with a le containing the acquired
data, a le containing p ointers to each segment is written by narecseg . The format
of the segmentation le is described in Appendix B.2.
The narecseg program accepts the same -u , -a , -i , -o , and -s options that are
accepted by narecord . In addition, the -n can be used to sp ecify how many seg-
ments should be recorded before stopping. By default, one segment is recorded.
Note that the audio data is always stored using the raw le format. Other formats
are not currently supp orted by narecseg .