Top-Level Documentation for HUB-4 Mandarin Speech Data
------------------------------------------------------

This CD-ROM contains a portion of the acoustic data designated as the
training set for the 1997 DARPA HUB-4 Mandarin Benchmark.

All acoustic files contain a 1024-byte NIST SPHERE header, followed by
linear PCM sample data; the samples are 16-bit linear PCM with the
high-byte first, and all recordings were done using a single channel
and 16-KHz sample frequency.  The sample data are not compressed.

Most files contain approximately 30 minutes of recorded material, and
some contain either 60 minutes or 120 minutes (approximately) from
CC-TV, KAZN-AM, or Voice of America broadcasts; since the sampling
format requires roughly 2 megabytes (MB) per minute of recording, the
file sizes are typically around 60 MB, with some files ranging up to
120 or 240 MB.