The first side of the Mitel tape consists of a number of tone and tone+noise based tests. The test suite synthesizes equivalent test data. Being digitally generated, this data is rather more predictable than the test data on the nasty old stretchy cassette tapes which Mitel sold.
The second side of the Mitel tape contains fragments of real speech from real phone calls captured from the North American telephone network. These are considered troublesome for DTMF detectors. A good detector is expected to achieve a reasonably low number of false detections on this data. Fresh clean copies of this seem to be unobtainable. However, Bellcore produce a much more aggressive set of three cassette tapes. All six side (about 30 minutes each) are filled with much tougher fragments of real speech from the North American telephone network. If you can do well in this test, nobody cares about your results against the Mitel test tape.
A fresh set of tapes was purchased for these tests, and digitised, producing 6 wave files of 16 bit signed PCM data, sampled at 8kHz. They were transcribed using a speed adjustable cassette player. The test tone at the start of the tapes is pretty accurate, and the new tapes should not have had much opportunity to stretch. It is believed these transcriptions are about as good as the source material permits.
PLEASE NOTE
These transcriptions may be freely used by anyone who has a legitimate copy of the original tapes. However, if you don't have a legitimate copy of those tapes, you also have no right to use this data. The original tapes are the copyright material of BellCore, and they charge over US$200 for a set. I doubt they sell enough copies to consider this much of a business. However, it is their data, and it is their right to do as they wish with it. Currently I see no indication they wish to give it away for free.
1.5.6