These comments here refer to version 1.2 of the MOCHA-TIMIT database.
Descriptions of the changes made to the labels published in that release are given in two spreadsheets: notes for msak0 and fsew0.
There were 8 sentences with incorrect text for each subject, which were given phonetic transcriptions derived from the corrected text with manual alignment.
One sentence held corrupted EMA data and was discarded.
There were some cases where the database's automatic alignment had failed and others where the dictionary transcripts did not match the utterances.
These were identified and corrected manually, with particular attention to alveolar obstruents, /t/, /d/ and /n/, which suffered high levels of elision and deletion.
The revised transcriptions are distributed as compressed tar files:
label files for msak0 and fsew0.
The readme file provides further help and documentation.