METHOD OF COMBINING CORPORA TO ACHIEVE CONSISTENCY IN PHONETIC LABELING

The present invention is a method of combining corpora to achieve consistency in phonetic labeling. Corpora are received. A first corpus is selected from the corpora. Generating a phonetic transcript if the first corpus does not include one. A second corpus is selected from the corpora. Generating a phonetic transcript if the second corpus does not include one. Each allophone in the second corpus is identified. At least one allophone is identified for each phone in the second corpus. For each phone in the second corpus, the allophone to which it most closely matches is identified. Each phone symbol in the phone transcript of the second corpus is replaced with a symbol for the corresponding identified allophone. The first corpus and second corpus are combined, including their phonetic transcripts, and designated as the first corpus. If there is another corpus in the corpora to be processed return to the step of selecting another second corpus.