Evaluation of speaker-independent phoneme recognition on TIMIT database using TDNNs