New speech recognition models PLPL-v2.1,
WER (Word Error Rate) = 14%
TOTAL Percent correct = 85.74% WER (Word Error Rate) = 16.01% Accuracy = 83.99%.
The accuracy mentioned above can be obtained for recordings having the quality of at least:
16kHz / 16bit mono, wav format, noise smaller than -40dB,
when the vocabulary used in the recordings is similar to the one found in TV and radio programmes.