Automatic Speech Recognition
audio
vosk
kaldi
nextcloud
kyteinsky's picture
initial commit
22028d0 verified
Japanese small model for Vosk
WER / CER
%WER 12.68 [ 8861 / 69878, 744 ins, 1581 del, 6536 sub ] exp/chain_a/tdnn/decode_test_csj_look_fast/wer_10_0.0
%WER 20.84 [ 23841 / 114379, 1803 ins, 9373 del, 12665 sub ] exp/chain_a/tdnn/decode_test_tedjp_look_fast/wer_9_0.0
%CER 9.52 [ 11016 / 115745, 1707 ins, 3026 del, 6283 sub ] exp/chain_a/tdnn/decode_test_csj_look_fast/cer_10_0.0
%CER 17.07 [ 32724 / 191731, 3502 ins, 14728 del, 14494 sub ] exp/chain_a/tdnn/decode_test_tedjp_look_fast/cer_8_0.0