- need kaldi & montreal forced aligner
- should create folders in root directory:
org_audio
,segmented_audio
,train-result
- run
bash pipeline-train.sh
- the model will be zip in
model.zip
in the root directory - run
bash pipeline-test.sh
background music: piano music played by Tim Shevlyakov
rock music: Always there for you
country music: Mama tried
training set: tedlium s5_r3
note lexicon dictionary
comes from kaldi/egs/tedlium/s5_r3/data/local/lang_nosp/align_lexicon.txt
note kaldi-scp/*.scp
comes from kaldi/egs/tedlium/s5_r3/data/train
, kaldi/egs/tedlium/s5_r3/data/test
, and kaldi/egs/tedlium/s5_r3/data/dev
note text
comes from kaldi/egs/tedlium/s5_r3/data/train/text
, kaldi/egs/tedlium/s5_r3/data/dev/text
, and kaldi/egs/tedlium/s5_r3/data/test/text
note only partial data (under org_audio
, segmented_audio
, train-result
and text
) are uploaded, please use run kaldi/egs/tedlium/s5_r3/run.sh
to get the full dataset, and run prepare_convert_to_wav.py
to convert wav from sph files