align_shiro_readme.txt

To create phonemap from abair xml:
cat ../abair-gitea/abair-corpora/cmg_ga_mu_corpusbeag/textproc_xml/*.xml | egrep 'symbol=".+"' | sed 's/<phoneme symbol="//g' | sed 's/"\/>//g' | sort | uniq > irish_test/gle_sampa.csv


The phonelist csv file has to be utf-8


To build model and align, with word and segments labs:


python align_shiro.py make-def irish_test/gle_sampa.csv llf_test/
python align_shiro.py make-index-from-xml llf_test/wav/ llf_test/textproc_xml/ llf_test/align
python align_shiro.py make-word-index-from-xml llf_test/wav/ llf_test/textproc_xml/ llf_test/align
python align_shiro.py feats llf_test/wav/ llf_test/align
python align_shiro.py train llf_test llf_test/align/feats llf_test/align/index.csv
python align_shiro.py align-feats llf_test llf_test/align/feats/ llf_test/align llf_test/align/index.csv
python align_shiro.py labs llf_test/align llf_test/lab
python align_shiro.py word-labs llf_test/align llf_test/wordlab




python3 align_shiro.py align-file aligner_models/nnc_en/ test_data/nnc_test/wav/nnc_arctic_0033.wav "hh iy ah n f ow l d ah d ah l ao ng t ay p r ih t ah n l eh t er sil ah n d hh ae n d ah d ih t t uw g r eh g s ah n" --sil

python3 align_shiro.py word-align-file aligner_models/nnc_en/ test_data/nnc_test/wav/nnc_arctic_0033.wav "He hh iy, unfolded ah n f ow l d ah d, a ah, long l ao ng, typewritten t ay p r ih t ah n, letter l eh t er, sil sil, and ah n d, handed hh ae n d ah d, it ih t, to t uw, Gregson g r eh g s ah n" --sil


python3 align_shiro.py align-file aligner_models/hb_sv/ test_data/jessica/wav/stts-001.wav "sil J A: S Ö: K k T t E0 SJ Y D d I: E N A F Ä3 R sil"
python3 align_shiro.py word-align-file aligner_models/hb_sv/ test_data/jessica/wav/stts-001.wav "jag J A:, sökte S Ö: K k T t E0, skydd SJ Y D d, i I:, en E N, affär A F Ä3 R" --sil


NOTE
There is a problem with the "ax" phoneme - seems to cause segmentation fault - yes and it doesn't exist in the nnc training data
So: Need to make sure that all phones in phonedef.json actually occur in training data!