s452111-mlworkshops/metrics_service.sh

#!/bin/bash

# Prepare hypothesis file based on column 2
cut -f2 'wikiniews_results.tsv' > hypothesis.txt
# Prepare  reference file based on column 3
cut -f3 'wikiniews_results.tsv' > reference.txt

# awk transfer txt to .trn
awk 'BEGIN{FS=OFS="\t"}{print $0,"(sp1_"NR")"}' < reference.txt > reference.trn
awk 'BEGIN{FS=OFS="\t"}{print $0,"(sp1_"NR")"}' < hypothesis.txt > hypothesis.trn

# use sclite to calculate WER, actually we need only lines starting with 'Scores'
sclite -f 0 -r reference.trn trn -h hypothesis.trn trn -e utf-8 -i rm -o all stdout | grep "Scores:" > wer_results.txt

cat wer_results.txt | awk '{print ( ($7 + $8 + $9) / ($7 + $8 + $6) ) * 100;}' >> wer_per_line.txt
awk '{getline l < "wer_per_line.txt"; print $0"\t"l} ' wikiniews_results.tsv > wikinews_results.tsv

# calculate mean WER for all records
awk '{sum += $1; n++} END { print sum / n; }' < wer_per_line.txt >> wer.txt

# calculate SSR
awk '{if ($1 == 0) acc += 1;} END { print ( acc / NR) * 100; }' < wer_per_line.txt >> srr.txt

# trim the files to max N lines
tail -15 wer.txt
tail -15 srr.txt