#!/usr/bin/env bash
# Regression test for the whisper service: runs every fixture under
# tests/fixtures/dictations/ through /asr twice (with and without hotwords)
# and reports word error rate vs the golden `.expected.txt`.
#
# Usage:
#   ./scripts/regress_whisper.sh
#   WHISPER_URL=http://minerva.lan:9001 ./scripts/regress_whisper.sh
#   ./scripts/regress_whisper.sh --verbose   # also print each hypothesis
#
# Exit code is the sum of mismatches across all hotwords runs (0 = perfect).
set -euo pipefail

WHISPER_URL="${WHISPER_URL:-http://minerva.lan:9001}"
REPO_ROOT="$(cd "$(dirname "$0")/.." && pwd)"
DIR="$REPO_ROOT/tests/fixtures/dictations"
VERBOSE=0
[[ "${1:-}" == "--verbose" || "${1:-}" == "-v" ]] && VERBOSE=1

die() { echo "error: $*" >&2; exit 1; }
[ -d "$DIR" ] || die "fixtures dir missing: $DIR"
command -v python3 >/dev/null || die "python3 not found"

# POST a single fixture. If hotwords arg is non-empty, forward it as a form
# field so the whisper service can feed it into faster_whisper.
run_once() {
    local m4a="$1" hotwords="$2"
    if [ -n "$hotwords" ]; then
        curl -sS --max-time 120 -X POST "$WHISPER_URL/asr?output=txt&language=de" \
            -F "audio_file=@$m4a" \
            -F "hotwords=$hotwords"
    else
        curl -sS --max-time 120 -X POST "$WHISPER_URL/asr?output=txt&language=de" \
            -F "audio_file=@$m4a"
    fi
}

# Word error rate via Levenshtein. Prints "errors/ref_len".
wer() {
    python3 - "$1" "$2" <<'PY'
import re, sys, unicodedata

def norm(path):
    with open(path, encoding="utf-8") as f:
        s = f.read()
    s = unicodedata.normalize("NFC", s).lower()
    s = re.sub(r"[^\w\s]", " ", s, flags=re.UNICODE)
    return s.split()

def lev(a, b):
    m, n = len(a), len(b)
    if m == 0 or n == 0: return max(m, n)
    prev = list(range(n + 1))
    for i, ca in enumerate(a, 1):
        cur = [i] + [0] * n
        for j, cb in enumerate(b, 1):
            cost = 0 if ca == cb else 1
            cur[j] = min(cur[j - 1] + 1, prev[j] + 1, prev[j - 1] + cost)
        prev = cur
    return prev[n]

ref, hyp = norm(sys.argv[1]), norm(sys.argv[2])
print(f"{lev(ref, hyp)}/{len(ref)}")
PY
}

printf "%-38s %-13s %-13s\n" "fixture" "no_hotwords" "with_hotwords"
printf "%-38s %-13s %-13s\n" "--------------------------------------" "-------------" "-------------"

total_hw_errs=0
for m4a in "$DIR"/*.m4a; do
    name="$(basename "$m4a" .m4a)"
    exp="$DIR/$name.expected.txt"
    hw_file="$DIR/$name.hotwords.txt"
    [ -f "$exp" ] || { echo "skip $name (no .expected.txt)"; continue; }

    tmp_no="$(mktemp)"
    tmp_hw="$(mktemp)"
    trap 'rm -f "$tmp_no" "$tmp_hw"' EXIT

    run_once "$m4a" "" > "$tmp_no"
    wer_no="$(wer "$exp" "$tmp_no")"

    if [ -f "$hw_file" ]; then
        hw="$(tr -d '\n' < "$hw_file")"
        run_once "$m4a" "$hw" > "$tmp_hw"
        wer_hw="$(wer "$exp" "$tmp_hw")"
        total_hw_errs=$(( total_hw_errs + ${wer_hw%/*} ))
    else
        wer_hw="—"
        total_hw_errs=$(( total_hw_errs + ${wer_no%/*} ))
    fi

    printf "%-38s %-13s %-13s\n" "$name" "$wer_no" "$wer_hw"

    if [ "$VERBOSE" -eq 1 ]; then
        echo "  expected:      $(tr -d '\n' < "$exp")"
        echo "  no_hotwords:   $(tr -d '\n' < "$tmp_no")"
        if [ -f "$hw_file" ]; then
            echo "  with_hotwords: $(tr -d '\n' < "$tmp_hw")"
        fi
    fi

    rm -f "$tmp_no" "$tmp_hw"
done

echo
echo "Total hotwords-run word errors: $total_hw_errs"
exit "$total_hw_errs"