auto-report-skill/scripts/transcribe.sh

#!/bin/bash
# Transcribe audio recordings using local whisper server (with API key)
# Usage: ./transcribe.sh <meeting_dir> [<file.WAV> <output_name>]

set -euo pipefail

SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
: ${WHISPER_URL:?ERROR: WHISPER_URL not set (e.g., https://llm.lambda.coredump.ru/v1)}
: ${WHISPER_API_KEY:?ERROR: WHISPER_API_KEY not set}
MODEL="whisper-1"  # Изменено с конкретной модели на общее название
LANGUAGE="ru"

# Hotwords
HOTWORDS_FILE="${HOTWORKS_PATH:-$SCRIPT_DIR/hotwords.txt}"
HOTWORDS=""
if [ -f "$HOTWORDS_FILE" ]; then
    HOTWORDS=$(grep -v '^#' "$HOTWORDS_FILE" | grep -v '^$' | tr '\n' ',' | sed 's/,,*/,/g; s/^,//; s/,$//')
    echo "Loaded hotwords from $HOTWORDS_FILE"
fi

# ---------- argument parsing ----------
if [ $# -lt 1 ]; then
    echo "Usage: $0 <meeting_dir> [<file.WAV> <output_name>]"
    exit 1
fi

MEETING_DIR="$1"
WORK_DIR="$(cd "$SCRIPT_DIR/$MEETING_DIR" && pwd)"
OUTPUT_DIR="$WORK_DIR/transcription"
mkdir -p "$OUTPUT_DIR"

convert_to_mp3() {
    local output_mp3="$1"
    shift
    local inputs=("$@")
    if [ -f "$output_mp3" ]; then
        echo "  $output_mp3 already exists, skipping conversion"
        return
    fi
    if [ ${#inputs[@]} -eq 1 ]; then
        echo "  Converting ${inputs[0]} -> $output_mp3"
        ffmpeg -y -i "${inputs[0]}" -ac 1 -ar 16000 -b:a 64k "$output_mp3" 2>/dev/null
    else
        local listfile
        listfile=$(mktemp /tmp/ffmpeg_concat_XXXXXX.txt)
        for f in "${inputs[@]}"; do
            echo "file '$f'" >> "$listfile"
        done
        echo "  Concatenating ${#inputs[@]} files -> $output_mp3"
        ffmpeg -y -f concat -safe 0 -i "$listfile" -ac 1 -ar 16000 -b:a 64k "$output_mp3" 2>/dev/null
        rm -f "$listfile"
    fi
}

transcribe_file() {
    local mp3_file="$1"
    local name="$2"
    local json_file="$OUTPUT_DIR/${name}.json"

    if [ -f "$json_file" ]; then
        echo "  $name already transcribed, skipping"
        return
    fi

    echo "  Transcribing $name..."
    local started
    started=$(date +%s)

    # Form the full URL for transcription
    local full_url="${WHISPER_URL}/audio/transcriptions"

    local curl_args=(
        -s -w "%{http_code}" -o "$json_file"
        -X POST "$full_url"
        -H "Authorization: Bearer $WHISPER_API_KEY"
        -F "file=@${mp3_file}"
        -F "model=${MODEL}"
        -F "language=${LANGUAGE}"
        -F "response_format=verbose_json"
        -F "temperature=0.0"
        --max-time 3600
    )
    if [ -n "$HOTWORDS" ]; then
        curl_args+=(-F "hotwords=${HOTWORDS}")
    fi

    local http_code
    http_code=$(curl "${curl_args[@]}")
    local elapsed=$(( $(date +%s) - started ))

    if [ "$http_code" != "200" ]; then
        echo "    ERROR: HTTP $http_code"
        # Display error response body for debugging
        if [ -f "$json_file" ]; then
            cat "$json_file"
        fi
        rm -f "$json_file"
        return 1
    fi

    echo "    Done in ${elapsed}s"

    # Extract plain text and timestamped text
    python3 - "$json_file" "$OUTPUT_DIR" "$name" <<'PYEOF'
import json, sys, os

json_path = sys.argv[1]
output_dir = sys.argv[2]
name = sys.argv[3]

with open(json_path) as f:
    data = json.load(f)

segs = data.get("segments", [])

# Timestamped text
lines = []
for seg in segs:
    start = seg.get("start", 0)
    h, m, s = int(start // 3600), int((start % 3600) // 60), int(start % 60)
    lines.append(f"[{h:02d}:{m:02d}:{s:02d}] {seg['text'].strip()}")

txt_path = os.path.join(output_dir, f"{name}.txt")
with open(txt_path, "w") as f:
    f.write("\n".join(lines))

# Plain text
plain = " ".join(seg["text"].strip() for seg in segs)
plain_path = os.path.join(output_dir, f"{name}_plain.txt")
with open(plain_path, "w") as f:
    f.write(plain)

print(f"    {len(segs)} segments, {len(plain)} chars")
PYEOF
}

# ---------- manual mode (specific file) ----------
if [ $# -ge 3 ]; then
    WAV_FILE="$WORK_DIR/$2"
    NAME="$3"
    MP3_FILE="$OUTPUT_DIR/${NAME}.mp3"

    echo "=== Transcribing $2 as '$NAME' ==="
    convert_to_mp3 "$MP3_FILE" "$WAV_FILE"
    transcribe_file "$MP3_FILE" "$NAME"
    echo "=== Done ==="
    exit 0
fi

# ---------- auto mode ----------
echo "=== Auto-detecting audio sources in $WORK_DIR ==="

H2N_XY=$(find "$WORK_DIR" -maxdepth 1 -name "SR*XY.WAV" | head -1)
H2N_MS=$(find "$WORK_DIR" -maxdepth 1 -name "SR*MS.WAV" | head -1)
mapfile -t SARAMONIC_FILES < <(find "$WORK_DIR" -maxdepth 1 -name "*.WAV" ! -name "SR*" | sort)

SOURCES=()

if [ -n "$H2N_XY" ]; then
    echo "  Found H2n XY: $(basename "$H2N_XY")"
    SOURCES+=("h2n_xy:$H2N_XY")
fi
if [ -n "$H2N_MS" ]; then
    echo "  Found H2n MS: $(basename "$H2N_MS")"
    SOURCES+=("h2n_ms:$H2N_MS")
fi
if [ ${#SARAMONIC_FILES[@]} -gt 0 ]; then
    echo "  Found Saramonic files: ${SARAMONIC_FILES[*]##*/}"
    joined=$(printf "|%s" "${SARAMONIC_FILES[@]}")
    joined="${joined:1}"
    SOURCES+=("saramonic:$joined")
fi

if [ ${#SOURCES[@]} -eq 0 ]; then
    echo "Error: No WAV files found in $WORK_DIR"
    exit 1
fi

echo ""
echo "=== Step 1: Converting to mp3 ==="
for entry in "${SOURCES[@]}"; do
    name="${entry%%:*}"
    paths="${entry#*:}"
    mp3="$OUTPUT_DIR/${name}.mp3"

    IFS='|' read -ra files <<< "$paths"
    convert_to_mp3 "$mp3" "${files[@]}"
done

echo ""
echo "=== Step 2: Transcribing ==="
for entry in "${SOURCES[@]}"; do
    name="${entry%%:*}"
    mp3="$OUTPUT_DIR/${name}.mp3"
    transcribe_file "$mp3" "$name"
done

echo ""
echo "=== Done! ==="
echo "Results in: $OUTPUT_DIR/"