#!/usr/bin/env bash
# transcribe-audio — Velma API wrapper for audio transcription
# Converts audio files to timestamped, speaker-labeled text with emotion data
#
# Usage: scripts/transcribe-audio <audio-file>
#        scripts/transcribe-audio --help
#
# Environment:
#   VELMA_API_KEY   — API key (or stored in project .mcp.json)
#   VELMA_API_URL   — Override API endpoint (default: https://api.modulate.ai/v1/transcribe)
#
# Output:
#   stdout: Timestamped speaker-labeled text ([MM:SS-MM:SS] Speaker_N: text)
#   stderr: Path to full JSON response file (for emotion/confidence parsing)

set -euo pipefail

VELMA_API_URL="${VELMA_API_URL:-https://api.modulate.ai/v1/transcribe}"
SUPPORTED_FORMATS="mp3|m4a|wav"

# --- Help ---
if [[ "${1:-}" == "--help" || "${1:-}" == "-h" || $# -eq 0 ]]; then
  cat <<'USAGE'
Usage: scripts/transcribe-audio <audio-file>

Transcribes audio using the Modulate Velma API with speaker diarization
and emotion detection.

Arguments:
  <audio-file>   Path to audio file (.mp3, .m4a, .wav)

Options:
  --help, -h     Show this help message

Environment Variables:
  VELMA_API_KEY   Velma API key (or configure via /mos:setup transcription)
  VELMA_API_URL   Override API endpoint (default: https://api.modulate.ai/v1/transcribe)

Output:
  stdout   Timestamped speaker-labeled text
           Format: [MM:SS-MM:SS] Speaker_N: text
  stderr   Path to JSON file with full response (emotions, confidence scores)

Examples:
  scripts/transcribe-audio recording.mp3
  scripts/transcribe-audio ~/meetings/standup.m4a
  VELMA_API_URL=http://localhost:8080/transcribe scripts/transcribe-audio test.wav
USAGE
  exit 0
fi

# --- Validate input file ---
AUDIO_FILE="$1"

if [[ ! -f "$AUDIO_FILE" ]]; then
  echo "Error: File not found: $AUDIO_FILE" >&2
  exit 1
fi

EXTENSION="${AUDIO_FILE##*.}"
EXTENSION=$(echo "$EXTENSION" | tr '[:upper:]' '[:lower:]')  # lowercase (macOS bash 3.2 compat)

if [[ ! "$EXTENSION" =~ ^($SUPPORTED_FORMATS)$ ]]; then
  echo "Error: Unsupported format '.$EXTENSION'. Supported: .mp3, .m4a, .wav" >&2
  exit 1
fi

# Check file size (warn above 500MB)
FILE_SIZE=$(stat -c%s "$AUDIO_FILE" 2>/dev/null || stat -f%z "$AUDIO_FILE" 2>/dev/null || echo 0)
if [[ "$FILE_SIZE" -gt 524288000 ]]; then
  echo "Warning: File is $(( FILE_SIZE / 1048576 ))MB. Large files may take longer to process." >&2
fi

# --- Resolve API key ---
API_KEY="${VELMA_API_KEY:-}"

if [[ -z "$API_KEY" ]]; then
  # Try reading from project .mcp.json
  MCP_JSON=""
  for candidate in ".mcp.json" "../.mcp.json"; do
    if [[ -f "$candidate" ]]; then
      MCP_JSON="$candidate"
      break
    fi
  done

  if [[ -n "$MCP_JSON" ]] && command -v jq &>/dev/null; then
    API_KEY=$(jq -r '.env.VELMA_API_KEY // .mcpServers.velma.env.VELMA_API_KEY // empty' "$MCP_JSON" 2>/dev/null || true)
  fi
fi

if [[ -z "$API_KEY" ]]; then
  echo "Error: Velma API key not configured. Run /mos:setup transcription or set VELMA_API_KEY" >&2
  exit 1
fi

# --- Call Velma API ---
RESPONSE_FILE=$(mktemp /tmp/velma-response-XXXXXX.json)

HTTP_CODE=$(curl -s -w "%{http_code}" -o "$RESPONSE_FILE" \
  -X POST "$VELMA_API_URL" \
  -H "Authorization: Bearer $API_KEY" \
  -H "Accept: application/json" \
  -F "audio=@$AUDIO_FILE" \
  -F "diarization=true" \
  -F "emotions=true" \
  --max-time 600 \
  2>/dev/null) || {
    echo "Error: Network failure connecting to Velma API at $VELMA_API_URL" >&2
    rm -f "$RESPONSE_FILE"
    exit 1
  }

if [[ "$HTTP_CODE" -lt 200 || "$HTTP_CODE" -ge 300 ]]; then
  ERROR_MSG=$(jq -r '.error // .message // "Unknown error"' "$RESPONSE_FILE" 2>/dev/null || cat "$RESPONSE_FILE")
  echo "Error: Velma API returned HTTP $HTTP_CODE: $ERROR_MSG" >&2
  rm -f "$RESPONSE_FILE"
  exit 1
fi

# --- Validate response has segments ---
if ! jq -e '.segments' "$RESPONSE_FILE" &>/dev/null; then
  echo "Error: Velma API response missing 'segments' field" >&2
  rm -f "$RESPONSE_FILE"
  exit 1
fi

# --- Parse and format output ---
# Format timestamps as MM:SS
format_timestamp() {
  local seconds="$1"
  local mins=$(echo "$seconds" | awk '{printf "%d", $1 / 60}')
  local secs=$(echo "$seconds $mins" | awk '{printf "%02d", $1 - ($2 * 60)}')
  printf "%02d:%s" "$mins" "$secs"
}

# Output timestamped speaker-labeled text to stdout
jq -r '.segments[] | "\(.start_time)|\(.end_time)|\(.speaker_id)|\(.text)"' "$RESPONSE_FILE" | \
while IFS='|' read -r start end speaker text; do
  start_fmt=$(format_timestamp "$start")
  end_fmt=$(format_timestamp "$end")
  # Normalize speaker_id: speaker_1 -> Speaker_1
  speaker_label=$(echo "$speaker" | sed 's/speaker_/Speaker_/')
  echo "[$start_fmt-$end_fmt] $speaker_label: $text"
done

# Output JSON path to stderr for downstream tools
echo "$RESPONSE_FILE" >&2