transcribe_audio.sh — Pre-Context

#!/usr/bin/env bash
set -euo pipefail

# ------------------------------------------------------------
# transcribe-audio.sh
# Post-session transcription on Mac using whisper-cli
# Usage: transcribe-audio.sh [--de|--en|--auto] <audio.wav>
# ------------------------------------------------------------

WHISPER="$HOME/Transkriptionen/whisper.cpp/build/bin/whisper-cli"
MODEL="$HOME/Transkriptionen/whisper.cpp/models/ggml-large-v3-turbo.bin"
VAD_MODEL="$HOME/Transkriptionen/whisper.cpp/models/ggml-silero-v6.2.0.bin"

# ------------------------------------------------------------
# Argument parsing
# ------------------------------------------------------------

LANG_MODE="en"
AUDIO=""

while [[ $# -gt 0 ]]; do
  case "$1" in
    --de)   LANG_MODE="de"; shift ;;
    --en)   LANG_MODE="en"; shift ;;
    --auto) LANG_MODE="auto"; shift ;;
    -*)
      echo "Usage: transcribe-audio [--de|--en|--auto] <audio.wav>" >&2
      exit 2
      ;;
    *)
      AUDIO="$1"; shift ;;
  esac
done

if [[ -z "$AUDIO" ]]; then
  echo "❌ No audio file specified." >&2
  echo "   Usage: transcribe-audio [--de|--en|--auto] <audio.wav>" >&2
  exit 1
fi

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

# ------------------------------------------------------------
# Language options
# ------------------------------------------------------------

case "$LANG_MODE" in
  en|de) LANG_OPTS=(-l "$LANG_MODE") ;;
  auto)  LANG_OPTS=() ;;
esac

# ------------------------------------------------------------
# Output path: same dir as audio, .txt extension
# ------------------------------------------------------------

AUDIO_DIR="$(dirname "$AUDIO")"
AUDIO_BASE="$(basename "$AUDIO" .wav)"
TRANSCRIPT_BASE="$AUDIO_DIR/${AUDIO_BASE}_transcript"

echo "🎧 Audio:      $AUDIO"
echo "🗣️  Language:   $LANG_MODE"
echo "📝 Transcript: ${TRANSCRIPT_BASE}.txt"
echo "----"

# ------------------------------------------------------------
# Run whisper-cli
# ------------------------------------------------------------
  auto)  LANG_OPTS=() ;;
esac

# ------------------------------------------------------------
# Output path: same dir as audio, .txt extension
# ------------------------------------------------------------

AUDIO_DIR="$(dirname "$AUDIO")"
AUDIO_BASE="$(basename "$AUDIO" .wav)"
TRANSCRIPT_BASE="$AUDIO_DIR/${AUDIO_BASE}_transcript"

echo "🎧 Audio:      $AUDIO"
echo "🗣️  Language:   $LANG_MODE"
echo "📝 Transcript: ${TRANSCRIPT_BASE}.txt"
echo "----"

# ------------------------------------------------------------
# Run whisper-cli
# ------------------------------------------------------------

"$WHISPER" \
  -m "$MODEL" \
  "${LANG_OPTS[@]}" \
  --vad \
  -vm "$VAD_MODEL" \
  --output-txt \
  --output-srt \
  -of "$TRANSCRIPT_BASE" \
  -f "$AUDIO"


# Strip ANSI escape sequences and carriage returns from txt output
sed -i '' 's/\x1b\[[0-9;]*[mGKH]//g; s/\r//g' "${TRANSCRIPT_BASE}.txt"

echo "✅ Done: ${TRANSCRIPT_BASE}.txt"