Linux transcription (x11 & wayland) with whisper.cpp and built-in VAD. Records microphone and desktop, and transcribes existing .wav files with arguments.

transcribeee.sh

#!/usr/bin/env bash

set -euo pipefail

# CONFIG
DIR="${HOME}/.transcription"
mkdir -p "$DIR"

WHISPER_BIN="/home/user/source/git/whisper_gpu.cpp/bin/whisper-cli"
WHISPER_MODEL="/home/user/source/git/whisper_gpu.cpp/models/ggml-large-v3.bin"
VAD_MODEL="/home/user/source/git/whisper_gpu.cpp/models/ggml-silero-v5.1.2.bin"

# dependencies
for cmd in ffmpeg ffprobe pactl wpctl notify-send xsel; do
  command -v "$cmd" >/dev/null 2>&1 || { echo "Missing dependency: $cmd" >&2; exit 127; }
done
[ -x "$WHISPER_BIN" ] || { echo "Missing whisper binary: $WHISPER_BIN" >&2; exit 127; }
[ -f "$WHISPER_MODEL" ] || { echo "Missing model file: $WHISPER_MODEL" >&2; exit 127; }

notify() {
  notify-send "$MOTIVE" "$*" 2>/dev/null || echo "[$MOTIVE] $*"
}

transcribe() {
  local file="$1"
  [[ -f "$file" ]] || { notify "No file to transcribe: $file"; return; }

  notify "Transcribing $(basename "$file")"

  local start_time end_time duration audio_len
  start_time=$(date +%s)

  # Get duration in seconds (integer)
  audio_len=$(ffprobe -v error -show_entries format=duration \
    -of default=nw=1:nk=1 "$file" | awk '{printf "%d\n",$1}')

  local out_srt

  if (( audio_len < 20 )); then
    # short: no VAD
    out_srt=$("$WHISPER_BIN" \
      -m "$WHISPER_MODEL" \
      -f "$file"
    )
  else
    # long: VAD
    out_srt=$("$WHISPER_BIN" \
      -m "$WHISPER_MODEL" \
      -f "$file" \
      --vad \
      --vad-model "$VAD_MODEL" \
      --vad-threshold 0.6 \
      --vad-min-silence-duration-ms 200
    )
  fi

  end_time=$(date +%s)
  duration=$((end_time - start_time))

  # remove whisper header
  out_srt=$(printf '%s\n' "$out_srt" | sed '1d')

  # strip timestamps → TXT
  out_txt=$(printf '%s\n' "$out_srt" \
    | perl -pe 's/^\[\d{2}:\d{2}:\d{2}\.\d{3}\s+-->\s+\d{2}:\d{2}:\d{2}\.\d{3}\]\s*//')

  local srt_file="${file%.wav}.srt"
  local txt_file="${file%.wav}.txt"

  printf '%s\n' "$out_srt" > "$srt_file"
  printf '%s\n' "$out_txt" > "$txt_file"

  # clipboard
  if command -v xsel >/dev/null 2>&1; then
    printf '%s' "$out_txt" | xsel --clipboard --input
  elif command -v xclip >/dev/null 2>&1; then
    printf '%s' "$out_txt" | xclip -selection clipboard
  fi

  notify "Saved transcription: $(basename "$txt_file") (${duration}s)"
}

# --- MODE ---

if [[ "${1:-}" == "--retranscribe" ]]; then
  MOTIVE="Transcriber"
  shift
  if [[ "$#" -lt 1 ]]; then
    echo "Usage: $0 --retranscribe file1.wav [file2.wav ...]"
    exit 2
  fi

  for f in "$@"; do
    transcribe "$f"
  done

  exit 0
fi

# ORIGINAL RECORDING MODES
mode="${1:-dt}"

case "$mode" in
  dt|me|both) ;;
  *) echo "Usage: $0 [dt|me|both]  OR  $0 --retranscribe file.wav [...]"; exit 2;;
esac
MOTIVE="Recorder"

MIC_SRC="$(pactl get-default-source)"

# Fallback logic: wpctl if available, else pactl
if command -v wpctl >/dev/null 2>&1; then
  SINK="$(wpctl status | awk '/Default Configured Devices:/{f=1;next} f&&/Audio\/Sink/{print $NF; exit}')"
else
  SINK="$(pactl get-default-sink)"
fi
DESK_MON="${SINK:+$SINK.monitor}"

ts=$(date '+%F_%H-%M-%S')
wav_dt="${DIR}/${ts}-dt.wav"
wav_me="${DIR}/${ts}-me.wav"

echo "Mode: $mode — recording; PID will follow"
case "$mode" in
  dt)
    [[ -n "$DESK_MON" ]] || { notify "No desktop sink"; exit 1; }
    notify "Recording desktop audio" "with $DESK_MON"
    ffmpeg -hide_banner -loglevel error -f pulse -i "$DESK_MON" \
      -ar 16000 -ac 1 -c:a pcm_s16le "$wav_dt" &
    pid=$!
    ;;
  me)
    [[ -n "$MIC_SRC" ]] || { notify "No mic source"; exit 1; }
    notify "Recording microphone" "with $MIC_SRC"
    ffmpeg -hide_banner -loglevel error -f pulse -i "$MIC_SRC" \
      -ar 16000 -ac 1 -c:a pcm_s16le "$wav_me" &
    pid=$!
    ;;
  both)
    [[ -n "$DESK_MON" && -n "$MIC_SRC" ]] || { notify "Missing devices"; exit 1; }
    notify "Recording desktop and microphone" "with $DESK_MON and $MIC_SRC"
    ffmpeg -hide_banner -loglevel error \
      -f pulse -i "$DESK_MON" -f pulse -i "$MIC_SRC" \
      -map 0:a -ar 16000 -ac 1 -c:a pcm_s16le "$wav_dt" \
      -map 1:a -ar 16000 -ac 1 -c:a pcm_s16le "$wav_me" &
    pid=$!
    ;;
esac

echo "$pid"
wait "$pid" || true

# --- TRANSCRIBE ---
case "$mode" in
  dt) [[ -s "$wav_dt" ]] && transcribe "$wav_dt" || notify "No desktop recording";;
  me) [[ -s "$wav_me" ]] && transcribe "$wav_me" || notify "No mic recording";;
  both)
    [[ -s "$wav_dt" ]] && transcribe "$wav_dt"
    [[ -s "$wav_me" ]] && transcribe "$wav_me"
    ;;
esac

exit 0