Voice dictation setup for Ubuntu 24.04 (GNOME Wayland): faster-whisper + ydotool + parecord. Hard-won fixes for ydotoold daemon and wl-copy paste method.

install-dictation.sh

#!/usr/bin/env bash
# =============================================================================
# Voice Dictation Setup for Ubuntu 24.04 (Wayland)
# Stack: faster-whisper + ydotool + parecord
# Toggle: Ctrl+Shift+Space (configurable)
# =============================================================================
set -euo pipefail

# --- Config ------------------------------------------------------------------
WHISPER_MODEL="${WHISPER_MODEL:-small}"        # tiny|base|small|medium|large-v3
HOTKEY="${HOTKEY:-<Ctrl><Shift>space}"         # GNOME keybinding format
DICTATION_DIR="$HOME/faster-whisper-dictation"
VENV_PYTHON="$DICTATION_DIR/venv/bin/python"
AUDIO_FILE="/tmp/dictation_recording.wav"
PID_FILE="/tmp/dictation.pid"

# Colors
RED='\033[0;31m'; GREEN='\033[0;32m'; YELLOW='\033[1;33m'; NC='\033[0m'
info()    { echo -e "${GREEN}[✓]${NC} $*"; }
warn()    { echo -e "${YELLOW}[!]${NC} $*"; }
error()   { echo -e "${RED}[✗]${NC} $*" >&2; }
section() { echo -e "\n${YELLOW}══ $* ══${NC}"; }

# --- Checks ------------------------------------------------------------------
section "Pre-flight checks"

if [[ "$EUID" -eq 0 ]]; then
    error "No ejecutar como root. Usa tu usuario normal."
    exit 1
fi

# Detect display server
SESSION_TYPE="${XDG_SESSION_TYPE:-unknown}"
info "Session type: $SESSION_TYPE"
if [[ "$SESSION_TYPE" == "x11" ]]; then
    warn "Estás en X11. ydotool funciona pero xdotool también sería opción."
fi

# RAM check (small model needs ~2GB)
TOTAL_RAM_MB=$(free -m | awk '/^Mem:/{print $2}')
info "RAM disponible: ${TOTAL_RAM_MB}MB"
if [[ "$TOTAL_RAM_MB" -lt 2048 ]] && [[ "$WHISPER_MODEL" == "small" ]]; then
    warn "RAM limitada. Considera WHISPER_MODEL=base para este sistema."
fi

# --- Step 1: System dependencies ---------------------------------------------
section "Step 1: System dependencies"

sudo apt-get update -qq
sudo apt-get install -y \
    portaudio19-dev \
    python3-venv \
    python3-pip \
    git \
    ydotool \
    ydotoold \
    wl-clipboard \
    pulseaudio-utils \
    libnotify-bin \
    cmake \
    libevdev-dev

info "Dependencias instaladas."

# --- Step 2: ydotool permissions ---------------------------------------------
section "Step 2: ydotool + uinput permissions"

# udev rule
if [[ ! -f /etc/udev/rules.d/60-uinput.rules ]]; then
    sudo tee /etc/udev/rules.d/60-uinput.rules > /dev/null << 'EOF'
KERNEL=="uinput", MODE="0660", GROUP="input"
EOF
    sudo udevadm control --reload-rules
    sudo udevadm trigger
    info "Regla udev creada."
else
    info "Regla udev ya existe."
fi

# Add user to input group
if ! groups | grep -qw input; then
    sudo usermod -aG input "$USER"
    warn "Usuario añadido al grupo 'input'. NECESITAS CERRAR SESIÓN Y VOLVER A ENTRAR antes de continuar."
    warn "Después del re-login, ejecuta este script de nuevo."
    exit 0
else
    info "Usuario ya está en el grupo 'input'."
fi

# ydotoold systemd user service
mkdir -p ~/.config/systemd/user
cat > ~/.config/systemd/user/ydotoold.service << 'EOF'
[Unit]
Description=ydotoold - ydotool daemon
After=graphical-session.target

[Service]
ExecStart=/usr/bin/ydotoold
Restart=on-failure
RestartSec=3

[Install]
WantedBy=default.target
EOF

systemctl --user daemon-reload
systemctl --user enable --now ydotoold.service 2>/dev/null || true

if systemctl --user is-active --quiet ydotoold.service; then
    info "ydotoold daemon activo."
else
    warn "ydotoold no pudo iniciarse (puede ser normal sin sesión gráfica activa)."
fi

# --- Step 3: Python venv + faster-whisper ------------------------------------
section "Step 3: faster-whisper (Python venv)"

if [[ ! -d "$DICTATION_DIR" ]]; then
    mkdir -p "$DICTATION_DIR"
fi

if [[ ! -d "$DICTATION_DIR/venv" ]]; then
    python3 -m venv "$DICTATION_DIR/venv"
    info "Venv creado en $DICTATION_DIR/venv"
fi

"$DICTATION_DIR/venv/bin/pip" install --quiet --upgrade pip
"$DICTATION_DIR/venv/bin/pip" install --quiet \
    faster-whisper \
    pyaudio \
    sounddevice \
    numpy

info "faster-whisper instalado."

# Pre-download model (evita delay en primer uso)
section "Step 4: Pre-descarga del modelo Whisper ($WHISPER_MODEL)"
info "Descargando modelo '$WHISPER_MODEL'... (puede tardar varios minutos)"

"$VENV_PYTHON" - << PYTHON
from faster_whisper import WhisperModel
import sys
try:
    print(f"  Descargando modelo '{sys.argv[1] if len(sys.argv) > 1 else 'small'}'...")
    model = WhisperModel("$WHISPER_MODEL", device="cpu", compute_type="int8")
    print("  Modelo descargado correctamente.")
except Exception as e:
    print(f"  Error: {e}", file=sys.stderr)
    sys.exit(1)
PYTHON

info "Modelo listo."

# --- Step 5: Dictation scripts -----------------------------------------------
section "Step 5: Scripts de dictado"

mkdir -p ~/.local/bin

# dictate-start
cat > ~/.local/bin/dictate-start << SCRIPT
#!/usr/bin/env bash
AUDIO_FILE="$AUDIO_FILE"
PID_FILE="$PID_FILE"

if [ -f "\$PID_FILE" ]; then
    notify-send -i microphone-sensitivity-high "Dictation" "Ya grabando... presiona de nuevo para parar"
    exit 0
fi

notify-send -i microphone-sensitivity-high "Dictation" "Grabando... presiona hotkey para parar"

# Graba en WAV mono 16kHz (formato óptimo para Whisper)
parecord --channels=1 --rate=16000 --format=s16le "\$AUDIO_FILE" &
echo \$! > "\$PID_FILE"
SCRIPT
chmod +x ~/.local/bin/dictate-start

# dictate-stop
cat > ~/.local/bin/dictate-stop << SCRIPT
#!/usr/bin/env bash
VENV_PYTHON="$VENV_PYTHON"
AUDIO_FILE="$AUDIO_FILE"
PID_FILE="$PID_FILE"
WHISPER_MODEL="$WHISPER_MODEL"

if [ ! -f "\$PID_FILE" ]; then
    notify-send "Dictation" "No está grabando"
    exit 0
fi

# Detener grabación
kill "\$(cat "\$PID_FILE")" 2>/dev/null
rm -f "\$PID_FILE"
sleep 0.3

notify-send -i emblem-synchronizing "Dictation" "Transcribiendo..."

TEXT=\$("\$VENV_PYTHON" << 'PYTHON'
import sys
from faster_whisper import WhisperModel

model = WhisperModel("$WHISPER_MODEL", device="cpu", compute_type="int8")
segments, info = model.transcribe(
    "$AUDIO_FILE",
    beam_size=5,
    language=None,    # auto-detect; pon "es" o "en" para forzar idioma
    vad_filter=True,  # filtra silencios automáticamente
)
text = " ".join(seg.text.strip() for seg in segments)
print(text)
PYTHON
)

if [ -n "\$TEXT" ]; then
    # Pequeño delay para que el foco vuelva a la ventana objetivo
    sleep 0.2
    # ydotool type is unreliable on GNOME Wayland regardless of --key-delay value
    # Numeric keycodes (29:1 47:1...) don't work either
    # Correct method: wl-copy to clipboard + ydotool key ctrl+v (symbolic names required)
    echo -n "\$TEXT" | wl-copy && sleep 0.1 && ydotool key ctrl+v
    notify-send -i emblem-default "Dictation" "✓ \$(echo "\$TEXT" | cut -c1-60)..."
else
    notify-send -i dialog-warning "Dictation" "Sin voz detectada"
fi

rm -f "\$AUDIO_FILE"
SCRIPT
chmod +x ~/.local/bin/dictate-stop

# dictate-toggle (el que se bindea al hotkey)
cat > ~/.local/bin/dictate-toggle << SCRIPT
#!/usr/bin/env bash
PID_FILE="$PID_FILE"

if [ -f "\$PID_FILE" ]; then
    exec ~/.local/bin/dictate-stop
else
    exec ~/.local/bin/dictate-start
fi
SCRIPT
chmod +x ~/.local/bin/dictate-toggle

info "Scripts creados en ~/.local/bin/"

# --- Step 6: GNOME keyboard shortcut -----------------------------------------
section "Step 6: Hotkey GNOME ($HOTKEY)"

# Leer keybindings existentes para no pisarlos
EXISTING=$(gsettings get org.gnome.settings-daemon.plugins.media-keys custom-keybindings 2>/dev/null || echo "@as []")

# Determinar el próximo índice disponible
NEXT_IDX=0
while echo "$EXISTING" | grep -q "custom${NEXT_IDX}"; do
    NEXT_IDX=$((NEXT_IDX + 1))
done

CUSTOM_PATH="/org/gnome/settings-daemon/plugins/media-keys/custom-keybindings/custom${NEXT_IDX}/"

# Construir nueva lista de keybindings
if [[ "$EXISTING" == "@as []" ]] || [[ "$EXISTING" == "[]" ]]; then
    NEW_LIST="['${CUSTOM_PATH}']"
else
    # Insertar al final de la lista existente
    NEW_LIST=$(echo "$EXISTING" | sed "s|]$|, '${CUSTOM_PATH}']|")
fi

gsettings set org.gnome.settings-daemon.plugins.media-keys custom-keybindings "$NEW_LIST"
gsettings set "org.gnome.settings-daemon.plugins.media-keys.custom-keybinding:${CUSTOM_PATH}" name "Voice Dictation Toggle"
gsettings set "org.gnome.settings-daemon.plugins.media-keys.custom-keybinding:${CUSTOM_PATH}" command "${HOME}/.local/bin/dictate-toggle"
gsettings set "org.gnome.settings-daemon.plugins.media-keys.custom-keybinding:${CUSTOM_PATH}" binding "${HOTKEY}"

info "Hotkey registrado: $HOTKEY → dictate-toggle"

# Desactivar conflicto IBus (Ctrl+Space es el más común)
if echo "$HOTKEY" | grep -qi "ctrl.*space"; then
    warn "Detectado Ctrl+Space: desactivando shortcut IBus para evitar conflicto..."
    gsettings set org.freedesktop.ibus.general.hotkey triggers "[]" 2>/dev/null || true
fi

# --- Step 7: Smoke test -------------------------------------------------------
section "Step 7: Smoke test"

info "Probando ydotool..."
if ydotool type -- "" 2>/dev/null; then
    info "ydotool OK"
else
    warn "ydotool tuvo un warning (normal si ydotoold no está activo aún)"
fi

info "Probando parecord..."
if command -v parecord &>/dev/null; then
    info "parecord disponible"
else
    error "parecord no encontrado. Instala pulseaudio-utils."
fi

# --- Summary -----------------------------------------------------------------
section "Instalación completa"

cat << EOF

  Uso:
    $HOTKEY   → Empieza a grabar
    $HOTKEY   → Para y transcribe (texto aparece donde está el cursor)

  Scripts:
    ~/.local/bin/dictate-start   # start recording
    ~/.local/bin/dictate-stop    # stop + transcribe + type
    ~/.local/bin/dictate-toggle  # toggle (el que usa el hotkey)

  Modelo Whisper: $WHISPER_MODEL  (cambia con WHISPER_MODEL=medium ./install.sh)
  Idioma: auto-detect  (edita dictate-stop para forzar: language="es")

  Logs del daemon:
    systemctl --user status ydotoold
    journalctl --user -u ydotoold -f

  Troubleshooting:
    - Si el texto no aparece: verifica que estás en el grupo 'input' (groups | grep input)
    - Si la primera transcripción es lenta: el modelo se carga en frío; las siguientes son más rápidas
    - Para GNOME Wayland con IBus activo: Settings → Keyboard → elimina shortcuts de Ctrl+Space

EOF

info "Done. Prueba hablando con $HOTKEY en cualquier ventana."