the-only-working-whisper-typer-ubuntu

bash

#!/bin/bash
# Voice dictation: run once to start recording, run again to stop + transcribe.
# Text goes to clipboard — paste with Ctrl+V wherever you want.
# Usage: bind to a hotkey in system settings

PID_FILE="/tmp/dictate.pid"
AUDIO_FILE="/tmp/dictate.wav"
API_KEY="$(cat ~/.config/openai-api-key 2>/dev/null)"
PROXY="http://localhost:8888"

if [ -f "$PID_FILE" ]; then
    # Stop recording
    kill "$(cat "$PID_FILE")" 2>/dev/null
    rm -f "$PID_FILE"
    notify-send -t 1000 "🎤" "Transcribing..."

    # Send to OpenAI Whisper API
    RESULT=$(curl -s --proxy "$PROXY" \
        -H "Authorization: Bearer $API_KEY" \
        -F "file=@${AUDIO_FILE}" \
        -F "model=whisper-1" \
        -F "language=ru" \
        "https://api.openai.com/v1/audio/transcriptions" | jq -r '.text // empty')

    rm -f "$AUDIO_FILE"

    if [ -n "$RESULT" ]; then
        echo -n "$RESULT" | nohup xclip -selection clipboard &>/dev/null &
        notify-send -t 2000 "🎤 Ctrl+V to paste" "$RESULT"
    else
        notify-send -t 2000 "🎤" "Nothing recognized"
    fi
else
    # Start recording
    parecord --format=s16le --rate=16000 --channels=1 "$AUDIO_FILE" &
    echo $! > "$PID_FILE"
    notify-send -t 1000 "🎤" "Recording..."
fi