How LLM Can Fix Your Posture

The author, a system engineer, developed a voice-to-text system to combat typing fatigue and improve productivity. He spends hours daily on his workstation, typing commands, prompts, and notes, which caused physical discomfort and created a bottleneck between thought and action. The solution involves an Android app sending recognized speech over local WiFi to a Python service on his workstation, which then types the text at the cursor's position. The key insight was the superior performance of Android's built-in speech recognition compared to other solutions like Whisper or Vosk. Android's native speech-to-text is fast, accurate, runs on device hardware, and handles language switching seamlessly. This system significantly increased his productivity, tripling the speed of tasks like writing prompts, commit messages, and documentation. Beyond productivity, the tool dramatically improved his physical well-being, allowing him to work standing for half the day, something uncomfortable with traditional typing. The Android app uses Kotlin and the SpeechRecognizer API, while the workstation service is a lightweight Python script using xdotool for keyboard input simulation. The system is entirely local, ensuring data privacy and low latency without internet dependency. While excellent for general text input, it doesn't work well for coding due to variable names and syntax, nor in noisy environments or for highly technical terms. However, with AI tools like Claude Code, the need to type code manually has also diminished for the author. This self-built tool, costing only a weekend of development, proved to be highly impactful, removing friction from daily tasks and improving both efficiency and physical health.