For developers who hate the idea of their private thoughts being processed by third-party servers, Inaudio offers a compelling alternative. The newly released desktop application performs all speech-to-text and text-to-speech operations locally, eliminating the need for cloud subscriptions or data transmission. As noted by creator shikki841 on DEV.to, the primary motivation was frustration with existing tools that consistently ship audio off to external clouds.

Core Functionality and Architecture

Inaudio operates on a simple premise: press a keyboard shortcut, speak, and watch the text appear. It maintains a searchable history of all dictated content, allowing users to retrieve specific sentences from previous sessions. The application is built using Electron, React, and TypeScript, with the heavy lifting for speech processing handled by sherpa-onnx. This stack ensures the app remains lightweight while delivering robust performance across operating systems. The tool also features a bidirectional audio interface. Beyond dictation, it can read clipboard content or pasted text aloud using on-device speech synthesis. This function is particularly useful for proofreading, as hearing clunky sentences often reveals errors faster than visual scanning. Additionally, it can transcribe system audio, such as meetings or videos, in real-time using a floating window that displays provisional grey text as it is heard.

Performance and Privacy

Users can download specific models to run inference directly on their CPU. One notable option is NVIDIA's Nemotron streaming ASR model for English, which is tuned for approximately 560 ms latency. The application includes noise filtering capabilities, such as a voice detector that skips non-speech stretches and an optional cleanup model to filter background noise. Input sources are flexible, allowing users to select system audio, microphone, or specific virtual cables.

Key Takeaways

  • Inaudio is a fully local desktop app for dictation and read-aloud features.
  • It uses sherpa-onnx for speech processing and supports NVIDIA Nemotron models.
  • The tool offers searchable history and real-time transcription of system audio.
  • Being a new project with few stars, users should expect rough edges and platform-specific quirks.

The Bottom Line

If you value privacy and want to avoid subscription fatigue for basic dictation, Inaudio is worth a look. However, given its early stage and potential inconsistencies across Windows, macOS, and Linux, it is best suited for developers willing to contribute or tolerate minor bugs in exchange for total data sovereignty.