Voice typing on Linux has taken a step forward with the introduction of a new application called Speed of Sound, which utilizes OpenAI’s Whisper model. Despite voice typing being a prevalent feature on mobile devices, it has not widely gained traction on desktop platforms, primarily due to past inaccuracies and inefficiencies in speech-to-text technology.
With Whisper, released in 2022, users have experienced significantly improved performance in speech recognition, which has spurred the development of various audio-to-text applications, including those for podcast transcriptions and auto-subtitles. The Speed of Sound app leverages a smaller version of the Whisper model, allowing users to transcribe speech directly into any text field on their Linux systems simply by speaking.
The application operates by pressing a button or using a keyboard shortcut to begin recording. After speaking, the recorded audio is then converted into text, which appears in the active text application. This functionality is compatible with major desktop environments, including GNOME and KDE, across both X11 and Wayland. Users can personalize the model by providing details like their writing style and specific vocabulary for better recognition.
Speech processing is done locally and offline, ensuring that no audio data leaves the device. However, it’s important to note that transcription is not entirely "real-time"; users must trigger the recording intentionally. If the accuracy is less than desired, additional models can be downloaded, or users may connect to cloud or self-hosted large language models for enhanced performance.
While the application is not a complete replacement for traditional typing, it serves as a useful tool for casual note-taking or composing drafts without being tied to a keyboard. It offers a refreshing alternative for creating written content, allowing a more natural flow of ideas.
Speed of Sound is open-source and can be installed from Flathub and the Snap Store, with alternative packaging options available on its GitHub releases page.