RealtimeSTT is a robust Python library for low-latency speech-to-text, built for applications that need to turn live microphone audio into text with only a few lines of code. With over 10,000 stars on GitHub and active maintenance into 2026, it combines voice activity detection, wake-word activation, and fast transcription into a recorder component that drops neatly into voice assistants and dictation tools.
The library wraps a complete realtime pipeline: audio is captured continuously, a voice activity detector decides when someone is actually speaking, and a transcription engine converts utterances with minimal delay. The default general-purpose engine is faster-whisper, and optional engines can be enabled through install extras when their dependencies and models are present. Optional wake-word support (via Porcupine) enables hands-free activation, callbacks deliver both intermediate and finalized transcripts, and the recorder handles details like pre-recording buffers so the start of a sentence is never clipped. The library is the speech front end of the author’s popular RealtimeTTS/RealtimeSTT ecosystem and is a common building block in open-source voice-assistant projects such as local ChatGPT voice clones.
Free. RealtimeSTT is open source under the MIT license. Transcription runs locally on your own hardware; optional wake-word services may require their own free or paid keys.
Developers building voice assistants, dictation features, accessibility tools, or live-captioning prototypes in Python who need dependable streaming speech recognition without wiring together VAD, audio capture, and Whisper themselves.