Suramak is a high-performance, local-first transcription application built natively in Rust using egui and whisper.cpp (via whisper-rs). Designed for complete privacy and offline operation, it processes audio entirely on-device without telemetry or cloud dependencies.
Key Features
- ⚡ Neural Engine Hardware Acceleration: Fast path inference on Apple Silicon via Core ML encoder packages (
--engine coreml), alongside Metal and multi-threaded CPU fallback tiers. - 🎙️ Multi-Source Audio Mixing: Mix multiple microphones or combine microphone and system audio (ideal for meeting transcription) with measured-clock drift correction.
- 🎯 Per-App Process Audio Taps: Capture audio exclusively from a specific running application via macOS CoreAudio process taps without capturing unwanted background sounds.
- 🔴 Live Provisional Transcription: Incoming audio streams are decoded in ~4-second overlapping sliding windows during capture for instantaneous feedback, followed by an automatic full-precision pass upon completion.
- 🎬 Zero-FFmpeg Native Media Decoding: Import video (
.mp4,.mov,.mkv,.webm) and Opus voice notes with native decoders compiled directly into the binary—no externalffmpegbinary dependencies required. - ✨ Audio-Aware AI Polish: Optional post-processing via OpenAI-compatible endpoints that passes chunked audio segments alongside text to accurately resolve names, homophones, and industry terminology.
- 📝 Comprehensive Export Options: Export timestamped transcripts in Markdown, TXT, JSON, and standard subtitle formats (SRT and VTT).
- 💧 Modern Native Interface: Liquid glass interface integration on macOS with floating record controls and interactive waveform playback with segment-following auto-scroll.