Apple Silicon Powers Private On-Device Transcription Apps

Two new macOS applications demonstrate that Apple Silicon can run high‑accuracy speech‑to‑text models locally, eliminating the need for cloud services. Qwen Scribe, an open‑source project, uses the Qwen3‑ASR model on the Mac's Metal GPU and provides drag‑and‑drop transcription, language detection and SRT export while keeping audio and text on the device. The app is distributed as source code and requires local building; signed binaries are planned for a future release. Nabla has launched a commercial clinical dictation app built with Swift and Apple’s Core ML tools. It runs entirely on the Apple Neural Engine, integrates directly with Epic EHR systems and supports existing dictation hardware, offering low‑latency, privacy‑preserving documentation for clinicians.