Sonicribe v2.4.0: WhisperKit, Faster Transcription, Lower RAM
Sonicribe v2.4.0 brings WhisperKit support for Apple Neural Engine acceleration, 2-3x faster transcription via optimized whisper.cpp, lower idle RAM, and one-click model downloads.
Sonicribe Team
Product Team
Table of Contents
Sonicribe v2.4.0 is Here
Version 2.4.0 is the biggest performance update to Sonicribe since launch. It ships a new transcription engine option, measurable speed gains, lower memory footprint, and a smoother model setup experience.
Download v2.4.0What's New in v2.4.0
WhisperKit — Apple Neural Engine Support
Sonicribe now supports WhisperKit as an alternative STT engine alongside whisper.cpp. WhisperKit runs Whisper models natively on Apple's Neural Engine (ANE), offloading transcription from the CPU/GPU entirely.
On Apple Silicon Macs (M1 and later), this means:
- Lower CPU and GPU utilization during transcription
- Consistent performance even under heavy workloads
- Full parity with whisper.cpp accuracy on supported models
To switch engines, open Settings → STT Engine and select WhisperKit. Models download separately for each engine.
2–3x Faster Transcription
whisper.cpp has been updated to the latest upstream version with additional optimizations tuned for Apple Silicon. Benchmarks on the same audio file:
| Model | v2.2.1 | v2.4.0 | Speedup |
|---|---|---|---|
| Large-v3 Turbo | 8.2s | 3.1s | ~2.6x |
| Large-v3 | 22s | 9s | ~2.4x |
| Medium | 5.1s | 2.0s | ~2.6x |
Results on M3 Pro with 30-second audio clip. Intel Macs will see smaller but meaningful gains.
Lower Idle RAM Usage
Sonicribe no longer holds the model in memory between recordings by default. Instead, the model loads on demand and releases after a short idle period. This reduces background RAM usage by approximately 150–300 MB depending on model size.
The behavior is configurable: go to Settings → Memory to pin the model in memory if you prefer faster response on short intervals.
One-Click WhisperKit Model Download
WhisperKit models can now be downloaded directly from within Sonicribe. Open Settings → Models, find the WhisperKit section, and click Download next to any model. No manual file management required.
How to Update
If auto-update is enabled, Sonicribe will prompt you on next launch. To update manually:
2. Run the installer — your settings and license are preserved
3. Relaunch Sonicribe
Full Changelog
Added- WhisperKit engine option (Apple Neural Engine acceleration on M1+)
- One-click model download for WhisperKit models in Settings
- whisper.cpp updated to latest upstream with Apple Silicon tuning (2–3x faster)
- Idle RAM reduced by 150–300 MB via on-demand model loading
- Minor UI fixes in Settings panel
- Improved error messaging when model download fails
Feedback
Found a bug or have a feature request? Reach out via the in-app feedback button or email support@sonicribe.online.
Ready to transform your workflow?
Join thousands of professionals using Sonicribe for fast, private, offline transcription.