Glimpse v0.9.6

Features

• Whisper models can now run on the Apple Neural Engine (ANE) on Apple Silicon Macs, making transcriptions faster and more power-efficient.
• Library overhaul: A new cleaner UI, support for assigning speakers, and better performance.

Improvements

• Much of Glimpse in this update has been optimized for an even smoother experience :)
• Local Whisper transcription is up to 30% faster after decoder tuning.
• Improved performance and reduced memory use during transcription, especially for longer recordings and cloud speech providers.
• Improved recording pill responsiveness and reduced unnecessary background work while recording.
• Improved History performance when loading recordings or importing transcripts.
• Parakeet now transcribes long recordings in chunks, improving transcriptions past around 5 minutes.

Changes

• Onboarding setup is down to one question and on Apple Silicon the Neural Engine encoder comes included automatically.
• Distil whisper models are now under experimental (They're not great at following dictionary rules).