Evaluate VibeVoice-ASR and other local transcription backends

[hdosys]

VibeVoice-ASR may be worth evaluating as an alternative local backend, particularly for long-form or multilingual transcription:

• https://huggingface.co/microsoft/VibeVoice-ASR
• https://www.youtube.com/watch?v=BYPlfLQm0CQ&t=1441s

A useful comparison would measure latency, memory usage, technical-vocabulary accuracy, multilingual accuracy, and GPU requirements against Whisper, Parakeet, and Canary. VibeVoice may be too resource-intensive for Handy's typical hardware, but its architecture and results are relevant to future backend support.

[cjpais]

This is a pretty big model and not really suited to most people without GPU's or lower end GPU's

[hdosys]

Based on my testing, Parakeet performs well for casual English dictation but loses accuracy on technical vocabulary, medical terminology, and some non-English languages. Whisper Large V3 Turbo with Vulkan runs well on my AMD Ryzen 7 7840HS system, although Handy's current Whisper integration is not yet reliable there.

Canary may be another useful option for GPUs with approximately 12 GB of VRAM. VibeVoice appears more resource-intensive and may be better suited to long-form workloads.

My current preference would be to stabilize Whisper support and benchmark Canary, Parakeet, and VibeVoice against the same representative workload before choosing a default direction.

[cjpais]

Supporting large models at the moment is not a priority. There are relatively few complaints that I have heard on the whole relative to the number of people using the app

There are plenty of people who want free alternatives.

[intech]

@User-3090 @cjpais I encountered the same issue and implemented a canary architecture along with model support #1022