add supporting NPU

[hossein-mohseni]

Hey bro, thanks for creating such a great project. I’m even typing this message using the tool right now. My only request is to add support for NPU. For example, my Asus Zenbook laptop has an NPU, and it would be much faster to use it instead of the CPU. I don’t have a dedicated GPU, just the built-in GPU of the Gen 14 Intel processor.

[infiniV]

Thanks for the kind words, glad it's working well for you — love that you typed this with it!

NPU support is on my radar. The tricky part is that VoiceFlow uses faster-whisper (CTranslate2 under the hood), which currently only targets CPU and CUDA — there's no NPU backend upstream. Getting Intel NPU acceleration would likely mean routing through something like OpenVINO with a separate model runtime, so it's a bit more involved than flipping a device flag.

I don't have an NPU-equipped machine to test against, which makes this harder to validate on my end, but I'll look into what's feasible. If anyone with the hardware wants to explore an OpenVINO path, PRs are very welcome.

Will update here as I dig into it.