search

natrys / whisper.el

Speech-to-Text interface for Emacs using OpenAI's whisper model and whisper.cpp as inference engine.
138 stars 10 forks source link

Support distilled model #15

Open edmundmiller opened 9 months ago

edmundmiller commented 9 months ago

https://github.com/ggerganov/whisper.cpp/pull/1424

It'd be nice to give this a shot.

NightMachinery commented 2 months ago

I have the distilled model downloaded. How do I make whisper.el use that?

~/code/misc/whisper.cpp/models/ggml-distil-large-v3.bin
NightMachinery commented 2 months ago

I got it to work:

  (setq
   whisper-install-whispercpp 'manual
   whisper-install-directory (expand-file-name "~/code/misc/")
   ;; =whisper.cpp= must be found inside `whisper-install-directory'.

   whisper-model "distil-large-v3"
   whisper-language "en"
   whisper-translate nil
   ;; whisper-use-threads (/ (num-processors) 2)

   whisper--ffmpeg-input-device ":0"
   ;; Use `rk/select-default-audio-device' to set this interactively.
   )