SlackBuilds Repository

15.0 > Multimedia > systran-faster-whisper-xxl-bin (245.4)

SYSTRAN Faster-Whisper-XXL is an implementation of OpenAI's Whisper
Speech-to-Text Model.

This is a binary repack.

The default executable is installed into /opt/Faster-Whisper-XXL,
while /usr/bin/whisper is a wrapper, setting the model cache directory
to be ~/.cache/faster-whisper-xxl (which is usually what you want).

This implementation downloads the models automatically from
https://huggingface.co/collections/Systran/faster-whisper
but you can download them yourself and copy into that directory.

The installation of this programs is large (5.6 Gb), and models are
also large (faster-whisper-large-v3 is 3.1 Gb), and they also eat a
lot of RAM or Video RAM when run (on the order of 6Gb).

The full list of models currently available on HuggingFace is:

tiny
tiny.en
base
base.en
small
small.en
medium
medium.en
large-v1
large-v2
large-v3
large
large-v3-turbo
turbo
distil-large-v2
distil-medium.en
distil-small.en
distil-large-v3
distil-large-v3.5


NOTE that larger models do _NOT_ necessarily mean better recognition.
In fact the authors themselves suggest that large-v2, or even medium
might sometimes give better quality than large-v3.
Actually, even "tiny" works quite decently.

Optional deps let you run whisper on the GPU instead of the CPU.
(Much fasta of course.)
cudatoolkit_13, rocmtoolkit_*, available on SBo, or maybe something
for Intel too

This requires: 7zip

Maintained by: Lockywolf
Keywords: stt,whisper,openai,faster-whisper,speech-to-text
ChangeLog: systran-faster-whisper-xxl-bin