feat: Add support for quantized Whisper model and update audio transc… #1508
Conversation
…ription workflow - Integrate whisper-rs library for improved audio transcription - Add WhisperLargeV3TurboQuantized transcription engine - Modify STT processing to use whisper-rs context and state - Update Cargo.toml to include whisper-rs with GPU support - Refactor transcription methods to work with new whisper-rs workflow - Add download function for quantized Whisper model - Update CLI and core audio transcription engine to support new quantized model
|
@EzraEllette is attempting to deploy a commit to the louis030195's projects Team on Vercel. A member of the Team first needs to authorize it. |
| whisper_model.pcm_to_mel(audio, 2)?; | ||
| let (_, lang_tokens) = whisper_model.lang_detect(0, 4)?; | ||
| let lang_token = get_lang_token(lang_tokens, languages)?; | ||
| params.set_language(get_lang_str(lang_token)); |
|
cool |
|
whisper-rs should support Vulkan hardware acceleration? Would you include Vulkan, hipblas and other features? |
Yes, will you collaborate with me to test on Linux? |
Of course! |
|
I've built with the Vulkan flag, and HW acceleration also works. |
3 participants
…ription workflow
name: pull request
about: submit changes to the project
title: "[pr] "
labels: ''
assignees: ''
description
brief description of the changes in this pr.
related issue: #587
how to test
add a few steps to test the pr in the most time efficient way.
-a whisper-large-v3-turbo-quantizedif relevant add screenshots or screen captures to prove that this PR works to save us time (check Cap).
if you are not the author of this PR and you see it and you think it can take more than 30 mins for maintainers to review, we will tip you between $20 and $200 for you to review and test it for us.