From 1ea9cc6d5ec2fe7efc95800431d1e96bc197e92b Mon Sep 17 00:00:00 2001 From: Pierre Guillot Date: Fri, 13 Dec 2024 16:13:35 +0100 Subject: [PATCH] docs: update whisper-manual.md --- docs/whisper-manual.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/docs/whisper-manual.md b/docs/whisper-manual.md index 5cb6681..a1180f4 100644 --- a/docs/whisper-manual.md +++ b/docs/whisper-manual.md @@ -25,7 +25,7 @@ The Whisper plugin is an implementation of the [Whisper](https://github.com/open The Whisper plugin analyses the text in the audio stream and generates markers corresponding to phrases, words or tokens (depending on the *Split Mode* parameter). The *Suppress Non-Speech Tokens* parameter controls whether non-speech tokens are generated (only usable with *Split Mode* on *Tokens*). -The lightweight ggml-base-q5_1 model is embedded in the plugin and the other q5 models (tiny, small, medium, and large_v2) will be installed on your system. The *Model* parameter is used to select which model to use. You can also download and use other models that may be more appropriate to your needs. Please, refer to the following section dedicated to models. +The lightweight ggml-base-q5_1 model is embedded in the plugin and the other models will be installed on your system. The *Model* parameter is used to select which model to use. You can also download and use other models that may be more appropriate to your needs. Please, refer to the following section dedicated to models. The Whisper Vamp Plugin has been designed for use in the free audio analysis application [Partiels](https://forum.ircam.fr/projects/detail/partiels/). @@ -44,7 +44,7 @@ Use the installer for your operating system. The plugin dynamic library (*whispe ## Models -By default, the plugin embeds the ggml-base-q5_1, which requires little space and offers fairly fast calculation and the other q5 models (tiny, small, medium, large_v2, and large_v3) are installed on your system. +By default, the plugin embeds the ggml-base-q5_1, which requires little space and offers fairly fast calculation and other models small-q5_1, medium-q5_0, large-v2-q5_0, and large-v3-turbo-q5_0 (Mac & Linux only) are installed on your system. It is possible to download other models that are potentially more robust or more adapted to your needs. To do this, you can use the `download-ggml-model.sh/cmd` scripts from Georgi Gerganov's [Whisper.cpp](https://github.com/ggerganov/whisper.cpp) project and provided with the plugin package. Once downloaded, these models must be installed in the repository: - Linux: `~/.config/Ircam/whispermodels` or `/opt/Ircam/whispermodels`