Plugins

This section defines settings for Speech To Text and Object Detection.

Before Speech To Text and Object Detection can be configured and used, Python3 needs to be installed. Once Python3 is installed, you may want to use a virtual environment (venv) to keep it separate for Kdenlive from other uses on your system.

If you have installed Python3 already jump right to the configuration page of Speech To Text or Object Detection.

Python installation

Installation on Linux

On most Linux distributions Python is installed by default. You can check if that is the case for your system by running python3 --version in a terminal. The following are the basic steps for installing Python3 on Ubuntu. If your distribution is not Ubuntu-based please refer to the specific documentation or search the Internet for installation instructions.

$ sudo apt updates
$ sudo apt install python3

The crucial third-party Python package you may need is pip. Python 3.4 and later include pip by default but it does not hurt to check by running command -v pip in a terminal (some distributions use pip for Python2 and pip3 for Python 3). If pip is missing, you can install it with

$ python3 -m ensurepip --upgrade

In case of issues please refer to the pip installation guide.

Nota

In the following paragraphs, pip is the generic term for all versions of pip, including pip3. Please use the correct pip command for your OS.

Installation on Windows

Download Python from the official Python download page.

Speech To Text

This section is used for setting up the Speech-to-Text feature of Kdenlive and for managing the various models for the two engines VOSK and Whisper.

Avvertimento

Speech To Text does not work with version 21.04.2 due to Vosk API issues. Use version 21.04.1 or 21.04.3 and later versions.

Speech Engines

There are two speech engines available: VOSK and Whisper. OpenAI’s Whisper is a speech recognition module for general use trained on a large dataset of diverse audio and is capable of performing speech translation, and language identification.

Whisper is slower than VOSK on CPU, but it is more accurate than VOSK. Whisper creates sentences with punctuation marks, even in Base mode.

You need to set up the models to be used by these engines.

Suggerimento

If you are using the flatpak version of Kdenlive, you may experience problems with installing the speech models. The sandbox approach of flatpak prevents kdenlive from running pip. There is a possible workaround using $ flatpak run --command=/bin/bash org.kde.kdenlive and then $ python -m ensurepip followed by $ python -m pip install -U openai-whisper torch (courtesy of Veronica Explains). Your mileage may vary.

The other option is to select Whisper and then click on Install multilingual translation. This will download and install the necessary dependencies and make Kdenlive aware of the location of your installation of Python and pip. After that you can follow the instructions for setting up VOSK and Whisper here.

VOSK

Python is working, but VOSK is not yet usable due to missing speech models

You need to download a speech model first from the alphacephei[1] download page. Follow the link (2) and download the models you need.

By default, the language models are installed into the following folders:

Linux:

~/.local/share/kdenlive/venv/Lib

Windows:

%LocalAppData%\kdenlive\venv\Lib

If you want to use a specific folder, check Custom models folder (3) and specify it in the text field below or click on Open file dialog to navigate to the target folder. If you use the default folder, it will be displayed for informational purposes in the models folder text field.

If you have installed VOSK in an earlier Kdenlive version, and now you have chosen the venv folder for Python, you can delete the previously installed VOSK libraries by using the following command in a terminal:

$ pip uninstall vosk srt

Click on Add Model and enter the path to the file(s) you downloaded.

Click on Open file dialog to open the file manager of your OS to navigate to where you downloaded the files and select the model file you want to add.

Alternatively, drag & drop the language model you want from the alphacephei[1] download page to the model window (4), and Kdenlive will download and extract it for you into the default folder or the custom folder you specified.

Nota

The models are compressed files (.zip) and can be several GB big. Depending on your internet connection, download times could be long. After the download, the files need to be extracted which depending on your system configuration can also take a long time. Kdenlive will appear unresponsive but is working in the background. Please be patient.

Once models have been installed, Kdenlive displays the size of the model folder. Click on Models folder to open the models folder with the file manager of your OS.

Whisper

Install missing dependencies

When you switch to Whisper for the first time you have to install the missing dependencies (about 2GB to download).

After that you need to download one or more speech models.

Whisper is installed but no speech model has been downloaded

1:

Whisper is selected for speech recognition

2:

When left at Autodetect, Kdenlive will try to figure out which language to use for speech recognition. If this gives the wrong results, select the correct language here.

3:

You can switch between using the CPU or your GPU for speech recognition. A GPU supporting CUDA is required for GPU speech recognition.

4:

Only for GPU. When Kdenlive detects a NVIDIA GTX 16xx graphic card it disables half precision (FP16) automatically. If you have issues with using GPU you can switch off half precision.

5:

You can have Whisper translate the text to English. If you need translation to other languages, you need to click on Install multilingual translation. This will enable SeamlessM4T[2] and download and install its models (around 10GB of data). Processing will happen offline from then on.

Click on Manage models or go with the recommendation of using the turbo model by clicking on Download (1.4GB). More information about the available models is on the Whisper source code page.

Whisper download and manage models

Kdenlive shows the download process.

Installed speech models have a solid circle. You can delete them by clicking on Uninstall model

Available models have a hollow circle. You can install them by clicking on Install model.

When all is configured correctly you get this screen: All green!

Path where Whisper is installed:

Linux:

~/.local/share/kdenlive/venv/Lib

Windows:

%LocalAppData%\kdenlive\venv\Lib

The Whisper speech models are stored here:

Linux:

~/.local/share/kdenlive/opencvmodels

Windows:

%AppData%\kdenlive\opencvmodels

To download and start subtitle translation follow these steps.

You can check for updates by clicking on Check configuration

If you have installed Whisper in an earlier Kdenlive version, and now you have chosen the venv folder for Python, you can delete the previously installed Whisper libraries by using the following command in a terminal:

$ pip uninstall openai-whisper

Nota

If you get consistent messages during speech recognition about missing model files, check where clicking on the link next to Models folder takes you. If it is ~/.cache where there is a folder Whisper containing all the models you downloaded, simply copy this folder to where the error message says they are missing (most likely: ~/.var/app/org.kde.kdenlive/cache)

Object Detection

The first time you use object detection the plugin has to be installed.

The plugin gets installed in folder:

Linux:

~/.local/share/kdenlive/venv-sam

Windows:

%LocalAppData%\kdenlive\venv-sam

If all is properly installed it should look like so:

When all is configured correctly you get this screen: All green!

1:

Model You can download different models for object detection.

2:

Device You can switch between using the CPU or your GPU for object detection. A GPU supporting CUDA is required for GPU object detection. Kdenlive tries to find your graphic card automatically. If the graphic card supports CUDA (Nvidia GPU) it will install the necessary driver so you can use it. If your graphic card doesn’t support CUDA or is not detected by Kdenlive you only see CPU. GPU Support If your Nvidia GPU is not listed, try installing an alternative by following these steps.

3:

Offload video to CPU to save GPU Memory If you have long clips on which Kdenlive has to detect objects you may get a crash. If enabled, this offloads some of the memory used to the RAM instead of the GPU memory.

4:

Use system packages only When enabled, Kdenlive will use the SAM2 version installed on your system. For advanced users only, as you will have to properly setup everything by yourself.

5:

Models folder When you click on the link the folder will be opened where the SAM2 models are stored including the size of this folder. Delete all models will delete the content of the model folder.

6:

Plugin size When you click on the link the folder will be opened where the venv-sam Python scripts are stored including the size of this folder. Uninstall plugin will uninstall venv-sam Python.

GPU manual installation

If Kdenlive doesn’t detect your NVIDIA GPU automatically you can try to install it manually.

GPU acceleration only works with CUDA (Compute Unified Device Architecture) which is a proprietary NVIDIA parallel computing architecture.

Check the CUDA version you have installed:

• open a command shell (Windows: Windows+R enter cmd and hit enter)

• Enter nvidia-smi –-version

You should get something like this:

Here the CUDA version is 12.8

Now click on GPU Support

You can select a CUDA version equal to or lower than what was detected

Select a CUDA version which is lower than what was detected.

Click Apply

On the next window click Continue

Kdenlive starts downloading all necessary files for using the GPU. Once finished you should see your GPU under Device like so:

Successfully installed GPU

---

[1] (1,2)

The alphacephei download page with the different language models looks like this:

[2]

This is meta’s SeamlessM4T model