This is meant as an editing extension for Davinci Resolve.
Among other things, it transcribes timelines directly from Resolve by rendering to Audio Only (mp4/wav) and then passing the rendered file to OpenAI Whisper, a state-of-the-art speech recognition model.
The tool saves the transcription into both a detailed JSON and an SRT file ready to be imported into Davinci Resolve (compatible with other software aswell).
The transcription and translation functions also work without the need to have Resolve installed on the machine.
The app is in this stage raw and not polished at all, but it already helps us in the editing room, so we simply wanted to share it for free for anyone to use!
Yes, the tool runs locally and there's no need for any additional account.
The results we get with Whisper are significantly better than any other Speech-to-Text model (including Google, AWS etc.) out there and the models are free to use. According to OpenAI, the models have been trained on data in 98 different languages (cca. 65% of data in English) and show strong Automated Speech Recognition results in ~10 languages. More info here: https://github.com/openai/whisper/blob/main/model-card.md
Our tests show that on a Macbook Pro M1 a 30-second timeline is transcribed on average in approx. 1 minute, but results may vary.
On a Windows workstation with a GTX1070, the transcription time is around a quarter of the length of the audio file (4-minute audio is transcribed in approx. 1 minute).
Our installation is on MacOS 12.6 running on M1, but the scripts should run fine on other CPUs and GPUs. For both production and development we're currently using Python 3.9.13.
Note: Whisper worked fine on Python 3.10.2, but we ran into problems when trying to install some packages which we're planning to use for future developments.
The tool only works on Resolve Studio 18.
Note: Unfortunately, only the Studio version of Resolve supports external scripting and Resolve versions earlier than 18 do not support Python 3.6+
We recommend running the tool inside a virtual environment like virtualenv. This is not required, but it prevents messing up your Python installation.
**Important Disclaimer: you need to be comfortable with using the Terminal on Mac OS, or the Command Prompt on Windows. You should be fine even with little to no experience, but keep in mind that you are installing stuff on your machine and there is a (very) slight chance that you'll affect your Operating System's overall performance or even the performance of your other apps. In an unlikely worst-case scenario, some stuff might not work anymore at all and you'll need pro help to fix them back.
In the Terminal:
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
brew install python@3.9
brew install python-tk@3.9
brew install git
brew install ffmpeg
pip install virtualenv
First, go to the Folder you want to install StoryToolkit in via Finder. Right-click on it and select "New Terminal at Folder". Once you get another terminal window open, run:
git clone https://github.com/octimot/StoryToolkitAI.git
This should download the app in the folder that you chose.
Now create a virtual environment (to prevent messing up with other python packages you may have installed on your OS for other stuff):
virtualenv -p python3.9 venv
Right now, your installation folder should contain 2 other folders, and the tree should look like this:
YOUR_INSTALLATION_FOLDER
+- StoryToolkitAI
+- venv
Now enable the virtual environment (this means that all the packages you'll install now via pip will be contained in the virtual environment, meaning that for the tool to work you'll ALWAYS have to activate the virtual environment first using the following command!)
source venv/bin/activate
pip install git+https://github.com/openai/whisper.git
For more info regarding Whisper installation, please check https://github.com/openai/whisper
pip install -r StoryToolkitAI/requirements.txt
If you are running the tool on a machine with an NVIDIA CUDA GPU, make sure you install Torch with CUDA:
pip uninstall torch
pip cache purge
install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu116
Note: If Resolve is not turned on or not available, the transcription and translation functions will work on normal wav files too. Simply press the transcribe or translate buttons and follow the process.
Inside the virtual environment, you should now be able to start the tool:
python StoryToolkitAI/app.py
Note: After restart of the machine or your terminal window, never forget to activate the virtual environment before starting the app. In the folder where you created venv, run:
source venv/bin/activate
Detailed Windows installation instructions coming soon.
Open a project in Resolve. Then, go back to the folder where you installed the tool, activate the virtual environment (if you created one) and then start the tool:
source venv/bin/activate
python StoryToolkitAI/app.py
A simple GUI should appear on the screen:
Go back to Resolve and open the Timeline that you want to transcribe, then click the "Transcribe Timeline" button and follow the process from there.
Once the process has started, it needs a bit of time to transcribe. As soon as it's done, it will save an SRT with the transcription and a JSON with the full result generated by Whisper.
When the transcription is ready, you can choose a Media Bin in Resolve where the app will automatically import the generated SRT file.
The tool also supports direct translation to English by clicking the "Translate Timeline to English" button. However, it will not generate any original language transcription together with the translation.
There seems to be a problem with the Resolve libraries on some Windows 10 machines, but we're trying to find a fix. If you're running the tool on a Windows 10 machine with Resolve 18 and it works, please get in touch.
To report any issues, please use the Issues tab here on Github: https://github.com/octimot/StoryToolkitAI/issues
This tool is developed and maintained by Octavian Mot (https://mots.us).
Feel free to get in touch or contribute.
Ideally, it should evolve by incorporating other machine learning models such as CLIP and GPT-3 to assist editors in their work.