Intron health promises to build africas first speech powered electronic medical record. Current ASR models do not perform very well with african accent and this is due to the lack of diverse speech datasets that are are fully representative of anglophone african countries.
This project uses open access data to build a dataset of free speech that simulates a conversation between people, an interview or a speech.
Clone the project
git clone https://link-to-project
Install yt-dlp.
on a linux VM do:
brew install yt-dlp/taps/yt-dlp
Go to the project directory
cd Intron_accent_data
Install dependencies
conda env create -f requirements.yml
Go to Dataset directory
cd dataset
Run the get_audio script to download youtube dataset
chmod +x get_audio.sh
./get_audio.sh
Run the get_transcripts script to download data get_transcripts
chmod +x get_transcripts.sh
./get_transcripts.sh