tosingithub/AfriSpeech-TTS

African accented clinical and general domain TTS # AfriSpeech-TTS #### African Digital Voices: Pan-African parameter-efficient multi-accent multi-speaker TTS By Intron Innovation By BioRAMP Contributor List: [] #### Progress - [x] Preprocess data - [x] Setup repo and starter scripts - [] Fine-tune backbone - [] Create accent-aware architecture #### Abstract [draft] Recent advances in speech synthesis have enabled many useful applications like audio directions in Google Maps , voice cloning, screen readers, and automated content generation on social media platforms like Tik-tok and Instagram. However many of these systems are dominated by voices sourced from data rich geographies with personas representative of their source data. Although 3000 of the world's languages are domiciled in Africa, African voices and personas are extremely under-represented in these systems. Recent transformer-based state-of-the-art models are large, compute-intensive, and data hungry. Training a representative diversity of models for multiple African accents in resource constrained settings is infeasible and prohibitive. Parameter-Efficient approaches therefore gained popularity using only 1.2% to 0.8% of original trainable parameters to achieve competitive performance in voice synthesis. We present Afro-TTS, the first pan-African accented English speech synthesis system able to generate speech in 100+ African accents, with 200 personas representing the rich phonological diversity across the continent for downstream localized application in Education, Public Health, and Automated Content Creation. ### How to Access the Data Train, dev, and test sets have been uploaded to an s3 bucket for public access. Here are the steps to access the data 1. If not installed already, download and install `awscli` for your platform (linux/mac/windows) following the instructions here 2. Create a download folder e.g. `mkdir AfriSpeech-TTS` 3. Request aws credentials to access the data by sending an email with title "AfriSpeech-TTS S3 Crede …