Vakyansh

Vakyansh is an end-to-end open-source speech toolkit for Indic languages developed by EkStep Foundation with Thoughtworks, described in a 2022 arXiv paper. The team built automatic pipelines for data creation, training, evaluation, and deployment, producing 14,000 hours of speech data across 23 Indic languages and wav2vec 2.0-based ASR models for 18 languages, plus Glow-TTS and HiFi-GAN text-to-speech models.\n\nAll models, code, and data recipes were open-sourced with the goal of enabling speech-first applications for low-resource Indian languages, and the checkpoints remain a foundation for Indic ASR fine-tunes.