Part of a broader ecosystem of NLP infrastructure for Southern Quechua — alongside QuechuaTok. This project focuses on audio enhancement and speech processing for Quechua language data, addressing the severe underrepresentation of indigenous languages in speech ML.
Southern Quechua has virtually no speech data in the major public corpora used to train speech models. The challenge is not just collecting audio — it is building the tooling to clean, align, and enhance recordings from field conditions, where equipment is variable and acoustic environments are complex.
The audio enhancement pipeline targets noise reduction, speech-to-noise ratio improvement, and segment alignment for downstream ASR training. It is designed to run on low-resource hardware and integrate directly with the QuechuaTok tokenization pipeline.