Northeast STT Multilingual Speech to Text Model

About Model

Northeast STT is a multilingual Speech-to-Text (STT) model developed by MWire Labs to advance automatic speech recognition for the indigenous languages of Northeast India. Built on Whisper Medium, the model is fine-tuned using 150,483 speech-text pairs collected from the ARTPARK IISc Vaani corpus and additional speech datasets developed by MWire Labs. The model supports eight Northeast Indian languages: Khasi, Garo, Mizo, Kokborok, Nagamese, Assamese, Chakma, and Wancho. Rather than training separate models for each language, Northeast STT provides a unified multilingual model capable of transcribing speech across multiple low-resource languages within a single framework. By combining these datasets with large-scale public speech corpora, the model improves speech recognition coverage for languages that remain underrepresented in existing AI technologies. Northeast STT is intended for automatic transcription, language documentation, corpus creation, accessibility tools, digital archives, academic research, and multilingual speech-enabled applications. It provides one of the broadest multilingual Speech-to-Text models currently available for Northeast Indian languages and serves as a foundation for future speech technologies in the region.