Odia remains an underrepresented language in conversational AI, with existing speech systems struggling to handle strong dialectal variations, regional accents, and noisy real-world speech data. Most available Odia speech datasets are limited in scale, biased toward standard forms, and recorded in controlled environments, leading to poor performance in natural conversations. As a result, Odia speakers experience inaccurate recognition, unnatural responses, and limited access to voice-based AI technologies.

ବାର୍ତ୍ତାଳାପ ( Bartālāpa) is a community-driven initiative to collect, curate, and process Odia speech data for developing advanced speech technologies.

The project focuses on building robust models for speech recognition, text-to-speech, and conversational AI, making Odia language technologies accessible to the common people while preserving linguistic and cultural heritage.

The scope of ବାର୍ତ୍ତାଳାପ (Bartalapa) is centered on Odia speech dataset preparation and ASR research, with a particular emphasis on dialectal variation and noisy real-world speech.

The project includes:

• Collection and curation of diverse Odia speech data covering multiple regional dialects, accents, speaking styles, and background noise conditions
• Annotation and quality control of speech transcripts with rich metadata (dialect, region, noise type, recording setup)
• Design and benchmarking of Odia ASR systems, including dialect-aware and noise-robust models, to study performance gaps and mitigation strategies
• Dataset standardization and documentation to support reproducibility and broader research use

As one component of the project, selected datasets and evaluation protocols will be adapted for participation in international shared tasks (e.g., IWSLT-style low-resource or dialectal ASR tracks), enabling external benchmarking and community engagement.