đ Exciting News: Two New Open-Source Bambara ASR Models!
Weâre thrilled to announce the release of two lightweight Bambara ASR models from our early fine-tuning experiments! These models achieve near-SOTA performance for open-source Bambara ASR (as of Feb 2025) while remaining efficient enough for real-world deployment.
đ§ Why Does This Matter?
Bambara is a severely under-resourced language. While ASR research has made rapid progress in major languages, open-source models for Bambara remain rare. Our goal is to share our fine-tuning experiments to help push research forward and support the development of ASR for African languages.
đïž Model 1: Parakeet-114M TDT-CTC
đ Hugging Face: RobotsMali/soloni-114m-tdt-ctc
đ§ Model Size: ~114M parameters
đ WER (TDT Decoder): 66%
đ WER (CTC Decoder): 40.6%
đĄ Key Takeaway: Despite being small, the CTC branch outperforms previous open-source models.
đ§ Model 2: QuartzNet-15×5
đ Hugging Face: RobotsMali/stt-bm-quartznet15x5
đ§ Model Size: ~19M parameters
đ WER: 46.5%
đĄ Key Takeaway: This lightweight model is optimized for real-time ASR in low-resource environments.
đ€ Final Thoughts
We hope this contributes to Bambara ASR research and encourages more collaboration. đ Stay tuned for future updates!
đ© Want to contribute or learn more? Reach out via RobotsMali or join the discussion on GitHub!
