AuroraASR-1M, a new speech recognition model, has been released. It features a lightweight design with 949K parameters and a weight-shared Conformer-CTC architecture. The model achieves 11.43% Word Error Rate on the LibriSpeech test-clean dataset and operates at approximately 436x real-time speed on a single CPU thread. It is licensed under Apache 2.0 and is available with INT8 weights at just 1.06 MB. AI
IMPACT Offers a highly efficient and accessible speech recognition solution for local deployment.
RANK_REASON Release of a new, specific AI model with performance metrics and architecture details. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →