IBM has released updates to its Granite Speech collection, introducing a new 2-billion parameter model, granite-speech-5.0-2b, and improved versions of two existing models. The new 2B model offers enhanced accuracy, robustness, and control for transcription and translation across multiple languages, including keyword biasing and requested punctuation. The granite-speech-5.0-470m-turboctc model has been updated to be multilingual and faster, achieving a higher RTFx score on the Open ASR Leaderboard. Additionally, granite-speech-4.1-2b-plus has been refined for better speaker attribution and timestamp accuracy in long-form recordings. AI
IMPACT These updates enhance speech processing capabilities, potentially improving enterprise applications requiring accurate transcription, translation, and speaker attribution.
RANK_REASON Frontier-lab model release with system card [lever_c_demoted from frontier_release: ic=1 ai=1.0]
- Apple Silicon
- granite-speech-4.1-2b-plus
- Granite Speech 5.0
- granite-speech-5.0-2b
- granite-speech-5.0-470m-turboctc
- IBM
- llama.cpp
- mlx-audio
- Open ASR Leaderboard
- transformers
- vLLM
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
