Granite Speech 5.0: Transcribe, Translate, and Attribute more Accurately and Faster
huggingface ·
IBM has released three significant updates to its Granite Speech collection, enhancing capabilities for transcription, translation, and speaker attribution. The new granite-speech-5.0-2b model offers higher accuracy and robustness against noise, supporting multilingual transcription and translation with features like keyword biasing and language ID prompting. Additionally, the updated granite-speech-5.0-470m-turboctc is now multilingual and significantly faster, achieving nearly 1.6x speed improvements while maintaining high throughput for on-device use. The revised granite-speech-4.1-2b-plus focuses on multi-speaker recordings, delivering improved speaker attribution, tighter timestamps, and better performance on long-form audio. These models are designed to cover diverse enterprise needs, from high-accuracy translation to rapid edge processing and detailed meeting analysis. All models are available under the Apache 2.0 license and integrate with popular frameworks like Transformers and vLLM. IBM recommends pairing these tools with Granite Guardian for enhanced risk detection in production environments.
huggingface