The development
What happened
IBM announced three updates to its Granite Speech collection: the new granite-speech-5.0-2b, a revised granite-speech-5.0-470m-turboctc, and updated weights for granite-speech-4.1-2b-plus. The releases target multilingual transcription and translation, faster transcription, and speaker attribution, respectively. Hugging Face’s announcement describes the models’ capabilities and reported benchmark improvements; the supplied evidence does not include full evaluation results.
Key facts
IBM describes granite-speech-5.0-2b as the same-parameter-count successor to granite-speech-4.1-2b, trained on 390,000 hours of public audio plus synthetic data.
The new 2B model supports transcription in English, French, German, Spanish, Portuguese and Japanese, and translation to and from English in those languages, plus English-to-Italian and English-to-Mandarin.
The revised 470M Turbo CTC model retains its footprint and adds multilingual transcription in English, Japanese, Spanish, German, French, Italian and Portuguese.
For updated granite-speech-4.1-2b-plus, IBM reports 7% overall reductions in both word diarization error rate and speaker-attributed WER; its architecture, prompt format and API are unchanged.
IBM says all three models are released under Apache 2.0 and have native Transformers support.
The context
Why it matters
The updates address distinct speech workloads without requiring users to treat the collection as a single-purpose transcription model. The speaker-attribution changes are reported as benchmark improvements, but the supplied announcement omits full evaluation details, so their size and applicability cannot be independently assessed here.