Launches
Teach your transcriber 2,000 names: IBM updates Granite Speech for multilingual work
Developing 2 sources
IBM's Apache 2.0 speech-language model for multilingual transcription and translation, with language prompts and keyword lists of up to 2,000 terms.
What happened lately
The model card dates release to October 20, 2026. IBM recommends segments of about two minutes or less and Q8_0 for llama.cpp; it warns that lower-bit quantization loses accuracy on audio longer than roughly a minute. Benchmark improvements are IBM-reported results, not AITrending reproductions.