IBM releases granite-speech-5.0-2b for multilingual transcription and speech translation
IBM has released granite-speech-5.0-2b, a model with 2 billion parameters for transcription in six languages and speech translation. It offers prioritization of up to 2 000 terms and an Apache 2.0 license. According to the company, it handles noise and longer translations better.
IBM has released granite-speech-5.0-2b, a model with 2 billion parameters. It supports transcription in English, French, German, Spanish, Portuguese and Japanese. It enables translation in both directions between English and the other five languages; it also translates from English into Italian and Mandarin Chinese.
The prompt can specify the input language, request punctuation and capitalization, or provide a list of up to 2 000 terms for the model to prioritize during transcription and translation. The model is available under the Apache 2.0 license and is supported by Transformers, vLLM, llama.cpp and mlx-audio, the last of which is for Apple Silicon.
According to IBM, the model handles noise, echo, distant microphones and longer or conversational translations better than its predecessor. The company reports an average word error rate of 5.06 % on public tests of short English recordings on the Open ASR Leaderboard. It describes the result as unofficial and dates it to 15 October 2026; this does not establish the same accuracy in other languages.
Why it matters
The ability to provide a custom vocabulary is useful for recordings containing product names, personal names and abbreviations. The open license and support from several tools give development teams the option to integrate transcription and translation into their own applications. However, the reported result for short English recordings does not, on its own, determine quality on specific company data.
Two audiences, two different impacts
What this means
For individuals
When working with recordings in supported languages, you can specify the language and desired text formatting in the prompt and provide your own list of technical terms.
For a business
Teams developing speech processing applications gain access to a model under the Apache 2.0 license, supported by several deployment tools. They can assess how to integrate it into their existing transcription and translation workflow.
DevelopmentCheck the original
Event sources
clearly official source · 1 publisher, 0 independent. We count feeds from the same owner only once.