🔍 Read the full analysis: Granite Speech 5.0 Brings Faster, More Accurate Transcription And Translation on ThorstenMeyerAI.com
TL;DR
IBM announced three updates to its Granite Speech collection: a new multilingual 2-billion-parameter model, a faster multilingual Turbo CTC model, and revised weights for a model focused on speaker attribution and timestamps. The performance improvements cited are IBM’s own evaluations; independent validation and results across specific customer workloads are not provided.
IBM has announced three updates to its Granite Speech collection: a new multilingual 2-billion-parameter model for transcription and translation, a faster multilingual revision of its 470-million-parameter Turbo CTC model, and updated weights for a model aimed at speaker labeling and timestamps. The releases address different speech-processing tasks, as detailed in the original analysis, while the reported performance figures come from evaluations cited by IBM, not independent testing.
The new granite-speech-5.0-2b succeeds the 4.1 2B model at the same parameter count. IBM says it transcribes English, French, German, Spanish, Portuguese and Japanese, and supports translation to and from English for those languages, plus translation from English into Italian and Mandarin. The company says it trained the model on 390,000 hours of public audio, alongside synthetic data for Japanese speech recognition, keyword biasing and translation.
IBM says the 5.0 model handles noise, reverberation and distant microphones better than its predecessor. It supports prompts identifying the input language and keyword lists of up to 2,000 terms to guide transcription and translation. When formatted output is requested, the company reports capitalization F1 scores ranging from 89.7 to 99.5 across its test sets. The model is released under Apache 2.0 and is supported in Transformers, vLLM, llama.cpp and mlx-audio for Apple Silicon.
The revised granite-speech-5.0-470m-turboctc adds Japanese, Spanish, German, French, Italian and Portuguese transcription alongside English, as well as native punctuation and capitalization. IBM reports a speed score of 19,870 RTFx on the Open ASR Leaderboard, compared with about 12,600 at the model’s August launch, while saying accuracy remained at the same level. For the updated granite-speech-4.1-2b-plus, IBM reports a 24% reduction in word error rate on long-form test sets, a 7% reduction in two speaker-attribution error measures, and an average timestamp shift reduced by more than half. The company says that update keeps the same architecture, prompt format and API.
Three Models for Different Audio Tasks
The releases give organizations separate options for workloads that place different demands on speech systems. The 5.0 2B model is positioned for multilingual transcription and translation; Turbo CTC is aimed at high-volume transcription where speed matters; and the Plus update targets recordings where speaker attribution and timing are important. That division may help teams choose a model for a particular job rather than expect one system to perform equally well across every audio task.
Those capabilities can affect how transcripts are used after recognition. More accurate timestamps may help align captions or locate passages in recordings, while speaker labels may make meetings and support calls easier to review. Keyword biasing could help with names, product terms and acronyms that are often misrecognized. These are plausible applications of the features described, but the supplied material does not quantify their effect in customer deployments.
IBM’s performance figures offer prospective users a reference point, not a guarantee of results on their own recordings. A model that performs well on public test sets may behave differently with specific accents, specialist vocabulary, overlapping speakers or recording equipment. Organizations still need to test representative audio and account for transcript review, computing costs and the consequences of recognition errors.
transcription software for multilingual audio
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How the Granite Models Differ
Granite Speech is IBM’s collection of models for speech recognition and related tasks. IBM introduced Turbo CTC in August as an English-only, 470-million-parameter encoder designed for speed. The October revision retains the same model size while adding six transcription languages and changes intended to support difficult acoustics and more readable output.
The new 5.0 2B model takes a broader multilingual direction, combining transcription with translation features. The 4.1 2B Plus update retains its existing architecture and interface while changing model weights for long-form recognition, speaker attribution and timestamps. IBM says all three models are available under an Apache 2.0 license and have native support in Transformers. It points users to the models’ documentation for evaluation results, architecture and training details, and recommends pairing them with Granite Guardian in production deployments that need additional risk detection.
IBM reports that the 5.0 2B model ranked second among open-weight ASR models on average English word error rate, at 5.06%. The company labels this result unofficial and says it averages public English short-form test sets on the Open ASR Leaderboard as of October 15, 2026. Its Turbo CTC speed comparison uses nine public English short-form test sets and is dated October 2026. Word error rate and RTFx measure different aspects of performance: lower word error rate and higher RTFx indicate better results on their respective measures.
“The model is the successor to granite-speech-4.1-2b at the same parameter count.”
— IBM, in its Granite Speech announcement
As an affiliate, we earn on qualifying purchases.
Limits of the Reported Tests
The announcement does not provide independent validation of IBM’s reported improvements. The cited short-form leaderboard results also do not establish how the models will perform across all accents, language varieties, recording conditions or business workflows. IBM’s stated improvements for noise and distant microphones are not accompanied in the supplied material by a detailed breakdown of those test conditions.
Some benchmark details remain incomplete. The capitalization F1 range for the 5.0 2B model is not broken down by test set in the source material. IBM says Turbo CTC’s accuracy was unchanged as its RTFx score increased, but the full benchmark table is not included here. The hardware and deployment setup used to produce the speed figures are also unspecified. It is therefore not possible from the supplied information to assess how closely those results would match a particular user’s environment.
professional speech recognition microphone
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Testing Before Deployment
IBM directs users to each model’s documentation for evaluation results, architecture and training details. The next practical step for prospective users is to compare the models on audio representative of their own workloads, including relevant languages, accents, microphones and specialist terms. That testing can show whether reported benchmark gains carry over to a specific use case and what review is needed before transcripts are relied on.
The announcement does not identify a separate upcoming release date or independent evaluation milestone. Further detail may become available through the model documentation and subsequent testing by users. Until then, IBM’s benchmark results should be read as publisher-reported measurements, with real-world performance still dependent on the audio and deployment conditions.
noise-canceling microphone for transcription
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did IBM announce?
IBM announced one new Granite Speech model and updates to two existing models, targeting multilingual transcription and translation, faster transcription, and improved speaker attribution and timestamps.
Which languages does the new 5.0 2B model support?
IBM says it transcribes English, French, German, Spanish, Portuguese and Japanese. It supports translation to and from English for those languages, as well as translation from English to Italian and Mandarin.
What does the Turbo CTC update change?
The 470-million-parameter Turbo CTC revision adds transcription in six languages beyond English, along with native punctuation and capitalization. IBM reports a higher RTFx speed score and says accuracy remained at the same level.
Are the performance improvements independently verified?
Not in the supplied source material. The performance figures are reported by IBM, and the announcement does not provide independent validation or enough detail to determine how results will transfer to every recording environment.
What should organizations test before using the models?
Organizations should evaluate the models on audio similar to their own, including relevant accents, languages, noise levels, overlapping speakers and specialist vocabulary. They should also decide how transcripts will be reviewed and what errors are acceptable for their workflow.
Primary source: Hugging Face · via ThorstenMeyerAI.com
