🔍 Read the full analysis: Granite Speech 5.0: Transcribe, Translate, And Attribute More Accurately And Faster on ThorstenMeyerAI.com
TL;DR
IBM announced a new 2-billion-parameter multilingual speech model and updates to two existing Granite Speech models. The company reports gains in speed, transcription accuracy, speaker attribution and timestamp alignment, but the cited figures come from IBM’s evaluations and do not establish performance across every real-world audio setting.
IBM has announced three updates to its Granite Speech collection: a new multilingual 2-billion-parameter model, a faster multilingual version of its 470-million-parameter Turbo CTC model, and revised weights for a model intended for recordings with multiple speakers. The releases target distinct speech tasks, and IBM’s reported performance improvements are based on its own evaluations and cited public benchmarks, as detailed in the original analysis.
The new granite-speech-5.0-2b replaces 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, along with translation from English into Italian and Mandarin. The company says training used 390,000 hours of public audio, as well as synthetic data for Japanese speech recognition, keyword biasing and translation.
IBM says the model is designed to handle noise, reverberation and distant microphones better than its predecessor. Users can prompt it to specify the input language and provide keyword lists of up to 2,000 terms to guide transcription or translation. When formatted output is requested, IBM 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 transcription for Japanese, Spanish, German, French, Italian and Portuguese alongside English, as well as native punctuation and capitalization. IBM reports 19,870 RTFx on the Open ASR Leaderboard, compared with about 12,600 at the model’s August launch, while describing accuracy as unchanged. For 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 more than a halving of average timestamp shift across speech. The company says this update retains the same architecture, prompt format and API.
Three Models, Three Audio Workloads
The releases separate three needs that often compete in speech processing: multilingual translation and transcription, high-throughput transcription, and accurate speaker and timestamp information. Organizations could choose a model according to the audio task rather than relying on one system for every recording. IBM presents the 5.0 2B model for multilingual work, Turbo CTC for speed, and the revised 4.1 2B Plus for longer recordings and speaker attribution.
These capabilities can affect what teams do with transcripts after recognition. More precise timestamps may help align captions or locate passages in audio, while speaker labels may make meeting and support-call records easier to review. Keyword biasing could help with names, acronyms and product terminology. Those are plausible uses of the described features, not measured outcomes in customer deployments: the supplied material does not quantify their impact in particular organizations.
IBM’s benchmark results offer one point of reference, but they do not guarantee the same results on a company’s own audio. Teams would need to test representative recordings and consider the review required for transcripts, especially where errors in names, speaker identity or translated content could matter. The releases also make model choice a practical trade-off: speed, language coverage and transcript detail are different priorities.
speech to text transcription software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How the Granite Updates Differ
IBM introduced Turbo CTC in August as an English-only, 470-million-parameter encoder designed for speed. The October revision keeps the same parameter count and adds six transcription languages, alongside punctuation and capitalization features. The new 5.0 2B model succeeds the 4.1 2B model and adds the described translation capabilities, while the 4.1 2B Plus update focuses on diarization, timestamps and long-form audio.
IBM says the 5.0 2B model ranked second among open-weight automatic speech recognition models by average English word error rate, with a score of 5.06%. The company calls the 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. Lower word error rate indicates fewer recognition errors on that measure; higher RTFx indicates faster processing relative to audio duration. The two metrics describe different aspects of performance.
IBM says all three models are available under Apache 2.0 and have native support in Transformers. It directs users to the models’ cards for evaluation results, architecture and training details, and recommends pairing them with Granite Guardian for production deployments that need additional risk detection. The source material does not provide customer deployment results or independent validation of the announced figures.
“The model is the successor to granite-speech-4.1-2b at the same parameter count.”
— IBM, in its Granite Speech announcement
multilingual speech translation device
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Benchmark Limits and Open Questions
The reported results have not been independently validated in the supplied material. Public benchmark scores do not establish how the models perform across different accents, language varieties, specialized vocabulary or recording conditions, including overlapping speech and noisy environments. IBM’s 5.0 2B capitalization scores are given as a range across test sets, without a breakdown by dataset here.
The source also does not specify the hardware or deployment conditions behind the Turbo CTC RTFx results, and it does not include the full benchmark table needed to inspect the speed and accuracy comparison. IBM’s statement that accuracy was maintained is its own characterization. No customer case studies, pricing information or detailed deployment requirements are provided in the supplied material.
As an affiliate, we earn on qualifying purchases.
Testing and Deployment Details
The models are available under Apache 2.0, with IBM pointing users to their model cards for evaluation results, architecture and training details. Prospective users can compare the models on audio resembling their own workloads, including the languages, microphones and speaking conditions they expect to encounter. The source material does not give a timeline for further releases or announce an independent evaluation.
For teams considering production use, the next practical step is to measure results on representative recordings and decide how transcripts, translations, speaker labels and timestamps will be reviewed. IBM recommends pairing the models with Granite Guardian where additional risk detection is needed. The scope and effectiveness of that pairing in specific deployments are not detailed in the announcement.
professional speech transcription service
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 model and revisions to two existing models in its Granite Speech collection, 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, plus translation from English into Italian and Mandarin.
How much faster is the updated Turbo CTC model?
IBM reports 19,870 RTFx on the Open ASR Leaderboard, compared with about 12,600 at its August launch. The comparison is based on public English short-form test sets; the supplied material does not specify the hardware conditions.
Are the reported gains independently verified?
The source material attributes the performance figures to IBM’s evaluations and cited public benchmarks and provides no independent validation. Results may vary on audio outside those test sets.
What is the license for the models?
IBM says all three models are released under the Apache 2.0 license. It lists support in Transformers and, for the new 5.0 2B model, vLLM, llama.cpp and mlx-audio for Apple Silicon.
Primary source: Hugging Face · via ThorstenMeyerAI.com
