🔍 Read the full analysis: Use Granite Speech 5.0 To Transcribe, Translate, And Attribute Audio More Efficiently on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
IBM has announced one new model and updates to two existing models in its Granite Speech collection. The releases target multilingual transcription and translation, faster transcription, and improved speaker attribution and timestamps; performance figures cited in the announcement come from IBM’s evaluations.
IBM has announced three updates to its Granite Speech collection: a new multilingual 2-billion-parameter model for transcription and translation, a faster multilingual version of its 470-million-parameter Turbo CTC model, and revised weights for a model focused on speaker attribution and timestamps. The releases give developers options for different audio tasks, while the performance figures published with the original analysis are IBM-reported evaluation results, not independent validation.
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. It also supports translation from English to Italian and Mandarin. The company says the model was trained on 390,000 hours of public audio, with synthetic data added for Japanese speech recognition, keyword biasing and translation.
IBM says the 5.0 model is more resilient than its predecessor to noise, reverberation and distant microphones, and can take prompts specifying the audio’s language. Users can provide keyword lists of up to 2,000 terms to guide transcription and 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 the Apache 2.0 license and is supported in Transformers, vLLM, llama.cpp and mlx-audio for Apple Silicon.
The updated 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 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. The revised granite-speech-4.1-2b-plus targets multi-speaker and long-form audio. 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. The update keeps the architecture, prompt format and API unchanged, according to IBM.
Three Models, Different Audio Workloads
The releases separate three needs that can be difficult to optimize at once: multilingual accuracy and translation, high-speed transcription, and useful speaker and timing information. Organizations processing meetings, customer calls or other recorded speech may choose a model according to whether they need translation, throughput, or clearer attribution and alignment. Those are potential applications of the features IBM describes; the announcement does not quantify results in customer deployments.
Improvements to timestamps and speaker labels can make transcripts easier to search, review or align with audio, while keyword biasing may help with product names, acronyms and other specialized terms. But benchmark gains alone do not establish that a system will perform well on a particular organization’s recordings. Teams will still need to test representative audio and account for transcription review, deployment costs and the consequences of errors.
Products Worth Considering

Sonos Roam (Gen 1) Portable Smart Speaker, White

Sonos One Two-Room Set Smart Speakers with Built-In Alexa, Black

LG UltraGear 27-inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDR True Black 400
As an affiliate, we earn on qualifying purchases.
How the Granite Releases Differ
IBM introduced the 470-million-parameter Turbo CTC model in August as an English-only encoder designed for speed. The October revision retains the model size and adds six transcription languages, alongside punctuation and capitalization. The new 5.0 2B model is positioned for multilingual transcription and translation; the 4.1 2B Plus revision is aimed at long recordings where speaker attribution and timing matter.
IBM says the 5.0 2B model ranks second among open-weight ASR models on average English word error rate, with a score of 5.06%. The company describes this as an unofficial comparison based on public English short-form test sets on the Open ASR Leaderboard, dated October 15, 2026. Its Turbo CTC speed comparison uses nine public English short-form test sets and is also dated October 2026. Word error rate and RTFx measure different things: lower word error rate indicates fewer transcription errors, while higher RTFx indicates faster processing relative to audio duration.
IBM says all three releases use the Apache 2.0 license and have native support in Transformers. It directs users to the model cards for evaluation results, architecture and training details. The company also recommends pairing the models with Granite Guardian for production deployments that need additional risk detection; that recommendation is not a reported measure of speech-recognition performance.
“The model is the successor to granite-speech-4.1-2b at the same parameter count.”
— IBM, in its Granite Speech announcement
What the Published Tests Do Not Show
The supplied announcement material does not include independent validation of IBM’s performance claims. Public short-form benchmarks cannot establish how the models perform across every accent, language variety, recording environment or business workflow. The results also do not by themselves show how accurately the systems handle overlapping speakers or highly specialized vocabulary in a specific organization’s recordings.
Some reported details remain limited. IBM gives the 5.0 2B capitalization F1 score as a range across test sets, without a further breakdown in the supplied material. For Turbo CTC, the announcement says speed increased while accuracy stayed the same, but the supplied information does not include the full benchmark table needed to examine that comparison. It also does not specify the hardware or deployment conditions behind the RTFx results. No customer deployment results or quantified effect on business workflows are provided.
Testing Before Deployment
IBM points prospective users to the models’ model cards for fuller evaluation results, architecture and training information. The next practical step for organizations considering the releases is to test them on audio resembling their own use case, including relevant languages, accents, noise levels and speaker patterns. That testing can help establish whether the published benchmarks translate to their recordings and what human review is needed.
The announcement does not specify a further release date or a scheduled independent evaluation. The models are described as available under Apache 2.0 with support in the named software tools; readers can check IBM’s model documentation for current versions and implementation details. Any future performance claims should be read alongside their test sets, comparison baselines and hardware conditions.
Key Questions
What did IBM announce?
IBM announced one new Granite Speech model and updates to two others: a multilingual 2B model for transcription and translation, a multilingual 470M Turbo CTC model, and revised 4.1 2B Plus weights focused on 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.
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 the model’s August launch. The cited comparison is based on nine public English short-form test sets and dates to October 2026; IBM says accuracy remained at the same level.
Are the reported benchmark results independently verified?
The supplied material presents the figures as IBM-reported evaluations and does not provide independent validation. Results on public test sets may not predict performance on every accent, recording environment or business workflow.
What license applies to the models?
IBM says all three releases use the Apache 2.0 license. The 5.0 2B model is listed as supported in Transformers, vLLM, llama.cpp and mlx-audio for Apple Silicon; IBM says the models have native support in Transformers.
Primary source: Hugging Face · via ThorstenMeyerAI.com
