🔍 Read the full analysis: How To Put Granite Speech 5.0 To Work For Transcription And Translation on ThorstenMeyerAI.com
TL;DR
IBM has announced a new 2-billion-parameter Granite Speech model for multilingual transcription and translation, alongside updates to two existing models. The company reports gains in speed, word error rate, speaker attribution and timestamps, but the cited results come from IBM’s evaluations and public benchmarks.
IBM has announced three updates to its Granite Speech collection: a new 2-billion-parameter multilingual 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 target different audio workloads; the performance figures cited so far are IBM-reported results, not independent validation.
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. It also translates from English into Italian and Mandarin. According to the company, training used 390,000 hours of public audio, plus synthetic data for Japanese speech recognition, keyword biasing and translation.
IBM says the 5.0 model is better than its predecessor at handling noise, reverberation and distant microphones. It accepts prompts specifying the input language and keyword lists of up to 2,000 terms. 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 revised granite-speech-5.0-470m-turboctc adds transcription in Japanese, Spanish, German, French, Italian and Portuguese alongside English, as well as native punctuation and capitalization. IBM reports a 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 updated granite-speech-4.1-2b-plus is aimed at multi-speaker and long-form recordings. 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 its architecture, prompt format and API are unchanged.
Three Models, Three Audio Workloads
The updates give organizations options for distinct speech tasks rather than asking one model to serve every need. Translation and multilingual accuracy, high-volume transcription speed, and speaker identification with precise timing can require different capabilities. IBM is positioning the 5.0 2B model for multilingual transcription and translation, Turbo CTC for speed, and the 4.1 2B Plus update for recordings where speaker labels and timestamps matter.
Those features can affect what teams do with transcripts after recognition. More accurate timestamps may help align captions or locate sections of audio, while speaker labels can make meeting and support-call records easier to review. Keyword lists may help with names, product terms and acronyms. These are plausible applications of the announced features, but the source material gives no measurements from customer deployments showing how much they improve a particular workflow.
The benchmark figures offer a comparison point, not a guarantee of results on a company’s own audio. Teams will need to test against recordings representative of their languages, accents, microphones, background noise and specialist vocabulary. They should also account for the cost of running models and any human review needed before transcripts or translations are used.
transcription software with speaker labeling
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How the Granite Models Differ
The Granite Speech collection covers speech recognition and related tasks. IBM introduced Turbo CTC in August as an English-only, 470-million-parameter encoder designed for speed. The October update keeps the model size and adds six transcription languages, along with punctuation and capitalization. The new 5.0 2B model takes a different path, adding multilingual translation capabilities at a larger parameter count.
IBM’s reported comparisons use different measures. Word error rate measures transcription errors, with lower scores indicating better results; RTFx is a measure of processing speed, with higher scores indicating faster processing. The company says 5.0 2B placed second among open-weight ASR models for average English word error rate, at 5.06%. IBM describes this as unofficial, based on public English short-form test sets on the Open ASR Leaderboard as of October 15, 2026. The Turbo CTC speed comparison is also based on nine public English short-form test sets and is dated October 2026.
IBM says all three releases have an Apache 2.0 license and native support in Transformers. The company directs users to the model cards for evaluation results, architecture and training details, and recommends pairing the models with Granite Guardian for production deployments that need additional risk detection.
“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.
What the Evaluations Do Not Show
The performance claims in the supplied material come from IBM’s own announcement and evaluations; no independent validation is provided. Public short-form benchmark results do not establish how the models perform across all accents, language varieties, recording conditions or business workflows. The source material does not provide enough detail to assess performance on overlapping speech, specialist vocabulary or noisy recordings outside the described tests.
Some results also lack a full breakdown in the material provided. IBM gives the 5.0 2B capitalization F1 as a range across test sets but does not specify the score for each set here. It describes Turbo CTC’s accuracy as unchanged while reporting higher speed, but the full benchmark table is not included. The cited RTFx result is not accompanied by the hardware or deployment conditions used to produce it. Customer results and real-world operating costs are also not reported.
As an affiliate, we earn on qualifying purchases.
Model Cards and Deployment Testing
IBM points prospective users to the model cards for fuller evaluation results, architecture and training details. The next practical step for organizations considering deployment is to test the relevant model on audio similar to their own, including the languages, accents, recording environments and terminology they expect it to handle. That testing can help determine whether benchmark results translate to their needs and what review process is appropriate.
The supplied source material does not give a date for a further release or an independent evaluation. For now, the announced models, their published documentation and results from organization-specific trials are the clearest basis for judging performance. Whether the reported gains hold across wider production use remains to be established.
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 for Granite Speech?
IBM announced one new model and updates to two existing models: a multilingual 2B model for transcription and translation, a faster multilingual 470M Turbo CTC model, and revised weights for 4.1 2B Plus focused on speaker attribution and timestamps.
Which languages does Granite Speech 5.0 2B support?
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 into Italian and Mandarin.
What improvements does the updated Turbo CTC model claim?
IBM says the 470M Turbo CTC update adds six transcription languages beyond English, along with punctuation and capitalization. The company reports 19,870 RTFx on the Open ASR Leaderboard, compared with about 12,600 at its August launch, while stating that accuracy remained unchanged.
Are the reported results independently verified?
The supplied material does not include independent validation. IBM’s figures are based on its evaluations and public benchmarks, which may not predict performance on every accent, recording environment or business workflow.
Where can developers find the models and technical details?
IBM says the releases use the Apache 2.0 license and are supported in Transformers. The 5.0 2B model is also supported in vLLM, llama.cpp and mlx-audio for Apple Silicon. IBM directs users to the model cards for technical and evaluation details.
Primary source: Hugging Face · via ThorstenMeyerAI.com
