IBM refreshes Granite Speech lineup: new ASR model, quicker edge transcription, sharper speaker attribution
In short: IBM rolled out three updates to its Granite Speech family: a brand-new 2B speech-language model (granite-speech-5.0-2b), a revised 470M Turbo CTC model built for speed, and an improved 4.1-2B-Plus model tuned for multi-speaker recordings. Together they target high-accuracy transcription/translation, fast on-device recognition, and accurate speaker/timestamp labeling. All three remain under Apache 2.0.
This summary was generated automatically by AI from Hugging Face's publication. It is our own text, not a copy of the original — facts, figures and quotes belong to the source, linked above and below.
What changed?
- 1granite-speech-5.0-2b adds language-ID prompting, up to 2,000-term keyword biasing, stronger noise/reverb handling, and expanded translation (English paired with French, German, Spanish, Portuguese, Japanese, plus English-to-Italian and English-to-Mandarin); capitalization F1 ranges 89.7-99.5; it no
- 2Turbo CTC moved from English-only to seven languages (adding Japanese, Spanish, German, French, Italian, Portuguese), gained native punctuation/capitalization, handles reverberant/far-field audio better, and became roughly 1.6x faster — throughput rose from about 12,600 to 19,870 RTFx — without losi
- 34.1-2B-Plus keeps the same architecture, prompts and API (drop-in upgrade) while cutting diarization errors: WDER and cpWER each fall 7%, timestamp shift (AAS) is more than halved, long-form WER drops 24%, and keyword-biasing errors fall 45% on 20-term lists
| Parameter | Before | Now |
|---|---|---|
| granite-speech-5.0-2b WER | Not specified (succeeds granite-speech-4.1-2b) | 5.06% average English WER, #2 open-weight ASR on leaderboard |
| Turbo CTC speed (RTFx) | ~12,600 RTFx at launch | 19,870 RTFx (roughly 1.6x faster at same accuracy) |
| Turbo CTC languages | English only | English, Japanese, Spanish, German, French, Italian, Portuguese |
| 4.1-2B-Plus WDER/cpWER | WDER 4.40, cpWER 30.50 | WDER 4.10, cpWER 28.48 (7% drop each) |
| 4.1-2B-Plus long-form WER | 13.53 | 10.22 (24% lower) |
| 4.1-2B-Plus keyword biasing WER (20-term) | 11.79 | 6.48 (45% lower) |
| 4.1-2B-Plus timestamp shift (AAS, all speech) | 210.86 | 98.54 |
Why it matters
These updates widen the gap on three practical fronts enterprises care about: raw transcription accuracy in noisy conditions, inference speed for on-device or high-volume pipelines, and reliability of who-said-what labeling in multi-speaker calls — all common pain points for contact-center and meeting-transcription workloads.
What it means for AI agents and contact centers
For teams running phone-based voice agents and contact-center analytics, the keyword biasing, noise robustness and faster edge inference in these open models are worth testing against current STT pipelines, particularly for throughput and multi-speaker call QA — though note Lithuanian is not among the languages the source lists as supported.
🧪 Worth Testing
Open-weight models with concrete gains in noise robustness, keyword biasing, speed, and speaker/timestamp accuracy make for a solid benchmark against existing STT setups for call transcription and meeting analysis.
Sources
- Hugging FaceOfficialPrimary sourceOriginal article →„Granite Speech 5.0: Transcribe, Translate, and Attribute more Accurately and Faster“21 Oct 2026, 17:58
- Published by source
- 21 Oct 2026, 17:58
- Found by our system
- 21 Oct 2026, 18:19
- Summary generated
- 21 Oct 2026, 18:21
This article was written by AI from the original source. Facts, numbers and prices come from the source; missing values are marked “Not specified”. Legal notice, copyright and privacy