Key Findings
The granite-speech-5.0-2b model achieves an average English Word Error Rate (WER) of 5.06%, ranking it second among top open-weight ASR models as of October 21, 2026. For high-speed requirements, the granite-speech-5.0-470m-turboctc has reached 19,870 RTFx on the Open ASR Leaderboard, representing a 1.6x speed increase over its launch version. Additionally, the granite-speech-4.1-2b-plus update reduces Word Diarization Error Rate (WDER) and speaker-attributed WER (cpWER) by 7% each.
Technical Details
The 2B parameter model supports transcription for English, French, German, Spanish, Portuguese, and Japanese, with translation capabilities to and from English in those languages, plus English to Italian and Mandarin. It was trained on 390,000 hours of public audio and synthetic data. Key technical enhancements include:
- Keyword Biasing: Support for lists up to 2,000 terms to improve accuracy for product names and acronyms.
- Control: Language ID prompting for multilingual steering and request-based punctuation and capitalization (F1 range 89.7 to 99.5).
- Turbo CTC: A 470M-parameter encoder now supporting multilingual transcription (English, Japanese, Spanish, German, French, Italian, and Portuguese) with native punctuation.
- 4.1 2B Plus: Optimized for long-form speech, reducing WER on long-form test sets by 24% and cutting average timestamp shift (AAS) by more than half.
Background & Context
These models are designed for specific enterprise use cases: high-accuracy transcription and translation (5.0-2b), on-device or raw speed transcription (Turbo CTC), and multi-speaker recordings such as meetings or support calls (4.1-2b-plus). The models are released under the Apache 2.0 license.
Significance & Outlook
The collection provides native support for Transformers, vLLM, llama.cpp, and mlx-audio for Apple Silicon. For production deployments requiring risk detection, IBM recommends pairing these models with Granite Guardian.
Source: https://huggingface.co/blog/ibm-granite/granite-speech-5
Get our weekly technology intelligence — free
Receive an infographic that lets you judge at a glance whether each field’s analysis report is worth reading.
Subscribe Free — Weekly Tech Intelligence
By subscribing, you’ll receive Troy-Technical’s weekly technology intelligence newsletter.
- Your email and selected fields are used only to deliver the newsletter.
- We never share your information with third parties.
- You can unsubscribe anytime via the link in each email.
See our Privacy Policy for details.
Takes about a minute · Unsubscribe anytime
