IBM Granite Speech 5.0 Delivers Faster, More Accurate Multilingual ASR for Enterprise Workloads

IBM has refreshed its Granite Speech lineup with three models that span the enterprise speech stack: a new 2-billion-parameter flagship for premium accuracy, an updated 470-million-parameter Turbo CTC variant for high-throughput edge deployment, and an enhanced 2B-plus model purpose-built for speaker-attributed meeting and call transcription.

Granite Speech 5.0 2B: Accuracy, Robustness, and Control

The new granite-speech-5.0-2b succeeds the 4.1 2B model at the same parameter count. It transcribes English, French, German, Spanish, Portuguese, and Japanese, and translates to and from English across those languages plus English-to-Italian and English-to-Mandarin. Training leveraged 390,000 hours of public audio augmented with synthetic data for Japanese ASR, keyword biasing, and translation tasks.

Key advances over 4.1 include lower English word-error rates on short-form benchmarks, improved resilience to noise, reverberation, and far-field microphones, language-ID prompting to steer multilingual output, stronger long-form conversational translation, keyword lists up to 2,000 terms for domain-specific biasing, and optional punctuation and capitalization via prompt (capitalization F1 89.7–99.5 across test sets). The model ranks #2 among open-weight ASR models on average English WER (5.06%) per the Open ASR Leaderboard as of mid-October 2026.

Turbo CTC and 2B-Plus Updates

The revised granite-speech-5.0-470m-turboctc adds multilingual support, native punctuation and capitalization, and greater noise/reverb robustness while preserving its sub-real-time latency profile for on-device and high-throughput scenarios. Meanwhile, granite-speech-4.1-2b-plus gains improved speaker attribution, timestamp precision, keyword biasing, and long-speech stability for multi-participant recordings.

All three models ship under Apache 2.0 with first-class integration in Hugging Face Transformers, vLLM, llama.cpp, and mlx-audio for Apple Silicon, giving teams flexible deployment paths from cloud GPU clusters to edge devices.