← Back
Hugging Face
IBM releases Granite 4.0 1B Speech model with 50% fewer parameters and multilingual support
· releasemodelfeatureopen-source · huggingface.co ↗

Granite 4.0 1B Speech Released

IBM has announced Granite 4.0 1B Speech, a lightweight automatic speech recognition (ASR) and bidirectional speech translation (AST) model designed for enterprise applications on edge devices. The model reduces parameter count by 50% compared to its predecessor while improving performance across multiple benchmarks.

Key Improvements

  • Reduced model size: Down from 2B to 1B parameters, enabling deployment on resource-constrained devices
  • Higher accuracy: Delivers better English transcription accuracy (measured by Word Error Rate) despite smaller size
  • Expanded language support: Now covers English, French, German, Spanish, Portuguese, and Japanese
  • New features: Includes Japanese ASR support and keyword list biasing for improved recognition of proper nouns and acronyms
  • Faster inference: Supports speculative decoding for optimized performance

Performance & Availability

The model currently ranks #1 on the OpenASR leaderboard among open-source speech recognition systems. It is released under the Apache 2.0 license with native support in Hugging Face transformers and vLLM frameworks. Comprehensive evaluation results, architecture details, and usage examples are available on the model card.

Recommended Deployment

For production deployments requiring additional safety features, IBM recommends pairing Granite 4.0 1B Speech with Granite Guardian for risk detection capabilities. The model is immediately available on Hugging Face and ready for deployment.