← Back
Hugging Face
IBM releases Granite 4.0 1B Speech with 50% fewer parameters, achieves #1 on OpenASR leaderboard
· releasemodelfeatureopen-sourceperformance · huggingface.co ↗

Granite 4.0 1B Speech Released

IBM's Granite 4.0 1B Speech is now available, representing a significant advancement in compact speech recognition technology. This model is purpose-built for enterprise applications on resource-constrained devices, offering multilingual automatic speech recognition (ASR) and bidirectional speech translation (AST) capabilities.

Key Improvements and Features

Model Efficiency and Performance:

  • 50% fewer parameters than the predecessor (Granite 3.3-2B) while delivering higher English transcription accuracy
  • Faster inference through speculative decoding optimization
  • Strong performance across standard ASR benchmarks as measured by Word Error Rate (WER)
  • Ranked #1 on the OpenASR leaderboard among open speech recognition systems

Language and Capability Expansion:

  • Expanded language support now covers: English, French, German, Spanish, Portuguese, and Japanese
  • New Japanese ASR support addresses community demand
  • Keyword list biasing for improved recognition of names and acronyms—a frequently requested feature

Availability and Licensing

The model is released under an Apache 2.0 license with native support in transformers and vLLM, making it accessible for both research and production use. IBM recommends pairing the model with Granite Guardian 3.3-8B for production deployments requiring additional risk detection capabilities.

Full evaluation results, architecture details, training data, and usage examples are available on the model card. Developers can start using Granite 4.0 1B Speech immediately for edge deployment scenarios.