Introduction to Granite 4.0 1B Speech

Meanwhile, the latest addition to IBM’s Granite Speech collection, Granite 4.0 1B Speech, is making waves in the industry. Additionally, this compact speech-language model is designed for enterprise applications on resource-constrained devices. For example, it is built for multilingual automatic speech recognition (ASR) and bidirectional speech translation (AST).

Key Features of Granite 4.0 1B Speech

However, what sets Granite 4.0 1B Speech apart is its ability to deliver higher English transcription accuracy, faster inference through speculative decoding, and expanded language support. Furthermore, it now covers English, French, German, Spanish, Portuguese, and Japanese. Therefore, two new additions in this release are Japanese ASR support and keyword list biasing for improved recognition of names and acronyms.

Additionally, Granite 4.0 1B Speech recently ranked #1 on the OpenASR leaderboard, highlighting its strong performance among open speech recognition systems. Meanwhile, despite its small size, Granite 4.0 1B Speech achieves highly competitive results on standard English ASR benchmarks.

Performance and Accuracy

For instance, performance is measured using Word Error Rate (WER)—the percentage of words transcribed incorrectly—where lower scores indicate better accuracy. However, as shown in Chart 1, Granite 4.0 1B Speech delivers strong WER across multiple datasets while using far fewer parameters than many comparable models.

Finally, we recommend pairing it with Granite Guardian for production deployments requiring additional risk detection. Therefore, give it a try today and let us know what you think!