Accelerating the search for antimicrobial peptides with Google Antigravity
deepmind.google·6d ago
AI Engineering
Hugging Face Blog·10h ago·5 min read
Granite Speech 5.0: Transcribe, Translate, and Attribute more Accurately and Faster
TL;DR
IBM's Granite Speech 5.0 introduces a new speech-language model and updates existing models, enhancing transcription and translation accuracy and speed across multiple languages.
✦ Why It Matters
Evaluate and integrate Granite Speech 5.0 models to enhance your enterprise transcription and translation workflows today.
Key Takeaways
- Use granite-speech-5.0-2b for high accuracy in transcription and translation.
- Choose granite-speech-5.0-470m-turboctc for fast, on-device transcription.
- Opt for granite-speech-4.1-2b-plus for improved multi-speaker recording handling.
- Incorporate keyword biasing to enhance transcription accuracy for specific terms.
Full Summary
Granite Speech 5.0 features the new granite-speech-5.0-2b model and updates to granite-speech-5.0-470m-turboctc and granite-speech-4.1-2b-plus. The new model offers enhanced automatic speech recognition (ASR) with improved accuracy, noise robustness, and multilingual capabilities, supporting languages like English, French, and Japanese.
It was trained on 390,000 hours of audio data, achieving a word error rate (WER) of 5.06%, ranking it among the top ASR models. The Turbo CTC update focuses on speed and now supports multiple languages with native punctuation.
The granite-speech-4.1-2b-plus model improves speaker attribution and timestamp accuracy, making it ideal for multi-speaker environments. These advancements cater to enterprise needs for high-quality transcription and translation.