Close Menu
    DevStackTipsDevStackTips
    • Home
    • News & Updates
      1. Tech & Work
      2. View All

      Top 10 Use Cases of Vibe Coding in Large-Scale Node.js Applications

      September 3, 2025

      Cloudsmith launches ML Model Registry to provide a single source of truth for AI models and datasets

      September 3, 2025

      Kong Acquires OpenMeter to Unlock AI and API Monetization for the Agentic Era

      September 3, 2025

      Microsoft Graph CLI to be retired

      September 2, 2025

      ‘Cronos: The New Dawn’ was by far my favorite experience at Gamescom 2025 — Bloober might have cooked an Xbox / PC horror masterpiece

      September 4, 2025

      ASUS built a desktop gaming PC around a mobile CPU — it’s an interesting, if flawed, idea

      September 4, 2025

      Hollow Knight: Silksong arrives on Xbox Game Pass this week — and Xbox’s September 1–7 lineup also packs in the horror. Here’s every new game.

      September 4, 2025

      The Xbox remaster that brought Gears to PlayStation just passed a huge milestone — “ending the console war” and proving the series still has serious pulling power

      September 4, 2025
    • Development
      1. Algorithms & Data Structures
      2. Artificial Intelligence
      3. Back-End Development
      4. Databases
      5. Front-End Development
      6. Libraries & Frameworks
      7. Machine Learning
      8. Security
      9. Software Engineering
      10. Tools & IDEs
      11. Web Design
      12. Web Development
      13. Web Security
      14. Programming Languages
        • PHP
        • JavaScript
      Featured

      Magento (Adobe Commerce) or Optimizely Configured Commerce: Which One to Choose

      September 4, 2025
      Recent

      Magento (Adobe Commerce) or Optimizely Configured Commerce: Which One to Choose

      September 4, 2025

      Updates from N|Solid Runtime: The Best Open-Source Node.js RT Just Got Better

      September 3, 2025

      Scale Your Business with AI-Powered Solutions Built for Singapore’s Digital Economy

      September 3, 2025
    • Operating Systems
      1. Windows
      2. Linux
      3. macOS
      Featured

      ‘Cronos: The New Dawn’ was by far my favorite experience at Gamescom 2025 — Bloober might have cooked an Xbox / PC horror masterpiece

      September 4, 2025
      Recent

      ‘Cronos: The New Dawn’ was by far my favorite experience at Gamescom 2025 — Bloober might have cooked an Xbox / PC horror masterpiece

      September 4, 2025

      ASUS built a desktop gaming PC around a mobile CPU — it’s an interesting, if flawed, idea

      September 4, 2025

      Hollow Knight: Silksong arrives on Xbox Game Pass this week — and Xbox’s September 1–7 lineup also packs in the horror. Here’s every new game.

      September 4, 2025
    • Learning Resources
      • Books
      • Cheatsheets
      • Tutorials & Guides
    Home»Development»Machine Learning»IBM Releases Granite 3.3 8B: A New Speech-to-Text (STT) Model that Excels in Automatic Speech Recognition (ASR) and Automatic Speech Translation (AST)

    IBM Releases Granite 3.3 8B: A New Speech-to-Text (STT) Model that Excels in Automatic Speech Recognition (ASR) and Automatic Speech Translation (AST)

    April 18, 2025

    As artificial intelligence continues to integrate into enterprise systems, the demand for models that combine flexibility, efficiency, and transparency has increased. Existing solutions often struggle to meet all these requirements. Open-source models may lack domain-specific capabilities, while proprietary systems sometimes limit access or adaptability. This shortfall is especially pronounced in tasks involving speech recognition, logical reasoning, and retrieval-augmented generation (RAG), where technical fragmentation and toolchain incompatibility create operational bottlenecks.

    IBM Releases Granite 3.3 with Updates in Speech, Reasoning, and Retrieval

    IBM has introduced Granite 3.3, a set of openly available foundation models engineered for enterprise applications. This release delivers upgrades across three domains: speech processing, reasoning capabilities, and retrieval mechanisms. Granite Speech 3.3 8B is IBM’s first open speech-to-text (STT) and automatic speech translation (AST) model. It achieves higher transcription accuracy and improved translation quality compared to Whisper-based systems. The model is designed to handle long audio sequences with reduced artifact introduction, enhancing usability in real-world scenarios.

    Granite 3.3 8B Instruct extends the capabilities of the core model with support for fill-in-the-middle (FIM) text generation and improvements in symbolic and mathematical reasoning. These enhancements are reflected in benchmark performance, including outperforming Llama 3.1 8B and Claude 3.5 Haiku on the MATH500 dataset.

    Technical Foundations and Architecture

    Granite Speech 3.3 8B uses a modular architecture consisting of a speech encoder and LoRA-based audio adapters. This design allows for efficient domain-specific fine-tuning while retaining the generalization capacity of the base model. The model supports both transcription and translation tasks, enabling cross-lingual content processing.

    The Granite 3.3 Instruct models incorporate fill-in-the-middle generation, supporting tasks such as document editing and code completion. Alongside, IBM introduces five LoRA adapters tailored for RAG workflows. These adapters support better integration of external knowledge, improving factual accuracy and contextual relevance during generation.

    A notable addition is adaptive LoRA (aLoRA), which reuses the key-value (KV) cache across inference sessions. This leads to a reduction in memory consumption and latency, particularly in streaming or multi-hop retrieval environments. aLoRA is designed to offer better trade-offs between computational overhead and performance in retrieval-heavy workloads.

    Benchmark Results and Platform Support

    Granite Speech 3.3 8B demonstrates superior performance over Whisper-style baselines in transcription and translation across multiple languages. The model performs reliably on extended audio inputs, maintaining coherence and accuracy without significant drift.

    In symbolic reasoning, Granite 3.3 Instruct shows improved accuracy on the MATH500 benchmark, outperforming comparable models at the 8B parameter scale. The RAG-specific LoRA and aLoRA adapters demonstrate enhanced retrieval integration and grounding, which are critical for enterprise applications involving dynamic content and long-context queries.

    IBM has made all models, LoRA variants, and associated tools open-source and accessible via Hugging Face. Additionally, deployment options are available through IBM’s watsonx.ai, as well as third-party platforms including Ollama, LMStudio, and Replicate.

    Conclusion

    Granite 3.3 marks a step forward in IBM’s effort to develop robust, modular, and transparent AI systems. The release targets critical needs in speech processing, logical inference, and retrieval-augmented generation by offering technical upgrades grounded in measurable improvements. The inclusion of aLoRA for memory-efficient retrieval, support for fill-in-the-middle tasks, and advancements in multilingual speech modeling make Granite 3.3 a technically sound choice for enterprise environments. Its open-source release further encourages adoption, experimentation, and continued development across the broader AI community.


    Check out the Model Series on Hugging Face and Technical details. Also, don’t forget to follow us on Twitter and join our Telegram Channel and LinkedIn Group. Don’t Forget to join our 90k+ ML SubReddit.

    🔥 [Register Now] miniCON Virtual Conference on AGENTIC AI: FREE REGISTRATION + Certificate of Attendance + 4 Hour Short Event (May 21, 9 am- 1 pm PST) + Hands on Workshop

    The post IBM Releases Granite 3.3 8B: A New Speech-to-Text (STT) Model that Excels in Automatic Speech Recognition (ASR) and Automatic Speech Translation (AST) appeared first on MarkTechPost.

    Source: Read More 

    Facebook Twitter Reddit Email Copy Link
    Previous ArticleRilasciata Ubuntu 25.04 “Plucky Puffin”: Arriva con GNOME 48 e kernel Linux 6.14
    Next Article OpenAI Releases a Practical Guide to Building LLM Agents for Real-World Applications

    Related Posts

    Machine Learning

    How to Evaluate Jailbreak Methods: A Case Study with the StrongREJECT Benchmark

    September 3, 2025
    Machine Learning

    Announcing the new cluster creation experience for Amazon SageMaker HyperPod

    September 3, 2025
    Leave A Reply Cancel Reply

    For security, use of Google's reCAPTCHA service is required which is subject to the Google Privacy Policy and Terms of Use.

    Continue Reading

    CVE-2025-6486 – TOTOLINK A3002R Stack-Based Buffer Overflow Vulnerability

    Common Vulnerabilities and Exposures (CVEs)

    CVE-2025-31189 – Apple macOS Sandbox Escalation

    Common Vulnerabilities and Exposures (CVEs)

    This tiny Bluetooth dongle gave me the ultimate in-flight movie experience

    News & Updates

    Why Your Automation Needs AI Decision-Making (And How Wordware Delivers)

    Development

    Highlights

    CVE-2025-48947: Session Cookies at Risk in Auth0 Next.js SDK

    June 4, 2025

    CVE-2025-48947: Session Cookies at Risk in Auth0 Next.js SDK

    A serious vulnerability has been uncovered in the widely-used Auth0 Next.js SDK—a library that helps developers implement authentication in their Next.js apps. Tracked as CVE-2025-48947 and rated CVSS …
    Read more

    Published Date:
    Jun 05, 2025 (3 hours, 14 minutes ago)

    Vulnerabilities has been mentioned in this article.

    CVE-2025-48947

    CVE-2025-29927

    CVE-2024-56332

    CVE-2024-51479

    CVE-2024-46982

    Introducing Amazon Bedrock AgentCore Identity: Securing agentic AI at scale

    August 15, 2025

    Threat Actors Allegedly Selling Baldwin Killer That Bypasses AV & EDR

    April 21, 2025

    CVE-2025-3601 – GitLab Denial of Service (DoS) in URL Processing

    August 27, 2025
    © DevStackTips 2025. All rights reserved.
    • Contact
    • Privacy Policy

    Type above and press Enter to search. Press Esc to cancel.