Genio AI Forum Pinned Announcement: Feature Updates and Community Feedback

NeuroPilot
1

Hi! :dragon_face: :dragon: :t_rex:
This post serves as the official bulletin for all Genio AI forum news and updates.

  • All updates regarding new features, platform changes, and resources for Genio AI and MediaTek Neuropilot will be posted here as they become available. Please check this post regularly to stay informed about the latest developments.

  • Community members are encouraged to reply below with any feedback, suggestions, or wishes related to MediaTek Neuropilot—the Genio platform’s AI accelerator software stack. Whether you have ideas for new features, requests for improvement, or simple thoughts to share, every comment is valued and will help drive continuous innovation.

MediaTek reviews all submissions and considers community input when planning roadmap and documentation updates. Together, let’s advance the capabilities of MediaTek AI! :clinking_beer_mugs:

IoT AI Development Status

Last Update: 2025-02-26

ETA of upcoming milestones:

  • RC0 is currently scheduled to be ready in 2026/3/E

Changelog

Official Release V1 (2025/12/30)

Updates & Fixes:

  • Yocto v25.1 is released.
    • ONNX Runtime is now integrated into IoT Yocto:
      • ONNX Runtime runs on CPU on Genio 510/700/1200.
      • ONNX Runtime supports NPU hardware acceleration on Genio 520/720.
      • ONNX Runtime runs on CPU on Genio 510/700/1200.
    • ncc-tflite is EOL : The on-device Neuron SDK compiler ncc-tflite is now deprecated and removed from IoT Yocto images. We encourage customers to use the ncc-tflite compiler running on Ubuntu OS(x86_64).
  • IoT AI Hub is released.
    • It’s a centralised site that provides the platform/os/model information for the IoT Platform.

PR4 (2025/10/27)

Updates & Fixes:

  • NNStreamer demo script is now compatible to Genio 720/520 EVK.
  • OnnxRuntime is now supported on Genio 720/520 EVK with CPU/NPU accleration(partial).

Known Issues:

  • Bad performance over Stable Delegate on G720/G520
  • Misleading warning caused by Neuron Adapter (libcmdl/neon driver..)

PR2 (2025/09/02)

  • Enabled NPU on Genio 720/520 EVK.(TensorFlow Lite with hardware acceleration)
5

Update for IoT Yocto v25.1-dev AI Development Status

6

What ASR stack/model families do you recommend or support for best performance and compatibility with the Genio NPU software stack on IoT Yocto v25.1? Any reference workflows or supported-model guidance would be appreciated. I’m looking for a solution with good NPU offload, simple integration, and best real-time performance on Genio 520.

Does Whisper get good NPU offload with ONNX Runtime + NeuronExecutionProvider?

I also noticed MediaTek’s BreezeApp uses k2-fsa/sherpa-onnx for ASR/TTS. Is sherpa-onnx expected to perform well and be compatible with the Yocto v25.1 ONNX Runtime + NeuronExecutionProvider?

8

Hi Aaron,
Thank you for your inquiry.
For now, Genio Platforms supports ASR with the model-compiling toolkit that allows users to get the .dla models for deployment.
For the audio preprocessing/ post-processing or the complete demo, Intelligo provides this kind of service for application development.

Onnx Runtime with GAI support is still on the draft of our project plan. The model-supporting information will be published on a quarterly basis (Mar./June./Sept./Dec.) on IoT AI Hub :slight_smile: !