Tutorial Broken: "Stable Audio Open Small" dependencies are expired (tf-nightly, ai-edge-litert-nightly)

[watsonville-ai]

I am trying to follow the "Generate audio with Stable Audio Open Small on LiteRT" learning path, but the Python environment is impossible to build.

The requirements.sh script fails because the specific nightly packages it depends on from March 2025 have been deleted from PyPI.

Specifically, these packages cannot be found:

    tf-nightly==2.19.0.dev20250208

    ai-edge-litert-nightly==1.1.2.dev20250305

This makes the entire tutorial impossible to complete.

Request: Could you please either:

    Update the requirements.sh and Python scripts to use current, available packages?

OR

    Provide the final converted int8 quantized .tflite models for download? This would be the best solution and would allow users to skip the broken conversion step. The float32 models available from other users produce only noise, as the audiogen C++ app seems to expect int8 data.

Thank you.

[AhSinan]

Hey @watsonville-ai,

Thank you for raising this issue.

Have you tried to use a recent commit? because these dependencies have been updated:

pip install ai-edge-torch==0.4.0 \
    "tf-nightly>=2.19.0.dev20250208" \
    "ai-edge-litert-nightly>=1.1.2.dev20250305" \
    "ai-edge-quantizer-nightly>=0.0.1.dev20250208"

The C++ program has been updated to work with the recent changes to the export script, which may explain why you are getting noise.

Hope this helps,