Anuj Solanki
GSoC Week 02 Summary
Date: 03 Jun 2024 - 09 Jun 2024
Introduction
This week, I made significant progress in integrating Whisper AI into the AGL (Automotive Grade Linux) platform. My efforts included creating new recipes, modifying existing ones, and resolving various dependency issues. While I successfully compiled Whisper AI on the AGL image, some challenges remain, particularly with the transcription process. Overall, this week was productive in laying the groundwork for further development and optimization.
Tasks I have completed this week:
Created the following recipes:
python3-numba
llvmlite
openai-whisper
openai-whisper-tiny
- Modified the recipes for PyTorch from the meta-python-ai layer.
- Packaged all recipes into the meta-whisper layer.
- Successfully compiled and tested Whisper AI on the agl-ivi-demo-flutter image.
- Resolved some issues in agl-service-voiceagent, including recipes for openblas, lapack, and vosk-kaldi. However, issues with python3-scipy remain unresolved.
Issues Faced:
Faced a dependency issue with llvmlite: llvmlite requires LLVM 14, while AGL uses LLVM 18. I resolved this by using a precompiled wheel for llvmlite.
Next Week Tasks:
Resolve the remaining issues in agl-service-voiceagent.
Begin integrating Whisper AI with agl-service-voiceagent.
Conclusion
This week has been productive in terms of setting up the foundational recipes and addressing challenges.
Resources:
-