Recently, the Hailo team compiled the Whisper models to Hailo8 and Hailo8L devices, opening new avenues for voice-based application development. Building on top of this, we are excited to share DeGirum/degirum_speech, the PySDK version of Whisper integration.
DeGirum Speech has been designed to make development easier with these features:
-
High-performance Whisper transcription on Hailo-8/Hailo-8L accelerators.
-
Transcribe from files, NumPy arrays, or a live microphone stream.
-
Optional audio denoising for noisy environments.
-
Optional Voice Activity Detection (VAD) to ignore silence.
-
Simple CLI to download models for offline use.
Looking forward to community’s feedback on this exciting package.