Interfaze Ships diffusion-gemma-asr-small, an Open-Source Diffusion ASR Model Transcribing Six Languages via DiffusionGemma’s Parallel Denoising Decoder

Short summary

Interfaze released diffusion-gemma-asr-small, an open-source multilingual ASR model that uses diffusion-based parallel denoising decoding instead of autoregression for speech transcription. The model adds a ~42M-parameter adapter to Google's frozen DiffusionGemma, supporting six languages with transcription cost determined by denoising steps rather than transcript length.

• •Open-source multilingual ASR model shipped by Interfaze
• •Uses diffusion-based decoding architecture instead of autoregressive approach
• •Single adapter covers six languages with step-based cost model

Generated with AI, which can make mistakes.