MeanFlowSE: one-step generative speech enhancement via conditional mean flow

Researchers have developed MeanFlowSE, a new generative AI model for speech enhancement that performs single-step inference instead of requiring multiple computational steps. The method achieves strong audio quality with substantially lower computational costs, making it suitable for real-time applications without needing knowledge distillation or external teachers.

• →MeanFlowSE eliminates the multistep inference bottleneck in generative speech enhancement by learning average velocity over finite intervals.
• →The single-step model achieves strong intelligibility, fidelity, and perceptual quality with significantly lower computational costs than multistep baselines.
• →The method requires no knowledge distillation or external teachers, simplifying the training process.
• →On VoiceBank-DEMAND dataset, the model demonstrates effectiveness for real-time speech enhancement applications.
• →The research code is open-sourced, enabling broader adoption and further development by the community.