CoolFace
Apppublic

cisco-ai/stupase-speech-enhancement

sourceHugging Faceupdated 2mo agoView on Hugging Face
18likes
App README

StuPASE: Studio-Quality Generative Speech Enhancement

This Space demonstrates StuPASE, a state-of-the-art generative speech enhancement model that removes noise and reverberation while preserving linguistic content and speaker identity, achieving studio-level perceptual quality.

How it works

Upload a noisy or reverberant speech recording (16 kHz mono recommended). StuPASE processes it through three stages:

  1. 1.DeWavLM-R — Low-hallucination phonetic enhancement (fine-tuned from WavLM)
  2. 2.CFM — Phonetic-guided acoustic enhancement via conditional flow matching
  3. 3.Mel Vocoder — Reconstructs the enhanced waveform from mel features

Model