sonique.Noise lab ↗

The listening room

Real-world speech.

Hear how Super Sonique v4 handles English speech from podcasts and YouTube. Switch between the original recording and the model output as you listen.

Recordings
100
Selection
Below MOS gate
EMA checkpoint
40,070

Selected recording

Loading recording…

SpectrogramOriginal
Frequency over time in the selected recording
Hz0 s

Loading audio…

About these clips

The library contains 50 podcast and 50 YouTube excerpts. Each failed its source manifest’s DNSMOSPro speech-quality gate of 3.0. The input MOS comes from scoring windows that overlap the excerpt. Voice activity and automatic transcription were checked on each clip.

The model output uses the eight-step Heun sampler. Both versions use the same playback gain and spectrogram scale. Press A or B while listening to switch at the same point in the recording.