SAWT restores damaged speech recordings to studio quality at 48 kHz, in any language, keeping the words and the voice of the recording. Each pair below is the same recording before and after. Every clip is levelled to the same loudness, so louder is not better. Headphones help.
Heard over a telephone line: 300 to 3700 Hz, line noise, the top two octaves gone. Quality predictor 3.05 to 4.60, bandwidth 3.7 to 23 kHz.
Damaged
Restored
A 48 kbps archive upload, band-limited at 22 kHz with a raised floor. Quality predictor 2.83 to 4.19, floor in pauses -39 to -61 dBFS, every phoneme matching the canonical text.
Damaged
Restored
Handheld recording outdoors: wind, traffic, a phone-grade microphone. Quality predictor 2.49 to 4.47.
Damaged
Restored
A telephone line: 300 to 3400 Hz, heavy compression, line noise. Quality predictor 1.17 to 4.56.
Damaged
Restored
A laptop webcam microphone in a small room, at its worst setting. Quality predictor 1.79 to 4.69.