Degraded Librispeech

Degraded Librispeech includes 34110 degraded speech samples obtained from 2650 clean speech sources extracted from Librispeech. DEGRADATIONS Background noise 0, 8, 15, 25, 40 dB Clipping 5, 10, 25, 40, 60 % of waveform samples Opus 8, 16, 32, 64, 128 kbps mp3 8, 16, 32, 64, 128 kbps DETAILS Loudness is normalized using EBU R 128. You can extract labels regarding the degradation type and intensity levels from the filenames. Degraded Librispeech has been used to develop NOMAD (Non-matching audio distance), a non-matching reference audio quality metric. NOMAD can also be used as a waveform generation loss function to improve speech quality, e.g., speech enhancement. References: NOMAD metric (ICASSP 2024): Ragano, Alessandro, Jan Skoglund, and Andrew Hines: NOMAD: Unsupervised Learning of Perceptual Embeddings for Speech Enhancement and Non-matching Reference Audio Quality Assessment [Paper],[Code] Original Librispeech dataset: Panayotov, Vassil, Guoguo Chen, Daniel Povey, and Sanjeev Khudanpur. "Librispeech: an asr corpus based on public domain audio books." In 2015 IEEE international conference on acoustics, speech and signal processing (ICASSP), pp. 5206-5210. IEEE, 2015.