English Spoken Digits Database under noise conditions for research: SDDN

A. Ouisaadane (1) , Said Safi (1) , M. Frikel (2)

1 USMS - Université Sultan Moulay Slimane
2 LAC - Laboratoire d'automatique de Caen

In this paper, we introduce a modified database for English spoken digits under all types of noise conditions (SDDN). This database was designed for use in scientific research, especially in the field of speech enhancement, noise robustness, background noise, speech recognition, noise reduction, signal processing. It was synthesized by a set of open source spoken numbers of the public on the network, in particular from Speech Commands Dataset v0.02 released by Google's TensorFlow and AIYteams. Various types of real and artificial noise were added for a SNR from 10 dB to -10 dB to make them more suitable for reality. The AURORA, CHiME3 and NOISEX-92 databases were used to select noise types.

Mots clés: Databases; Signal to noise ratio; Noise measurement; Speech enhancement; Speech recognition; 1/f noise; Matlab; SDDN; SNR; Speech Commands Dataset; Noise

Domaines: Traitement du signal et de l'image [eess.SP]

A. Ouisaadane, Said Safi, M. Frikel. English Spoken Digits Database under noise conditions for research: SDDN. 2019 International Conference on Wireless Technologies, Embedded and Intelligent Systems (WITS), Apr 2019, Fez, Morocco. pp.1-5, ⟨10.1109/WITS.2019.8723698⟩. ⟨hal-02368315⟩