Speech Amplitude Envelope for 17 languages

This dataset contains Speech Amplitude envelope computed on audio files extracted from ‘www.faithcomesbyhearing.com'

We selected recordings which did not contain any sound effects but only plain speech.

language	nb of files	duration
ar: arabic	7224 files	6.02 hours
en: English	6078 files	5.06 hours
es: Spanish	6767 files	5.64 hours
eu: Basque	6573 files	5.47 hours
fi: Finnish	7393 files	6.16 hours
fr: French	5606 files	4.67 hours
hi: Hindi	6783 files	5.65 hours
hy: Armenian	6346 files	5.28 hours
ja: Japanese	7758 files	6.46 hours
ko: Korean	7172 files	5.97 hours
nl: Dutch	6611 files	5.50 hours
ru: Russian	9039 files	7.53 hours
sv: Swedish	5322 files	4.43 hours
ta: Tamil	10571 files	8.81 hours
th: Thai	8072 files	6.72 hours
vi: Vietnamese	6268 files	5.22 hours
zh: Chinese mandarin	5831 files	4.86 hours

languages: ISO 639-1

Audiofiles were then transform to mono, downsampled to 16KHz and cut in pieces of 3 seconds.

Speech Amplitude envelopes were computed on the audiofiles using Matlab code.

Speech Amplitude envelope data was downsampled to 512Hz.

Each pickle file contains Speech Amplitude envelopes data for the specified language sampled at 512Hz.

Script to compute the Speech Amplitude envelopes is available here

references:

Gilbert, G., & Lorenzi, C. (2006). The ability of listeners to use recovered envelope cues from speech fine structure. The Journal of the Acoustical Society of America, 119(4), 2438-2444.