This repository contains the datasets list and scripts required for the ConferencingSpeech 2022 challenge. For more details about the challenge, please see our website.
baseline, this folder contains baseline system include inference model exported by inference scripts;eval, this folder contains evaluation scripts to calculate PLCC, RMSE and SRCC;Traing\Dev datasets, this folder contains training and development test datasets provied to the participant;Tencent Corpus, this dataset includes about 14,000 speech chinese speech clips with simulated (e.g. codecs, packet-loss, background noise) and live conditions.NISQA Corpus, the NISQA Corpus includes more than 14,000 speech samples with simulated (e.g. codecs, packet-loss, background noise) and live (e.g. mobile phone, Zoom, Skype, WhatsApp) conditions.IU Bloomington Corpus, there are 36,000 speech signals (18,000 each) extracted from COSINE and VOiCES datasets, each truncated between 3 to 6 seconds long. Note that the IU Bloomington corpus adopts ITU-R BS.1534 (MUSHRA) for subjective rating collection, which results in a score of 0-100 instead of 1-5. Thus, the IU Bloomington corpus will only be provided to participants as additional materials, and will NOT appear in this challenge as a evaluation test set. Participants can decide whether to use it according to their needs.PSTN Corpus, there are about 80,000 speech clips through classic public switched telephone networks, each truncated 10 seconds long.
To install requirements install Anaconda and then use:
conda env create -f envs.yml
This will create a new environment with the name "conferencingSpeech". Activate this environment to go on:
conda activate conferencingSpeech