Cog Av Hearing - Research Outputs - Datasets
Audiovisual ChiME3 CorpusA new AV-ChiME3 corpus is developed by combining Grid Corpus clean audio/video with ChiME3 noises for a wide-range of SNRs ranging from -12 to 12dB.
Primary Audiovisual DatasetThe main development of all strands of the project is taking place using the large audiovisual Grid Corpus, which was developed by project members from the University of Sheffield and has been very widely used. This corpus consists of a very large collection of audio waveform files and associated videos, with transcriptions and alignment data available.
Grid Corpus Homepage
Speech Performance evaluationTo evaluate the final approaches developed by our system, we make use of the data provided by the CHiME speech in noise challenge . This was a challenge from a number of years ago that made use of the Grid corpus, mixed with noise. It was widely used by a number of audio only speech processing systems, which therefore provides a benchmark to use. It is suitable as the Grid sentences also have associated visual data, and so we can compare.
CHiME Challenge Homepage