Abstract
Speaker identification and verification via the measurement of formant frequencies from a sound spectrogram are often based on rules of thumb. In this Letter, we investigate the effectiveness of synchrosqueezed spectrograms, which can make formants clearer, for speaker verification using an experiment involving human participants. In the experiment, 88 participants were presented with pairs of synchrosqueezed spectrograms and asked to determine whether each pair was produced by the same speaker or different speakers. It was found that synchrosqueezing improved the performance of speaker discrimination by significantly reducing mistakes for different speakers and is thus useful for speaker verification using spectrograms.