freevc uses pretrained speaker encoder, i.e., the speaker encoder is trained in advance on large datasets (e.g. voxceleb) with speaker verification task, while freevc-s does not. For their performance difference, please refer to table 1&2 in the paper.
freevc uses pretrained speaker encoder, i.e., the speaker encoder is trained in advance on large datasets (e.g. voxceleb) with speaker verification task, while freevc-s does not. For their performance difference, please refer to table 1&2 in the paper.