contentvec
contentvec copied to clipboard
speech self-supervised representations
- Conform latest upstream repository. I only tested that run imports, load models, and check some values loaded in the Tensors. I do *not* confirmation of operation. Regards.
According to the paper, the contrastive loss weight is set to 1e-5 * num_train_steps. However, in the given config, the 'loss_weights: [10,1]', where 1 is the contrastive loss weight. May...
Hello, could you please share what model is used for creating of ContentVec_500 model ? I assume, that this pretrained Contentvec_500 model is not pretrained from scratch, but finetuned. I'm...
I looked at the bar plots of the speaker identification task on the SUPERB benchmark (Figure 3 in the paper) for the three models, hubert, hubert-iter and contentvec. However, I...
Hey guys! I am working on fine-tuning a model produced by ContentVec. Firstly, I'm unsure if this is possible, but I've reached a stage where I encounter an error related...