HAESUNG JEON

https://www.kaggle.com/datasets/bryanpark/korean-single-speaker-speech-dataset KSS dataset, above link, has CC BY-NC-SA 4.0, not allowed for commercial use. but repo has apache 2.0 license, allowed for commercial use. as owner of this repo and...

codes for "Reducing Barriers to Self-Supervised Learning: HuBERT Pre-training with Academic Compute?"

5

I've read paper "Reducing Barriers to Self-Supervised Learning: HuBERT Pre-training with Academic Compute" [link](https://arxiv.org/pdf/2306.06672.pdf). and want to how pre-processing optimization is implemented. in Section 2.3 of paper. it's bit hard...

while other metrics accept (estimates, reference) in function call, SISDR takes (reference, estimates) in function call which makes confusing and easy-to-mistake. need change! https://github.com/descriptinc/audiotools/blob/49b8b6bb9259a99bda5f2138a1a4df055dd172ef/audiotools/metrics/distance.py#L83-L91

Thanks for releasing w2v-bert 2.0! i have two questions 1. is released checkpoint fine-tuned on seamless? or checkpoint trained on only self-supervised objective? 2. is there any data statistics to...

https://github.com/bshall/hifigan/blob/374a4569eae5437e2c80d27790ff6fede9fc1c46/hifigan/generator.py#L12-L16 file name of released models seem to be swapped. soft refers to hubert-discrete, discrete refers to hubert-soft actually it works good despite of that, are these link proper?

is pretrained weight of discriminator of base model available?

3

Thanks for nice work. @bshall I'm trying to train hifigan now, but it takes so long training it from scratch using other dataset. If discriminator of base model is also...