Max Morrison

@justinsalamon we can further narrow the dependency requirement to just two functions (librosa.sequence.viterbi and librosa.sequence._viterbi) and a custom error class (librosa.util.exceptions.ParameterError). Those 3 have no other dependencies within librosa. I...

Hi @jongwook. I think my discussions with @justinsalamon today prompted this issue. I used librosa.sequence.viterbi to decode discrete emissions directly from the posterior observations emitted by the network. I tried...

A PR has been started! Regarding the theoretical justification, yes: the output of softmax (the posterior distribution) should be used and not the raw logits. As the network was trained...

Currently, files are downloaded into `C:\tmp\nussl`. PyTorch relies on either a explicit string argument or two environment variables. [Here](https://pytorch.org/docs/stable/_modules/torch/utils/model_zoo.html#load_url) is the relevant code for PyTorch. Keras creates the folder `C:\Users\\.keras`....

Sorry for the wait. I'm getting a `FileNotFoundError`: `FileNotFoundError: [Errno 2] No such file or directory: '/media/ext/models/deep_clustering_vocal_44k_long.model'`

Ah, so that's why you had me run the DeepSeparation script. That makes sense to me now. For reference, that script with DeepSeparation worked just fine on Windows, whereas DeepClustering...

The inference speed is dependent on the compute you are using. If you want comparable results to those given in the paper, make sure your compute is comparable to the...

You do not need to modify `torchcrepe`. `ljspeech` is not a dataset that I implemented. You should check how you setup that dataset and compare it to the examples provided...

Yep, you were both correct. `librosa` updated the return type of `librosa.sequence.viterbi` from `np.int64` to `np.uint16`. Fixed in `torchcrepe` version 0.0.16.

For a new dataset, you'll make a function in [download.py](https://github.com/descriptinc/cargan/blob/master/cargan/data/download.py). This function downloads and unzips the dataset, then prepares a cache of 22.05 kHz wav files. See the other examples...