Hi, first of all - brilliant idea, looks very exciting, thank you very much for sharing this.
I just wanted to reach out with a couple stumbling blocks I had on my first time playing around with it.
The very first dataset that I tried to load errored out with a FileNotFoundException. After digging, it turned out that it was missing the /data directory in the root of the application (~/TorchStudio). Adding that folder fixed it, so it seems like installation missed creating an initial empty /data directory.
Now I've loaded the dataset, CMUARCTIC from torchaudio.datasets, I'm seeing samples in the "Formatting" pane, and some of it is what I expect, but I'm not seeing everything.
CMUARCTIC returns samples as (Tensor, int, str, str): (waveform, sample_rate, transcript, utterance_id)
In the formatting pane I only see two Tensors that I can visualize: "Tensor 0" and "Tensor 1". The visual for Tensor 0 is clearly showing the waveform from the sample, and Tensor 2 is clearly showing the sample rate as a "label" style visualization.
How can I look at / visualize the final two strings from the sample? The 'transcript' and 'utterance_id' strings?
And also - can I create new Tensors from these sample properties, to use during training? Say, for example I want to create a 2d Fourier spectrum from the waveform, and use that as input? Or turn the string labels into a one-hot vector to use as a target output?