Tiny dataset as an example of labeling?

[jaggzh]

Can anyone provide a tiny dataset example (just one might even do) where we can see the ideal positioning of the start/stop times for our text?

I'm not sure:

1. How close I should go to the start of the audio waveform
2. How close at the end (some phonemes extend a bit, even though the word is already completed and understandable).

Here's an example of what I'm doing in audacity (pixelated spectrogram for patient privacy):