Feature #878
annotation specification for annotators (and pyin tweaking)
| Status: | New | Start date: | 2014-02-28 | |
|---|---|---|---|---|
| Priority: | Normal | Due date: | ||
| Assignee: |  Rachel Bittner | % Done: | 0% | |
| Category: | - | |||
| Target version: | - |
Description
This is mainly a request to Rachel: I assume your annotators will need a description of exactly how to annotate, including:
- can a pitch track be interrupted for 1 or 2 frames at a voiced consonant (n, v, m, ...)?
- does reverb still count as pitched?
- should weakly-pitched sounds (maybe noises that actually end up having a pitch) be annotated?
- should only be pitches annotated that the annotator can identify?
-- for me it would be really great to have that sort of thing, then I could tweak pyin to behave more like you want it.
Only slightly related question: do you assume you'll have nicely normalised audio (i.e. can I tell pyin that very low amplitude audio is very unlikely to be pitched?).
History
#1
Updated by Rachel Bittner about 10 years ago
I guess we've gotten to the philosophical questions!
Obviously there's no right answer, but I'd opt for:
*can a pitch track be interrupted for 1 or 2 frames at a voiced consonant (n, v, m, ...)? yes - I'd prefer the melody to generally align with note onsets, even if the onset isn't initially pitched
*does reverb still count as pitched? No
*should weakly-pitched sounds (maybe noises that actually end up having a pitch) be annotated? No
*should only be pitches annotated that the annotator can identify? --> can you elaborate? I'm not sure what you mean
To address the second point - we won't necessarily give it normalized audio, but it might be a good idea to normalize it when reading into Tony? Or alternatively, give pyin the normalized audio? It doesn't seem like it would hurt and it would at least give you consistent input.
#2
Updated by Matthias Mauch about 10 years ago
*should only be pitches annotated that the annotator can identify? --> can you elaborate? I'm not sure what you mean
I used your Bob 2 recording to test pyin a bit, and sometimes, especially at the ends of notes, the voice is so low in volume that I can't really tell what is happening -- maybe he's doing a very quick upward glide -- I can't tell. Should I just annotate no pitch then, or the pitch pyin suggests...
However, my main point was that, once you've got your instructions for the annotators, it would be great if you could share them with us. My specific questions were more or less made up on the spot to give you an idea what sort of things I mean.
#3
Updated by Rachel Bittner about 10 years ago
Ah ok I see what you mean. We haven't made a formal annotation protocol yet but we should have that early next week and I'll share it with you as soon as we've settled on something.