Add AudioDecoder.get_samples_played_in_range() public method
#555
Conversation
facebook-github-bot
added
the
CLA Signed
|
|
||
| def get_samples_played_in_range( | ||
| self, start_seconds: float, stop_seconds: Optional[float] = None | ||
| ) -> AudioSamples: |
There was a problem hiding this comment.
I feel like start_seconds should default to the stream's beginning, but that can be done later.
There was a problem hiding this comment.
If we do that, we should also mirror that in the video API. We could try to play the same trick that range() does, where the semantics of the first parameter changes based on the number of parameters, but I'd (softly) rather not do that.
There was a problem hiding this comment.
Agreed, I'll update #150 eventually with more of these desirable default behaviors. That would be a good onboarding PR.
| def __repr__(self): | ||
| return _frame_repr(self) | ||
|
|
||
|
|
There was a problem hiding this comment.
Below: the audio decoding API returns the new AudioSamples class rather than a pure Tensor. I think it has the following benefits:
- users can keep track of the sample_rate within that struct without having to handle it separately
- for some edge-cases, like with our mp3 test asset, the stream's beginning isn't 0. So we also returns
pts_seconds, which may not always be equal tostart_seconds.
There was a problem hiding this comment.
I think this makes sense, and it mirrors the video API. In the video API, __getitem__() returns just the tensor, but the named methods return Frame or FrameBatch. I think we should probably do that for audio as well.
| sample_rate: int | ||
|
|
||
| def __post_init__(self): | ||
| # This is called after __init__() when a Frame is created. We can run |
There was a problem hiding this comment.
| # This is called after __init__() when a Frame is created. We can run | |
| # This is called after __init__() when an AudioSamples instance is created. We can run |
| if (startSeconds == stopSeconds) { | ||
| // For consistency with video | ||
| return AudioFramesOutput{torch::empty({0}), 0.0}; | ||
| return AudioFramesOutput{torch::empty({0, 0}), 0.0}; |
There was a problem hiding this comment.
Drive-by, the video APIs return something of shape (0, C, H, W) (where C H W are from the metadata). Here, (0, 0) is the best we can do, at least it preserves the number of dimensions.
| output_pts_seconds = first_pts | ||
|
|
||
| num_samples = frames.shape[1] | ||
| last_pts = first_pts + num_samples / self.metadata.sample_rate |
There was a problem hiding this comment.
Won't last_pts be a float here? Is that your intention, or should we call round()?
There was a problem hiding this comment.
Yes it's a float. I didn't add the _seconds everywhere, but they're all in seconds. I can do it if you think it improves clarity.
We do call round() just below, which I think is enough? Or is there an edge-case I'm missing?
3 participants
Towards #549