-
For a project that I'm working on that uses Lhotse for the data pipeline, I want to extend Lhotse's sampler to apply some additional constraints on the batches (mainly to enforce that each batch should have cuts from the same speaker). What would be the most straight-forward way to implement this? |
Beta Was this translation helpful? Give feedback.
Answered by
pzelasko
Oct 25, 2022
Replies: 1 comment 1 reply
-
I suggest to use RoundRobinSampler that consists of a list of DynamicBucketingSampler instances, one for each speaker. In case the speakers have a very small amount of data, use DynamicCutSampler instead. Would that work? |
Beta Was this translation helpful? Give feedback.
1 reply
Answer selected by
desh2608
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
I suggest to use RoundRobinSampler that consists of a list of DynamicBucketingSampler instances, one for each speaker. In case the speakers have a very small amount of data, use DynamicCutSampler instead. Would that work?