Add narrow band augmentation support #8806
Conversation
|
|
||
| # 2.c. On-the-fly Narrowband augmentation | ||
| if config.nb_augmentation: | ||
| nb_cuts = cuts.resample(config.sample_rate // 2) |
There was a problem hiding this comment.
I'm wondering if this SR shouldn't be hardcoded as 8000Hz since narrowband telephony is defined as 300Hz - 3400Hz, if somebody changes config.sample_rate it's not really going to go down to true narrowband.
|
jenkins |
jmayank1511
force-pushed
the
lhotse_agumentation_support
branch
from
228c86f
to
a4456cc
Compare
|
jenkins |
1 similar comment
|
jenkins |
| def __init__(self, sampling_rate, target_sampling_rate=8000, p=0.5, seed=0): | ||
| self.sampling_rate = sampling_rate | ||
| self.target_sampling_rate = target_sampling_rate | ||
| self.p = p |
There was a problem hiding this comment.
Double-check that p is valid, e.g., in (0,1)
| @@ -299,3 +310,24 @@ def _flatten_alt_text(cut) -> list: | |||
| text_instance.custom = {"text": data.pop("text"), "lang": data.pop("lang"), **data} | |||
| ans.append(text_instance) | |||
| return ans | |||
|
|
|||
|
|
|||
| class NarrowbandTransform: | |||
There was a problem hiding this comment.
This class is doing a resampling transform.
I'd suggest to consider naming it to something like ResamplingTransform.
|
|
||
| # 2.b. On-the-fly speed perturbation. | ||
| # mux here ensures it's uniformly distributed throughout sampling, | ||
| # and applying it here (before sampler/dataset) ensures optimal | ||
| # bucket allocation. | ||
| if config.perturb_speed: | ||
| cuts = CutSet.mux(cuts, cuts.perturb_speed(0.9), cuts.perturb_speed(1.1),) | ||
|
|
||
| # 2.c. On-the-fly Narrowband augmentation | ||
| if config.nb_augmentation: |
There was a problem hiding this comment.
I'd suggest naming this parameter downsampling augmentation, or something similar.
A band-pass augmentation (e.g., to simulate telephony frequency bandwidth) could be an additional augmentation, with corresponding lower and upper frequency.
|
|
||
| def __init__(self, sampling_rate, target_sampling_rate=8000, p=0.5, seed=0): | ||
| self.sampling_rate = sampling_rate | ||
| self.target_sampling_rate = target_sampling_rate |
There was a problem hiding this comment.
nit: target_sampling_rate may be a bit misleading, since the output signal will be at sampling_rate.
Consider using something like resampling_rate, which could fit well with ResamplingTransform
nit2: consider checking resampling rate is less than the input sample rate, and raise either a warning or even an error otherwise.
Signed-off-by: Mayank Jain <mayjain@nvidia.com>
for more information, see https://pre-commit.ci Signed-off-by: Mayank Jain <mayjain@nvidia.com>
Signed-off-by: Mayank Jain <mayjain@nvidia.com>
jmayank1511
force-pushed
the
lhotse_agumentation_support
branch
from
2b81941
to
6d6731e
Compare
for more information, see https://pre-commit.ci
|
This PR is stale because it has been open for 14 days with no activity. Remove stale label or comment or update or this will be closed in 7 days. |
|
This PR was closed because it has been inactive for 7 days since being marked as stale. |
What does this PR do ?
Add support for narrowband augmentations in lhotse dataloaders and fix small bug in noise augmentation
Add a one line overview of what this PR aims to accomplish.
Collection: [Note which collection this PR will affect]
Changelog
Usage
# Add a code snippet demonstrating how to use thisJenkins CI
To run Jenkins, a NeMo User with write access must comment
jenkinson the PR.Before your PR is "Ready for review"
Pre checks:
PR Type:
If you haven't finished some of the above items you can still open "Draft" PR.
Who can review?
Anyone in the community is free to review the PR once the checks have passed.
Contributor guidelines contains specific people who can review PRs to various areas.
Additional Information