Add new speech command recognition tutorial #1204
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Deploy preview for pytorch-tutorials-preview ready! Built with commit d07e292 https://deploy-preview-1204--pytorch-tutorials-preview.netlify.app |
Closed
8kHz with M5: Accuracy: 58%, 77% |
4kHz with M5: Accuracy: 60%, 70% |
3kHz with M5: Accuracy: 59%, 64% |
8kHz, n_channels=64: Accuracy 72%, 67% |
8kHz, stride=8: Accuracy 69%, 79% |
8kHz, stride=16: Accuracy: 68%, 73% |
8kHz, n_channel=32, Accuracy: 63%, 71% |
8 kHz, n_channel=32, stride=16: Accuracy 62%, 71% |
8kHz, n_channel=16, stride=16, Accuracy: 54%, 64% |
8kHz, simple linear, MFCC, Accuracy: 13%, 15% :) |
vincentqb
changed the title
vincentqb
changed the title
|
Follow-up to this one with more epochs :) 8kHz, stride=16, epoch=21, accuracy=86% |
Loop only with batch resample on GPU, CPU, num_workers=1, pin_memory=True, 1:22Loop only, CPU, num_workers=1, pin_memory=True, 5:50Linear, CPU, num_workers=0, pin_memory=False, 5:44Linear, GPU, num_workers=0, pin_memory=False, 5:18Linear, CPU, num_workers=1, pin_memory=True, 5:55Linear, GPU, num_workers=1, pin_memory=True, 4:59Linear, CPU, num_workers=0, pin_memory=False, 5:44Linear, CPU, SGD, num_workers=0, pin_memory=False, 5:58Linear, GPU, SGD, num_workers=0, pin_memory=False, 5:58 |
CPU Model, CPU Resample, 3:37GPU Model, CPU Resample, 2:10GPU Model, GPU Resample, 1:26 |
scheduler step at 20, accuracy=86% after 21 epochs, 25.44s/itno scheduler, accuracy=81% after 21 epochs |
vincentqb
force-pushed
the
speechtutorial
branch
6 times, most recently
from
2568d5e to
0b882f1
Compare
mthrok
reviewed
Nov 6, 2020
| information from executed cells disappear). | ||
|
|
||
| First, let’s import the common torch packages such as | ||
| ``torchaudio <https://github.com/pytorch/audio>``\ \_ that can be |
There was a problem hiding this comment.
Looks like the link format should be markdown, not RST format.
https://colab.research.google.com/notebooks/markdown_guide.ipynb#scrollTo=70pYkR9LiOV0
There was a problem hiding this comment.
Indeed, the conversion from colab to rst didn't render correctly here. Thanks for pointing it out! The tutorial is rendered first on pytorch.org, so I see other tutorials using rst links, e.g. here.
intermediate_source/speech_command_recognition_with_torchaudio.py
Outdated
Show resolved
Hide resolved
| # Let’s find the list of labels available in the dataset. | ||
| # | ||
|
|
||
| labels = sorted(list(set(datapoint[2] for datapoint in train_set))) |
There was a problem hiding this comment.
[not PR review comment] I wonder if the Dataset implementations should have this kind of attributes.
There was a problem hiding this comment.
(added mention in #910)
| # | ||
| # The actual loading and formatting steps happen in the access function | ||
| # ``__getitem__``. In ``__getitem__``, we use ``torchaudio.load()`` to | ||
| # convert the audio files to tensors. ``torchaudio.load()`` returns a |
There was a problem hiding this comment.
Is it worth mentioning the internal mechanism about torchaudio.load, which can be changed anytime in the future?
There was a problem hiding this comment.
My impression was that some details would help the user understand. I agree though this might be too detailed, especially since those are internal details. I've replaced this by the following to instead emphasize that there exists a torchaudio.load function in case someone just wants to load a file instead of using a dataset.
The actual loading and formatting steps happen when a data point is being accessed, and torchaudio takes care of converting the audio files to tensors. If one wants to load an audio file directly instead,
torchaudio.load()can be used. It returns a tuple containing the newly created tensor along with the sampling frequency of the audio file (16kHz for SpeechCommands).
intermediate_source/speech_command_recognition_with_torchaudio.py
Outdated
Show resolved
Hide resolved
intermediate_source/speech_command_recognition_with_torchaudio.py
Outdated
Show resolved
Hide resolved
4 tasks
vincentqb
changed the title
vincentqb
force-pushed
the
speechtutorial
branch
from
e9bc491 to
d07e292
Compare
4 tasks
2 tasks
rodrigo-techera
pushed a commit
to Experience-Monks/tutorials
that referenced
this pull request
Nov 29, 2021
* add new speech tutorial. * update with a few parameter tuned. model takes less than 10 min to run now. * feedback. * improve GPU performance. add interactive demo at the end. * feedback.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Add command recognition tutorial, see colab.
1h10 min using this5 min.SpeechCommands as it is currently in torchaudio doesn't have the train/valid/test split. We could decide to add that in before the release to simplify the tutorial. Add SpeechCommands train/valid/test split audio#966.Experiment with MFCC instead of resample only.Replace SPEECHCOMMANDS by YESNO dataset, and usetorchaudio.transforms.vadto segment audio into utterances.This pull request complements #572, see also deprecated tutorial.
cc @brianjo