-
Notifications
You must be signed in to change notification settings - Fork 2.9k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add figure for tuning & enrich the tuning section in doc #284
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Great! Thanks!
deep_speech_2/README.md
Outdated
``` | ||
|
||
- Tuning with CPU: | ||
|
||
```bash | ||
python tools/tune.py --use_gpu False | ||
``` | ||
The grid search will log the WER (word error rate) or CER (character error rate) at each point in the hyper-parameter space and their minima, and draw the error surface optionally. A proper hyper-parameters range should include the global minima of the error surface for WER/CER, as illustrated in the following figure. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
log-->print ?
remove and their minima
.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Done
deep_speech_2/README.md
Outdated
<br/>An example error surface for tuning on the dev-clean set of LibriSpeech | ||
</p> | ||
|
||
Usually, as the figure shows the variation of language model weight ($alpha$) mainly affect the performance of CTC beam search decoder. And a better procedure is first tuning on serveral data batches (the number can be specified) to find out the proper range of hyper-parameters, then change to the whole validataion set to carray out an accurate tuning. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
as the figure shows
--> as the figure shows,
is first tuning
--> is to first tune
mainly affect
--> significantly
alpha
--> $\alpha$
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Done
deep_speech_2/README.md
Outdated
|
||
Usually, as the figure shows the variation of language model weight ($alpha$) mainly affect the performance of CTC beam search decoder. And a better procedure is first tuning on serveral data batches (the number can be specified) to find out the proper range of hyper-parameters, then change to the whole validataion set to carray out an accurate tuning. | ||
|
||
After tuning, you can reset $\alpha$ and $\beta$ in the inference and evaluation modules to see if they really help improve the ASR performance. For more help |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
--> `$\alpha$ and
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Done
b54929d
to
626aaff
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Followed all comments. Thanks!
deep_speech_2/README.md
Outdated
``` | ||
|
||
- Tuning with CPU: | ||
|
||
```bash | ||
python tools/tune.py --use_gpu False | ||
``` | ||
The grid search will log the WER (word error rate) or CER (character error rate) at each point in the hyper-parameter space and their minima, and draw the error surface optionally. A proper hyper-parameters range should include the global minima of the error surface for WER/CER, as illustrated in the following figure. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Done
deep_speech_2/README.md
Outdated
<br/>An example error surface for tuning on the dev-clean set of LibriSpeech | ||
</p> | ||
|
||
Usually, as the figure shows the variation of language model weight ($alpha$) mainly affect the performance of CTC beam search decoder. And a better procedure is first tuning on serveral data batches (the number can be specified) to find out the proper range of hyper-parameters, then change to the whole validataion set to carray out an accurate tuning. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Done
deep_speech_2/README.md
Outdated
|
||
Usually, as the figure shows the variation of language model weight ($alpha$) mainly affect the performance of CTC beam search decoder. And a better procedure is first tuning on serveral data batches (the number can be specified) to find out the proper range of hyper-parameters, then change to the whole validataion set to carray out an accurate tuning. | ||
|
||
After tuning, you can reset $\alpha$ and $\beta$ in the inference and evaluation modules to see if they really help improve the ASR performance. For more help |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Done
Resolve #283