Speech recognition Template on New Dataset #2152
Replies: 8 comments
-
Hi @monaabdelazim, |
Beta Was this translation helpful? Give feedback.
-
Hiv@mravanelli, |
Beta Was this translation helpful? Give feedback.
-
Hi,
just comment the environmental corruption part (
https://github.com/mravanelli/speechbrain-1/blob/develop/templates/speech_recognition/ASR/train.yaml#L131)
in the yaml if you don't want environmental corruption with your own
dataset.
…On Fri, 19 Nov 2021 at 16:09, monaabdelazim ***@***.***> wrote:
***@***.***,
I have already created the mentioned files for my dataset. But when I
start running the training using the mentioned command python train.py
train.yaml
It starts downloading the RIrs dataset.
I guess I need to modify a certain parameter in the .yaml file or adding a
certain attribute to the command but I am totally confused.can you help me
with that issue?
—
You are receiving this because you were assigned.
Reply to this email directly, view it on GitHub
<#1141 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AEA2ZVXIQCS276YWHUXDDQLUM24KJANCNFSM5IKXP3PQ>
.
|
Beta Was this translation helpful? Give feedback.
-
@mravanelli I did but I recieved another error.. value error: the reference "env_corrupt" is not valid. How to get rid of this error? |
Beta Was this translation helpful? Give feedback.
-
Right, you have to comment also the corresponding part in train.py (lines
142-145
https://github.com/speechbrain/speechbrain/blob/develop/templates/speech_recognition/ASR/train.py#L142
)
…On Fri, 19 Nov 2021 at 18:12, monaabdelazim ***@***.***> wrote:
@mravanelli <https://github.com/mravanelli> I did but I recieved another
error.. value error: the reference "env_corrupt" is not valid. How to get
rid of this error?
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
<#1141 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AEA2ZVRVS7JUTSFGPFVMQZLUM3KUZANCNFSM5IKXP3PQ>
.
|
Beta Was this translation helpful? Give feedback.
-
What about lines 167-169? Shall I comment them too? |
Beta Was this translation helpful? Give feedback.
-
Yes, all the entries with environmental augmentation (if, for some reason,
you don't want it)
…On Fri, 19 Nov 2021 at 18:18, monaabdelazim ***@***.***> wrote:
What about lines 167-169? Shall I comment them too?
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
<#1141 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AEA2ZVUEMKOLFGTQGMLKZADUM3LL5ANCNFSM5IKXP3PQ>
.
|
Beta Was this translation helpful? Give feedback.
-
@mravanelli I followed your instructions and this is my new question. How to provide the training command with the LM and tokenizer created using my dataset(i.e is there such a parameter to add to the training command?)? |
Beta Was this translation helpful? Give feedback.
-
I am trying to run the ASR template on my own data. I have generated the tokenizer and the language model as mentioned in the tutorial . when I start training the model the RIRS dataset starts downloading. How to convience the toolkit to start training on my own dataset and to generate the LM and tokenizer ckpt files?
The log files are attached.
log.txt
env.log
Beta Was this translation helpful? Give feedback.
All reactions