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
Fixing some issues with chime7-task1 baseline #4925
Conversation
Codecov Report
@@ Coverage Diff @@
## master #4925 +/- ##
=======================================
Coverage 76.56% 76.56%
=======================================
Files 603 603
Lines 53753 53753
=======================================
Hits 41157 41157
Misses 12596 12596
Flags with carried forward coverage won't be shown. Click here to find out more. 📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more |
@popcornell About dipco, this fix works for me. Thank you! |
I got the following error at stage4
I guess it can be fixed by
However, I'm not sure why using an array variable i.e. |
I am using that variable because I want that when channels is undefined it is not passed so GSS uses all available ones. It runs fine on my side so this is strange. What does |
The following is enough for your case. It is not necessary to use array variable.
This script obviously can't work with "set -u" if Please check.
|
I am starting to add this pretrained model https://huggingface.co/popcornell/chime7_task1_asr1_baseline
Note that now i discarded sentences < 0.2 seconds (not many on dev, so it won't change much) so it is not true WER for the challenge. |
# ESPNet does not scale parameters with num of GPUs by default, doing it | ||
# here for you | ||
asr_batch_size=$(calc_int 128*$ngpu) # reduce 128 bsz if you get OOMs errors | ||
asr_max_lr=$(calc_float $ngpu/10000.0) | ||
asr_warmup=$(calc_int 40000.0/$ngpu) | ||
|
||
if [ $decode_only == 1 ]; then |
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.
if decode only i run data prep and GSS only on development sets
Indeed thanks Yoshiki ! |
LGTM |
Note that I will need to change the README.md too probably based on what we decide tomorrow about the channel selection stuff. |
Rules are not clear not if espnet/egs2/chime7_task1/asr1/run.sh Line 129 in ecf3bd3
The model remains the same (we want to discourage one model for each scenario) but you just select a bunch of microphones with some a priori knowledge for example of the topology of the array (e.g. linear in chime6 and i use only outer mics) |
OK, please make additional PRs. |
many thanks to Kamo Naoyuki @kamo-naoyuki .
I think there was a mistake on the close talk microphones for dipco dev (they were removed for eval).
I will also address your other useful comments but I am unsure about the sox one.