Sung-Lin Yeh
Sung-Lin Yeh
After carefully applying masking to the fbank and normalization, I found the encoder outputs vary with batch size. As @danpovey mentioned, the padding for cov layers can cause such difference....
> > @30stomercury could you take a look at that? I don't understand why the results vary with batch size here ... > > I think several reasons can cause...
Hi @GaetanLepage , I've solved it in #933. Please see the results I got below. It would be great if you can help us test the numbers as well. `batch_size=8`...
Great, @GaetanLepage thanks for the testing.
Thanks @TParcollet , I'm investigating kenlm and see if we can have BPE in ARPA format. I will add some to-dos later
I see, I will take a look.
Hi @mravanelli and @TParcollet, this is the current status of this pr. - [x] Support NgramScorer with kenlm. - [x] full scorer mode - [x] partial scorer mode - [x]...
@Gastron Sure, we can do it here. I will add it to the to-dos.
Hi @mravanelli , thank you for your suggestions. 1, 2. Currently I don't have a better way to manage importing part, I will finish the docstring first. 3. Yes, that...
Hi @mravanelli , I propose two ways to manage scorers in yaml. The second approach is similar to the way you suggested. ``` # Method1 scorer: !new:speechbrain.decoders.scorer.ScorerBuilder eos_index: !ref blank_index:...