From c24b6103896f58f09c65f0452b0cd7d0fb3ff0e3 Mon Sep 17 00:00:00 2001 From: Fangjun Kuang Date: Wed, 22 Dec 2021 23:47:24 +0800 Subject: [PATCH] Minor fixes. --- .github/workflows/run-pretrained-transducer-stateless.yml | 2 +- README.md | 2 +- egs/librispeech/ASR/transducer_stateless/decoder.py | 3 +++ 3 files changed, 5 insertions(+), 2 deletions(-) diff --git a/.github/workflows/run-pretrained-transducer-stateless.yml b/.github/workflows/run-pretrained-transducer-stateless.yml index 7b08dd470..7af2299a4 100644 --- a/.github/workflows/run-pretrained-transducer-stateless.yml +++ b/.github/workflows/run-pretrained-transducer-stateless.yml @@ -24,7 +24,7 @@ on: types: [labeled] jobs: - run_pre_trained_conformer_ctc: + run_pre_trained_transducer_stateless: if: github.event.label.name == 'ready' || github.event_name == 'push' runs-on: ${{ matrix.os }} strategy: diff --git a/README.md b/README.md index f0ead2779..931fb0198 100644 --- a/README.md +++ b/README.md @@ -80,7 +80,7 @@ We provide a Colab notebook to run a pre-trained RNN-T conformer model: [![Open Using Conformer as encoder. The decoder consists of 1 embedding layer and 1 convolutional layer. -The best WER with beam search with beam size 4 is: +The best WER using beam search with beam size 4 is: | | test-clean | test-other | |-----|------------|------------| diff --git a/egs/librispeech/ASR/transducer_stateless/decoder.py b/egs/librispeech/ASR/transducer_stateless/decoder.py index 4b1ec6ee6..f49f8c90b 100644 --- a/egs/librispeech/ASR/transducer_stateless/decoder.py +++ b/egs/librispeech/ASR/transducer_stateless/decoder.py @@ -75,6 +75,9 @@ class Decoder(nn.Module): Args: y: A 2-D tensor of shape (N, U) with blank prepended. + need_pad: + True to left padding the input. Should be True during training. + False to not pad the input. Used only during inference. Returns: Return a tensor of shape (N, U, embedding_dim). """