This commit is contained in:
Dongji Gao 2023-09-19 14:23:27 -04:00
commit 8c5c23ebe7

View File

@ -1,7 +1,7 @@
# Introduction
This is a weakly supervised ASR recipe for the LibriSpeech (clean 100 hours) dataset. We train a
conformer model using Bypass Temporal Classification (BTC)/Omni-temporal Classification (OTC) with transcripts with synthetic errors. In this README, we will describe
conformer model using [Bypass Temporal Classification](https://arxiv.org/pdf/2306.01031.pdf) (BTC)/Omni-temporal Classification (OTC) with transcripts with synthetic errors. In this README, we will describe
the task and the BTC/OTC training process.
Note that OTC is an extension of BTC and supports all BTC functions. Therefore, in the following, we only describe OTC.
@ -124,13 +124,26 @@ The bypass arc deals with substitution and insertion errors, while the self-loop
### Decoding
```
export CUDA_VISIBLE_DEVICES="0"
python conformer_ctc2/decode.py \
./conformer_ctc2/decode.py \
--exp-dir "${exp_dir}" \
--lang-dir "${otc_lang_dir}" \
--lm-dir "${lm_dir}" \
--otc-token "${otc_token}"
```
### Results (ctc-greedy-search)
| Traning Criterion | test-clean | test-other |
|------------|:-------:|:----:|
| CTC |100.0|100.0 |
| OTC | 11.89 | 25.46 |
### Results (1best, blank_bias=-4)
| Traning Criterion | test-clean | test-other |
|------------|:-------:|:----:|
| CTC |98.40|98.68 |
| OTC | 6.59 | 15.98 |
## Citations
```
@article{gao2023bypass,