Skip to content

Commit 3f1959f

Browse files
sangeet2020sangeet2020
committed
add clarity to training data; add computing power details
1 parent 9eadc9e commit 3f1959f

1 file changed

Lines changed: 3 additions & 0 deletions

File tree

‎recipes/RescueSpeech/README.md‎

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -14,6 +14,9 @@ python train.py hparams/robust_asr_16k.yaml --data_folder=<data_folder_path>
1414
```
1515
Here the data path should be the path to **uncompressed `Task_ASR.tar.gz`** downloaded from link above.
1616

17+
## Computing power
18+
Please note that running this recipe can be computationally demanding due to the Whisper ASR (`whisper-large-v2`) model with 906.5M parameters (compared to 1.5B parameters in the original model but feature encoder is frozen in our case). When fine-tuning both the Whisper and SepFormer models together, we used an Nvidia A100-80 GB GPU, which took approximately 15 minutes per epoch.
19+
1720
## Results
1821
During training, both speech enhancement and ASR is kept unfrozen- i.e. both ASR and ehnance loss are backpropagated and weights are updated.
1922

0 commit comments

Comments
 (0)