How make inference on a single WAV file using pre-train model?

[faranaziz]

No can find example for a single file.
Thanks you.

[sjscotti]

Hi
It has been awhile since I used it, but after you download the repo and install it as mentioned in https://github.com/haoxiangsnr/FullSubNet/blob/main/docs/prerequisites.md, you just follow the instructions on https://github.com/haoxiangsnr/FullSubNet/blob/main/docs/getting_started.md for Inference and Applying a Pre-trained Model .

as I remember you

1. download the checkpoint model fullsubnet_best_model_58epochs.tar from https://github.com/haoxiangsnr/FullSubNet/releases and put it in a folder you create (I'll refer to the path to this folder as <model>)
2. create a folder for your wave file (I'll refer to the path to this folder as <input>) and put your wave file in it
3. add the <input> path to the \recipes\dns_interspeech_2020\fullsubnet\inference.toml file as described in step 1 under Inference mentioned above

...
[dataset.args]
dataset_dir_list = [
    "<input>"
]
...

4. create a folder for your output (I'll refer to the path to this folder as <output>)
5. follow step 2 under Inference mentioned above

cd FullSubNet/recipes/dns_interspeech_2020

# One GPU is used by default
python inference.py \
  -C fullsubnet/inference.toml \
  -M <model> \
  -O <output>

Your denoised file will be in <output>

Good Luck!

[haoxiangsnr]

Hi
It has been awhile since I used it, but after you download the repo and install it as mentioned in https://github.com/haoxiangsnr/FullSubNet/blob/main/docs/prerequisites.md, you just follow the instructions on https://github.com/haoxiangsnr/FullSubNet/blob/main/docs/getting_started.md for Inference and Applying a Pre-trained Model .

as I remember you

1. download the checkpoint model fullsubnet_best_model_58epochs.tar from https://github.com/haoxiangsnr/FullSubNet/releases and put it in a folder you create (I'll refer to the path to this folder as <model>)
2. create a folder for your wave file (I'll refer to the path to this folder as <input>) and put your wave file in it
3. add the <input> path to the \recipes\dns_interspeech_2020\fullsubnet\inference.toml file as described in step 1 under Inference mentioned above

...
[dataset.args]
dataset_dir_list = [
    "<input>"
]
...

1. create a folder for your output (I'll refer to the path to this folder as <output>)
2. follow step 2 under Inference mentioned above

cd FullSubNet/recipes/dns_interspeech_2020

# One GPU is used by default
python inference.py \
  -C fullsubnet/inference.toml \
  -M <model> \
  -O <output>

Your denoised file will be in <output>

Good Luck!

@sjscotti Great, thanks for your attention and help.

[haoxiangsnr]

No can find example for a single file.
Thanks you.

Hi,

Thanks for your attention.

@sjscotti has given you a lot of details about how to enhance one file.

If you have any other questions, please feel free to contact me.