Installation

Installation for all features

First, follow Installation for Training.
Second, install the following packages:

pip install torchmetrics jiwer onnx onnxsim onnxscript
pip install git+https://github.com/openai/whisper.git

Third, install onnxruntime-gpu.
Make sure to install the version that matches your CUDA version.
If you cannot install the GPU version, you can install the CPU version instead.
However, DNSMOS and SCOREQ will run on CPU and the metrics_ns.py code will run very slow.

Minimal installation for training

Required by train.py and train_torchrun.py.

Refer to Installation for Training.

Minimal installation for calculating objective metrics

Required by metrics_ns.py.

First, follow Installation for Training.
Second, install the following pacakges:

pip install torchmetrics jiwer
pip install git+https://github.com/openai/whisper.git

Finally, install onnxruntime-gpu.
Make sure to install the version that matches your CUDA version.
If you cannot install the GPU version, you can install the CPU version instead.
However, DNSMOS and SCOREQ will run on CPU and the metrics_ns.py code will run very slow.

Minimal installation for ONNX exporting

Required by scripts/export_onnx.py and scripts/export_onnx_spec.py.

First, follow Installation for Training.
Second, install the following pacakges:

pip install onnx onnxsim onnxscript

Minimal Installation for ONNXRuntime (spec2spec version)

Required by scripts/test_onnx_spec.py.

First, install PyTorch. It doesn't need to be GPU version.
Second, install the following pacakges:

pip install numpy scipy librosa tqdm

Finally, install onnxruntime.
It doesn't matter whether you intsall a CPU version or a GPU version. Even if you install a GPU version, the code will run on CPU anyway.

Minimal Installation for ONNXRuntime (wav2wav version)

Required by scripts/test_onnx.py.

You don't need to install PyTorch in this case. First, install the following pacakges:

pip install numpy scipy librosa tqdm

Then install onnxruntime.
It doesn't matter whether you intsall a CPU version or a GPU version. Even if you install a GPU version, the code will run on CPU anyway.