Sep-TFAnet VAD :Single-Microphone Speaker Separation and Voice Activity Detection in Noisy and Reverberant Environments
To pull the Docker image mordehay/separation_vad:1.0
, use the following command:
docker pull mordehay/separation_vad:1.0
Once the image has been pulled, you can create and start a container:
docker run -it mordehay/separation_vad:1.0 /bin/bash
You will be placed into an interactive terminal inside the container.
With the container up and running, execute the only_inference.py
script. Here are the available arguments for the script:
python only_inference.py \
-c/--config [CONFIG_PATH] \
-r/--resume [CHECKPOINT_PATH] \
-d/--device [DEVICE] \
-sp/--save_test_path [SAVE_PATH] \
-o/--online [TRUE/FALSE] \
-ps/--precision_save [16/32] \
-pm/--path_mix [MIX_SIGNAL_PATH] \
-ikw/--inference_kw [INFERENCE_OPTIONS]
-
-c/--config
: Path to the config file.- Default:
config_with_vad.json
- Default:
-
-r/--resume
: Path to the latest checkpoint.- Default:
model_without_vad.pth
- Default:
-
-d/--device
: Indices of GPUs to enable.- Default: All GPUs.
-
-sp/--save_test_path
: Directory where separated signals will be saved.- Default: Current directory.
-
-o/--online
: Online mode toggle.- Default:
True
- Default:
-
-ps/--precision_save
: Precision of the separated signals.- Choices:
16
or32
- Default:
32
- Choices:
-
-pm/--path_mix
: Path to the mixed signal for running the separation. Required argument. -
-ikw/--inference_kw
: Options for the inference phase, such as:filter_signals_by_smo_vad
: Filter by smoothed VAD (Default:False
).filter_signals_by_unsmo_vad
: Filter by unsmoothed VAD (Default:False
).length_smoothing_filter
: Length of smoothing filter (Default:3
).threshold_activated_vad
: VAD threshold (Default:0.5
).return_smoothed_vad
: Toggle to return smoothed VAD (Default:False
).
This argument must be dictionary as JSON string for an exmaple:
'{"filter_signals_by_smo_vad": false, "length_smoothing_filter": 3, "return_smoothed_vad": true}'