torchaudio.models.conv_tasnet_base¶
- torchaudio.models.conv_tasnet_base(num_sources: int = 2) ConvTasNet[source]¶
Builds non-causal version of
ConvTasNet.The parameter settings follow the ones with the highest Si-SNR metirc score in the paper, except the mask activation function is changed from “sigmoid” to “relu” for performance improvement.
- Parameters:
num_sources (int, optional) – Number of sources in the output. (Default: 2)
- Returns:
ConvTasNet model.
- Return type:
