funasr/models/bicif_paraformer/cif_predictor.py
@@ -198,7 +198,7 @@ output2 = self.upsample_cnn(_output) output2 = output2.transpose(1, 2) output2, _ = self.self_attn(output2, mask) # import pdb; pdb.set_trace() alphas2 = torch.sigmoid(self.cif_output2(output2)) alphas2 = torch.nn.functional.relu(alphas2 * self.smooth_factor2 - self.noise_threshold2) # repeat the mask in T demension to match the upsampled length