File: ctc_decoders.py

package info (click to toggle)
pytorch-audio 2.6.0-1
  • links: PTS, VCS
  • area: main
  • in suites: forky, sid, trixie
  • size: 10,696 kB
  • sloc: python: 61,274; cpp: 10,031; sh: 128; ansic: 70; makefile: 34
file content (15 lines) | stat: -rw-r--r-- 452 bytes parent folder | download | duplicates (3) 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
 
from torch import topk


class GreedyDecoder:
    def __call__(self, outputs):
        """Greedy Decoder. Returns highest probability of class labels for each timestep

        Args:
            outputs (torch.Tensor): shape (input length, batch size, number of classes (including blank))

        Returns:
            torch.Tensor: class labels per time step.
        """
        _, indices = topk(outputs, k=1, dim=-1)
        return indices[..., 0]