HardNLPnlp-204

RNN-T Alignment Count

Problem

Return the number of monotonic RNN-T alignments between encoder_frames frames and target_tokens tokens: the binomial coefficient C(T + U, U).

Each alignment corresponds to a lattice path with T blank steps and U token steps. Negative inputs give 0.

Hint

Choose the positions of the token steps among all T + U steps.

case 1

in: [2,1]

expected: 3

case 2

in: [3,2]

expected: 10

case 3

in: [0,0]

expected: 1

case 4

in: [0,2]

expected: 1

case 5

in: [4,0]

expected: 1

Starter codePython

import math
def rnnt_alignment_count(encoder_frames, target_tokens):
    # Your code here
    pass

5 test cases · run your code against them in the editor