0

I am looking for algorithms/methods for performing a forced alignment of 2D sequences that have different lengths. I have extracted data that traces the mouth movements of different people saying the same thing. Since people say things at different speeds, the vectors obtained from the tracing have different lengths.

For each person I have a 2D vector with the position of a person's mouth relative to the x and y planes. Let us say person one said "hello" and the trace was [[1,3,4,6,7], [3,5,6,3,2]], whereas person two said "hello" as well and the trace was [[1,3,4,6,7,4,6,7], [3,5,6,3,2,2,6,3]].

Note that person two took longer to say "hello" compared to person one. I am trying to force align these sequences so they have the same length.

Any ideas?

Vicky
  • 969
  • 1
  • 11
  • 19
NeuralNew
  • 96
  • 1
  • 10
  • 2
    Possibly by interpolation or truncation. Truncation would be riskier if the values in the shorter trace are not guaranteed to be in the longer trace. Maybe do an auto-correlation and truncate the longer trace after time matching. Another possibility is interpolating the smaller trace to match the length of the longer trace. – MichaelTr7 Feb 23 '21 at 03:53
  • Thanks for the ideas! I am going to try them. I think interpolation and resampling is a good option for this issue. – NeuralNew Feb 26 '21 at 19:11

0 Answers0