Add forced alignement.

[Thomcle]

Forced alignment, as proposed by the Montreal Forced Aligner tool is complicated to implement in applications. Would it be possible to integrate forced alignment into whisper timestamp? Would it be possible to integrate this tool into whisper timestamp, i.e. to obtain a transcript but indicate the words and precise punctuation as input so that the model doesn't make any mistakes?

Thanks you.