What it means
Word alignment connects individual words to positions in a recording. A transcript answers what was said; alignment concerns when those words occur. A sentence-level subtitle interval alone does not establish each word's boundary.
Two inputs, different questions
Speech recognition can propose text and timing from audio. Forced alignment instead matches supplied text to audio using an acoustic model and related resources. The Montreal Forced Aligner workflow is one example; WhisperX combines recognition with an alignment stage. Neither description guarantees that every proposed word or boundary is correct.
Why equal division is not alignment
Dividing a ten-second sentence by its word count produces estimated slots. Speech contains pauses and words of different lengths, so those slots are not observations of the recording. Label such timings as estimates.
Review a real result
Check names, numbers, missing words, overlapping intervals and words close to a clip boundary against the audio. Preserve the original machine result and make corrections as a separate revision.
Topicut's public time shifter and format converter operate on supplied caption times. They do not listen to audio or perform word alignment. A successful conversion is not an alignment test.