Need for a verbatim / raw transcription mode for forced alignment

[bvegliantestudioeco]

Whisper produces transcriptions optimized for readability, which works very well
for most use cases.

However, for professional workflows involving forced alignment (e.g. Aeneas,
subtitle timing, e-learning pipelines), this creates a limitation.

What seems to be missing is a verbatim / raw transcription mode that:

• preserves fillers, hesitations and repetitions
• preserves exact word order
• avoids merging or rewriting spoken content
• prioritizes alignment correctness over readability

This would not require changing the default behavior.
A separate flag or decoding mode (e.g. --verbatim) could explicitly separate
transcription from normalization.

Today users are forced to choose between:

• readable text that drifts during alignment
• verbatim text with recognition errors

Curious to hear if this is something the team has considered,
or if there are recommended approaches for alignment-critical workflows.