Inspect transcript words without start or end: skip them in cuts

The inspect transcript returns words[], and start and end are optional in the schema. A pause cutter that does w["end"] on every word will raise a KeyError on an untimed one, or worse, treat a missing time as 0 and cut a minute of speech. Skip untimed words when you compute gaps, but treat an untimed word as a barrier: do not remove any gap beside it.

Whether untimed words occur in practice for your audio is not stated in the docs; the code below is defensive.

Rules

• A gap exists only between two timed words that are adjacent in the list.
• If an untimed word sits between two timed ones, the gap is unknown: keep it.
• Never default a missing start or end to 0.
• Log how many words you skipped so a bad transcript is visible.

Effect on the cut list

Hypothetical transcript of six words with one untimed word at index 3.

Python

def gaps(words, threshold=0.35):
    cuts, prev = [], None
    for w in words:
        s, e = w.get("start"), w.get("end")
        if s is None or e is None:
            prev = None                 # barrier: gap beside it is unknown
            continue
        if prev is not None and s - prev > threshold:
            cuts.append((round(prev, 3), round(s, 3)))
        prev = e
    return cuts

Notes

• A skipped word still costs nothing extra; the STT price is per audio minute ($0.01).
• If many words are untimed, cut with sentence segments[] instead: they have no gaps by construction.
• Trim rejects an output under 0.2 s, so still drop slivers.