Pad speech crops before diagnosing numeric ITN errors

Pad speech crops before diagnosing numeric ITN errors

When a spoken identifier loses its first digit, trace the value stage by stage before blaming inverse text normalization. Compare source audio onset, crop boundaries, frozen ASR text, ITN output, and argument parsing. A crop that starts at a speech-event timestamp with no leading padding can cut into the first word; downstream ASR then omits it, while ITN correctly normalizes the surviving words. Add a short pre-roll or VAD-onset rewind for identifier crops, then regenerate the ASR cache and replay the resolver before retraining. This does not address genuine ASR substitutions or ambiguous speech.

Applies when: speech-derived tool arguments show systematic leading-character loss. Keywords: ASR NeMo ITN audio crop pre-roll VAD identifier digits