Audio, Noise, and Responsible Inference

Audio can support event timing when a scoreboard is not visible, but a competition recording may also contain music, crowd reactions, commentary, echoes, advertisements, and sounds from another area. Every audio observation therefore needs its own confidence.

Separate the channels of evidence

A whistle-like sound may suggest a pause, an impact may align with a visible strike, and speech may contain a score announcement. None of these should silently override conflicting video evidence. The system should record which channel contributed to an inference and how reliable that channel was at the time.

• Detect persistent music and background noise before interpreting short events.
• Align audio and video clocks after edits or dropped frames.
• Treat speech recognition as uncertain when language or speaker identity is unclear.
• Require corroboration before reporting an inferred final score.

Sound is supporting evidence, not an automatic verdict.

MachX multimodal analysis principle