Audio Courtsider: in-browser models

Weights for the in-browser Live mode of Audio Courtsider (demo), which listens to soccer commentary and flags goals, penalties and red cards before the official data feed.

file	what
`text/model.onnx`	DistilRoBERTa event model (fp16): P(goal, shot, corner, yellow, red, penalty, offside, substitution) + which team, from a line of commentary with two lines of context
`audio/model.onnx`	causal temporal CNN over 64-band log-mel: P(goal in the last few seconds) from crowd + commentator sound
`audio/mel.bin`	the log-mel filterbank (librosa, 16 kHz, n_fft 1024, 64 bands, fmax 8 kHz)
`meta.json`	labels, thresholds, the audio calibrator curve and a feature self-test

Training data: SoccerNet action-spotting labels and SoccerNet-Echoes commentary (CC BY 4.0), ESPN 2025/26 play-by-play text, and (audio model) SoccerNet broadcast audio used under the SoccerNet NDA. For research and demonstration; not for betting.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support