bpms = core.tempo_frequencies(tg.shape[0], hop_length=hop_length, sr=sr)
// Weight the autocorrelation by a log-normal distribution
prior = np.exp(-0.5 * ((np.log2(bpms) - np.log2(start_bpm)) / std_bpm)**2)
// Kill everything above the max tempo
if max_tempo is not None:
max_idx = np.argmax(bpms < max_tempo)
After Change
// it would also make time aggregation much more natural
// Get the maximum, weighted by the prior
best_period = np.argmax(np.log(tg + 1e-10) + logprior[:, np.newaxis], axis=0)
tempi = bpms[best_period]
// Wherever the best tempo is index 0, return start_bpm
