I trained andrej karpathy's char-rnn on chunks of popular trap songs and farmersmanual material and sampled from these models during various stages of the training. Result is something between noise and a spectral approximation of the target plus some slower state switches (201510).
sample_exwe_0161_cv_fm_lstm_256_3L_03_lm_lstm_epoch27 .46767_ tr。
聲道：1
類型：mp3
neural、network、recurrent、char-rnn、generative
神經、網絡、復發、炭RNN、生成的
資料來源：freesound ( https://freesound.org )