Table 3 Objective evaluation comparison of three methods
From: MetaMGC: a music generation framework for concerts in metaverse
Metrics | Transformer XL | Transformer-XL+MC | Transformer-XL+DQN | POP909 Dataset | Melody_LSTM | RL-tuner |
|---|---|---|---|---|---|---|
Sound range | 39.25 | 37.12 | 39.09 | 53.04 | 37.46 | 53.04 |
Average of the number of different pitches | 19.6 | 20.62 | 24.1 | 36.74 | 18.6 | 36.74 |
Average number of pitches played simultaneously | 2.81 | 2.85 | 2.79 | 3.77 | 2.15 | 3.77 |
Retuning time step ratio | 71.2% | 72.7% | 74.2% | 74.9% | 64.8% | 74.9% |
Ratio of tones in the tonality | 49.3% | 58.5% | 61.05% | 61.3% | 40.5% | 61.3% |
Information entropy of pitch | 2.73 | 2.68 | 2.63 | 2.82 | 2.91 | 2.82 |
Rhythmic consistency with the first two bars | 94.3% | 97.1% | 97.3% | 98.8% | 81.3% | 98.8% |