Encodec config error and Encodec training on 16kHz data

[zhangchi2004]

1. It seems that the config for encodec model in stable_audio_tools/configs/model_configs/autoencoders/encodec_musicgen_rvq.json got the decoder strides reversed. It should be [8,5,4,4].
2. Has anyone tried to train the encodec model, especially on 16khz data? It didn't work well for me. Either the discriminator outperformed the generator causing gradient explosion, or the loss curves went normal but the generated samples were horrible.