Fix broken timbre transfer demo notebook + update to ddsp==3.5.0

[olaviinha]

Current timbre transfer demo notebook does not work in Colab. According to issue tracker hasn't worked since November 2022.

I see saving from Colab to Github did some minor unintended nugatory changes to character encodings, so...

Notable changes that fix the notebook are:

92: !pip install -qU ddsp==3.5.0 "hmmlearn<=0.2.7"

118: import tensorflow as tf

124: from tensorflow.python.ops.numpy_ops import np_config 125: np_config.enable_numpy_behavior()

347: audio_features_mod = {k: copy.copy(v) for k, v in audio_features.items()}

388: audio_features['loudness_db'].numpy(),

[olaviinha]

Hmm this makes the notebook work but it appears to produce rather poor results, at least with models trained on ddsp[data_preparation]==1.6.3.