ref_db=20, max_db=100, Where did these values come from? statistics?

Kyubyong / tacotron

A TensorFlow Implementation of Tacotron: A Fully End-to-End Text-To-Speech Synthesis Model

Apache License 2.0

1.83k stars 436 forks source link

Open alayamanas opened 5 years ago

alayamanas commented 5 years ago

mel = np.clip((mel - hp.ref_db + hp.max_db) / hp.max_db, 1e-8, 1) mag = np.clip((mag - hp.ref_db + hp.max_db) / hp.max_db, 1e-8, 1)

Is there any reason for using this normalization method ?