ref: c74876bbc699f57ef921b1fdc6467343955e9b1d
parent: f3eb6164551b17bc06e740d23e0821f77300e603
author: Jean-Marc Valin <jmvalin@jmvalin.ca>
date: Fri Oct 12 23:54:58 EDT 2018
Adding some instructions
--- /dev/null
+++ b/dnn/README
@@ -1,0 +1,18 @@
+In the src/ directory, run ./compile.sh to compile the data processing program.
+Then, run the resulting executable:
+./dump_data input.s16 exc.s8 features.f32 pred.s16 pcm.s16
+
+where the first file contains 16 kHz 16-bit raw PCM audio (no header)
+and the other files are output files. The input file I'm using currently
+is 6 hours long, but you may be able to get away with less (and you can
+always use ±5% or 10% resampling to augment your data).
+
+Now that you have your files, you can do the training with:
+./train_wavenet_audio.py exc.s8 features.f32 pred.s16 pcm.s16
+and it will generate a wavenet*.h5 file for each iteration.
+
+You can do the synthesis with:
+./test_wavenet_audio.py features.f32 > pcm.txt
+
+If you're lucky, you may be able to get the current model at:
+https://jmvalin.ca/misc_stuff/lpcnet_models/
--
⑨