Shortcuts

Reproduciblity

Overview

We use deterministic random number generation to ensure replicability, even of noise oscillators. Nonetheless, there are small numeric differences between the CPU and GPU. The mean-average-error between audio samples generated on CPU and GPU are < 1e-2.

Defaults

Reproducibility is currently guaranteed when batch_size is multiple of 32 and you use the default SynthConfig settings: sample_rate=44100, control_rate=441.

Train vs Test

If a train/test split is desired, 10% of the samples are marked as test. Because researchers with larger GPUs seek higher-throughput with batchsize 1024, \(9 \cdot 1024\) samples are designated as train, the next 1024 samples as test, etc.

All AbstractSynth forward() methods return three batched tensors: audio, latent parameters, and an is_train boolean vector.

Read the Docs v: stable
Versions
latest
stable
v1.0.1
v1.0.0
Downloads
pdf
html
epub
On Read the Docs
Project Home
Builds

Free document hosting provided by Read the Docs.