Path to the checkpoint directory.
Disposes of any untracked Tensors to avoid GPU memory leaks.
Loads variables from the checkpoint and builds the model graph.
Returns true iff model is intialized.
Predicts LogMelMagnitudes and IFreq from latent (z) and pitch conditioning.
A 4-D Tensor of latent (z) concatenated with one-hot pitch conditioning. Size [batch, 1, 1, nLatents].
Size of input batch.
a 4-D Tensor size [batch, time, freq, ch]. First channel is LogMelSpectrogram and second channel is InstantaneousFrequency.
Creates one random sample of LogMelSpecgram and IFreq from integer pitch.
Integer MIDI pitch number of sound to generate.
specgrams a 4-D Tensor size [batch, time, freq, ch]. First channel is LogMelSpectrogram and second channel is InstantaneousFrequency.
Include specgramsToAudio as a member method for API/export.
a 4-D Tensor size [batch, time, freq, ch]. First channel is LogMelSpectrogram and second channel is InstantaneousFrequency.
Float32Array of audio samples for first specgram in the batch.
Generated using TypeDoc
GANSynthconstructor.