Torchaudio.transforms.griffinlim output to tensorboard audio

I am trying to implement a VAE on audio and want to listen to the reconstructed audio via tensorboard. The input and output of my network is a spectrogram (computed with torchaudio.transforms.Spectrogram) so I assume I should use torchaudio.transforms.GriffinLim to get a listenable waveform.
However, when I send the output of torchaudio.transforms.GriffinLim to TensorBoard, I get warning: audio amplitude out of range, auto clipped. and there is no “audio” tab in TensorBoard.

It seems that GriffinLim outputs a waveform with values between -2 and 2, but I do not know what kind of audio format is supported by TensorBoard so I do not know what kind of transform I should perform after applying GriffinLim.
Does anyone have an idea?