These spectrograms are for the sentence "bright sunshine shimmers on the ocean". The first one shows the target acoustics for this sentence. The second one shows the acoustics output by a forward model that had been trained on this sentence, and on 229 other sentences. The third one shows the acoustics output by a forward model that was trained on 230 sentences, not including this one. The third spectrogram is the most important because it demonstrates that the model learned the general relationship between articulation and acoustics for this particular speaker in the MOCHA database.

Click on each spectrogram to play the corresponding wave files.

Target

Model trained on sentences including this one

Model trained on sentences NOT including this one