Articles

The Naval podcast team distills hour-long conversations into four minutes. See the test they use for what stays and how the editing process runs.

MelGAN generates raw audio from mel-spectrograms in real time. A look at how the model works and why speed matters for voice synthesis in editing.

Identical audio run through every major transcription service, with errors counted by hand. See how the results compared and why audio quality dominated.

Overdub predicts missing audio from surrounding context and a voice model. See how imputation works and why generated words blend into a real recording.