Fig. 3. We compared the effectiveness of decoding algorithms at reconstructing various audio aspects and generating intelligible audio using six different performance metrics.

The top-performing algorithm across all metrics was the LSTM RNN (red stars).