Add evaluation results with quality, accuracy, and speaker similarity metrics

initial commit