Open TTS Leaderboard: pick an AI voice by ear and check the cost

The Open TTS Leaderboard is a free Hugging Face Space that tests open text-to-speech models on three things: whether the words come out right (word error rate, checked by a speech recognition model), generation speed, and how close a cloned voice sounds to the reference clip. Its Listen tab lets you play the actual samples behind those scores, for the same passage across models.

The catch is that the default ranking is English word error rate. It doesn't tell you which voice sounds the most natural. This guide shows you how to compare voices with the names hidden, try the words they might trip over, and check what your videos would cost. I haven't run a listening study of my own. The scores are the leaderboard's.