“ElevenLabs believes that for achieving high performance in voice models, model architecture is more important than scale.”