INTRODUCING VOICE ACTING ARENA

Today, LAION and TTS Arena are launching Voice Acting Arena - a platform for comparing AI voice acting, with human listeners evaluating the acting quality.

Text-to-speech has made remarkable progress. Today's systems can sound clear, natural, fluent, and increasingly human.

But sounding human is not the same as giving a human actor level performance.

Most TTS systems are optimized for contexts where consistency is a strength: assistants, call centers, narration, customer support. The ideal voice is often clear, friendly, and professional.

Acting is different.

A convincing performance may need to sound frightened, exhausted, awkward, angry, vulnerable, restrained, or emotionally conflicted. It may include hesitation, a shaky breath, an uncomfortable pause, or a sentence that trails off. These aren't necessarily imperfections to eliminate - they can be what makes a performance feel real.

Voice Acting Arena asks a different question from traditional TTS evaluation: not just "Does this sound realistic?" but "Do I believe this performance?"

Same scene. Two performances. Your judgment.

Each matchup presents a scene, acting instructions, and two anonymous model-generated performances of the same script.

You might hear a dispatcher reassuring a frightened caller, someone trying to hide bad news, or a person recording a farewell while trying to sound normal.

Listen to both takes, then compare them on three dimensions:

• Overall acting preference: Which performance do you prefer overall?
• Instruction following: Which better matches the scene and acting direction?
• Genuineness and authenticity: Which feels more believable, rather than forced or mechanical?
• Vocal bursts (where requested): Which better delivers reactions such as laughter, sobs, gasps, or screams naturally and in line with the acting direction?

We keep these judgments separate deliberately. A performance can sound natural while missing the emotion. Another can follow the direction closely without feeling convincing. Sometimes the less polished take is the one that feels more true.

Building toward voices with human level acting performance

We want Voice Acting Arena to grow into an ongoing evaluation for increasingly capable voice models, with new models, scenes, emotions, and acting challenges.

Your votes will help show where today's systems already succeed, where they fall back into generic "pleasant TTS," and what still separates realistic speech from a convincing performance.

They will also support our research collaboration, including a planned paper, and our broader goal of advancing open voice acting models.

AI voices shouldn't just be able to speak. They should be able to perform, reflecting the richness and nuances of natural speech and professional acting

Put on headphones, try a few comparisons, and tell us which performances land.

Try Voice Acting Arena → https://voiceacting.ttsarena.org/