TTS VOICE DESIGN BENCHMARK

How well can TTS models bring characters to life?

An open benchmark measuring how convincingly TTS models create distinctive, role-fitting voices from natural-language character descriptions.

View Leaderboard Explore Dataset

LEADERBOARD

Voice Design Leaderboard

RankModel

Breeze TTS 2

78.0270898.5%

MiMo-V2.5-TTS

72.7820296.3%

Inworld Voice Design

64.6137898.2%

Fun-AudioGen-VD

63.8213698.6%

Eleven v3

58.8811199.7%

Qwen3-TTS-VD

58.488399.6%

VoxCPM2

55.2850998.5%

METRICS

Three measures of effective character voice design.

We evaluate role fit, voice diversity, and transcript pass. Together, they capture both individual performance and range across the full cast.

Get Evaluation Suite on GitHub

Role Fit

Whether the generated voice belongs to the character. Reviewers jointly rate age, texture, temperament, and delivery from 1 to 5 against the full role profile. Clips that fail the transcript gate are excluded.

Voice Diversity

The number of acoustically distinct voice clusters across the generated cast. This shows whether the cast sounds like different people or collapses into a few house voices. It complements Role Fit because a model may follow individual prompts well while repeatedly using the same few voices.

Transcript Pass

Whether each clip clearly speaks its intended line, judged separately from voice design quality. Minor pronunciation or clarity issues pass. Noticeable substitutions, omissions, or insertions are rated as degraded. Rewritten or skipped lines fail. The reported score is the share of clips rated Pass.

DATASET

1,000 characters across seven worlds of storytelling

Curated from CharacterCodex, each task pairs an anonymized character prompt with a natural preview script—testing distinctive voice creation without relying on known names or franchises.

Waffle chart of the 1,000 benchmark tasks grouped by media genre

See complete benchmark on Hugging Face

Literature, Stage and Lyric Works

18%

Novels • Plays • Short Stories • Songs • Poetry

Factual, News and Research Sources

17%

Biographies · Documentaries · Newspapers · Scientific Papers · Research Journals

Myth, Folklore and Historical Lore

14%

Mythology · Urban Legends · Historical Texts · Folklore · Fairy Tales

Illustrated and Animated Media

14%

Manga · Graphic Novels · Anime · Comic Books · Webcomics

Games and Interactive Play

13%

Video Games · Board Games · Card Games · Tabletop Role-Playing Games

Periodical and Web Media

13%

Magazines · Blogs · Online Articles

Film and Television

11%

Movies · Television Shows

LISTEN TO THE GAP

Same character. Same script. Which voice fits better?

Voice Design Prompt

Mature male, 50s-60s, Japanese-accented English. Dignified, measured, stoic tone. Clear texture with slight age rasp. Deliberate pacing, significant pauses. Conveys wisdom, authority, and visionary caution.

Script

We do not rebuild a nation with noise or haste. We place each stone with care, and when doubt rises, we answer it with discipline, patience, and an unshaken sense of duty.