最良のプロバイダー音声テキスト読み上げ（TTS）モデル

Relative Elo score of the models as determined by responses from users in Artificial Analysis' Speech Arena. Some models may not be shown due to not yet having enough votes.

Share of highlighted English spans that human reviewers judged as pronounced correctly. Each model is measured with a single voice, shown on the Pronunciation Robustness Overview page. Prompts are sent to each model as written, with no normalization on our side. Where a model offers a text normalization setting, we use its default. Spans fall into four categories:

• Standalone Terms (term pronunciation): brand, place and technical names pronounced correctly, e.g. Arkansas, façade, or genre.
• Contextually Appropriate (contextual disambiguation): words spelled the same but read differently depending on context, e.g. a wound that is bandaged vs. a bandage that is wound.
• Preserving Exact Sequences (sequence fidelity): codes, paths, emails and identifiers spoken exactly, e.g. .env.local or a.chen@ucsf.edu.
• Expanding Shorthand (text normalization): numbers, dates, units and notation read out naturally, e.g. 6'2", Chapter XVII, or 1 tsp of sugar.

Clips are each judged by up to three independent reviewers, and any reviewer who fails one of our attention checks has all of their votes removed. Accepted pronunciations include the variants valid in both US and UK English and other legitimate variations (e.g. quay as "key" or "kay", and SQLite as "ess-cue-ell-light" or "sequel-light"). Some models may not be shown due to not yet having results.

Share of highlighted English spans that human reviewers judged as pronounced correctly. Each model is measured with a single voice, shown on the Pronunciation Robustness Overview page. Prompts are sent to each model as written, with no normalization on our side. Where a model offers a text normalization setting, we use its default. Spans fall into four categories:

• Standalone Terms (term pronunciation): brand, place and technical names pronounced correctly, e.g. Arkansas, façade, or genre.
• Contextually Appropriate (contextual disambiguation): words spelled the same but read differently depending on context, e.g. a wound that is bandaged vs. a bandage that is wound.
• Preserving Exact Sequences (sequence fidelity): codes, paths, emails and identifiers spoken exactly, e.g. .env.local or a.chen@ucsf.edu.
• Expanding Shorthand (text normalization): numbers, dates, units and notation read out naturally, e.g. 6'2", Chapter XVII, or 1 tsp of sugar.

Clips are each judged by up to three independent reviewers, and any reviewer who fails one of our attention checks has all of their votes removed. Accepted pronunciations include the variants valid in both US and UK English and other legitimate variations (e.g. quay as "key" or "kay", and SQLite as "ess-cue-ell-light" or "sequel-light"). Some models may not be shown due to not yet having results.

Price per 1M characters of text. For detail on how we calculate price for providers which price based on inference time or subscription plans, see our methodology page.

Number of characters processed per second of generation time. Higher values indicate faster generation speeds.