Gradium AI发布新TTS模型：硬场景通过率81%，首字延迟216ms
Gradium AI Releases New Default TTS Model: 81.0% Hard-Case Pass Rate at 216 ms Time-to-First-Audio
Agent开发者的痛点就是数字和专有名词读错，这个模型在保持低延迟的同时把硬场景通过率做到了行业第一，直接解决语音交互最核心的可靠性问题，值得集成压测。
Voice agents fail on exactly the parts of a call that matter most: the order number, the callback digits, the email address the caller has to write down. Gradium AI has released a new text-to-speech model and made it the default across its API and Studio. The company reports an 81.0% human-rated pass rate on a 500-sentence hard-case set spanning five languages, ahead of Cartesia Sonic 3.6 at 75.1% and ElevenLabs v3 Conversational at 65.4%. Time to first audio is 216 ms at P50 on Coval, 170 ms faster than the model it replaces.
语音助手在通话中最关键的环节上频频出错：订单号、回拨数字、以及需要拨打者记下的电子邮件地址。Gradium AI 发布了一款新的文本转语音模型，并将其设为其 API 和 Studio 的默认模型。该公司报告称，在一个涵盖五种语言的 500 句高难度测试集中，人工评分通过率高达 81.0%，领先于 Cartesia Sonic 3.6 的 75.1% 和 ElevenLabs v3 Conversational 的 65.4%。在 Coval 基准测试中，首音频延迟的中位数（P50）为 216 毫秒，比其替代模型快 170 毫秒。
Is it deployable?
它是否可部署？
Yes, today, with no migration. Gradium switched the model on as the default across its API and Studio on August 31, 2026. Existing voices, including custom clones, keep working unchanged.
是的，今天即可部署，无需迁移。Gradium 已于 2026 年 8 月 31 日将其模型切换为 API 和 Studio 的默认设置。现有声音（包括自定义克隆声音）保持不变，继续正常工作。
The accuracy number
准确率数据
Gradium built a 500-sentence evaluation set and open-sourced it on Hugging Face under CC BY 4.0: 100 items across 10 criteria in five languages (EN, DE, FR, ES, PT). Seven atomic criteria cover spelling, acronyms, alphanumeric tokens, dates, regular numbers, large and floating numbers, and email. Three composite criteria (Orders, IT Ticket, Claims) stack several of those into one realistic agent turn.
Gradium 构建了一个包含 500 个句子的评估集，并在 Hugging Face 上以 CC BY 4.0 协议开源：涵盖五种语言（EN、DE、FR、ES、PT），每种语言 100 个项目，共 10 项标准。七项原子标准涵盖拼写、缩写、字母数字令牌、日期、普通数字、大数和浮点数，以及电子邮件地址。三项复合标准（订单、IT 工单、索赔）将其中几项组合成一个真实的智能体交互轮次。
Scoring is human and strict. A sentence passes only if an independent native-speaker rater hears every element pronounced correctly and completely; one dropped digit fails the sentence. Audio was loudness-normalized, order randomized, and raters capped at 40 comparisons with an enforced break.
评分由人工进行且非常严格。只有当独立的本语母语评审员听到每个元素都发音正确且完整时，该句子才算通过；只要漏掉一个数字，整个句子即判定失败。音频经过响度标准化处理，顺序随机打乱，每位评审员的比较次数上限为 40 次，并强制休息。
Pooled across the ten criteria and averaged over the five languages with equal weight: Gradium TTS 81.0%, Cartesia Sonic 3.6 75.1%, ElevenLabs v3 Conversational 65.4%, Fish Audio S2.1 Pro 49.5%, Inworld TTS 1.5 Max 46.5%. All generated in August 2026 with default settings.
综合十项标准并以五种语言等权重平均后得出：Gradium TTS 为 81.0%，Cartesia Sonic 3.6 为 75.1%，ElevenLabs v3 Conversational 为 65.4%，Fish Audio S2.1 Pro 为 49.5%，Inworld TTS 1.5 Max 为 46.5%。所有结果均使用默认设置在 2026 年 8 月生成。
The latency number
延迟数据
On Coval’s TTS benchmark, Gradium reports a 216 ms P50 time to first audio, 170 ms faster than the model it replaces. The more useful figure is the spread: a 30 ms p75-p25 interquartile range across 480 runs, the tightest of the five models tested. Cartesia Sonic 3.6 sits at 454 ms median with a 165 ms spread, 36% of its own median, and callers experience tail turns rather than medians.
在 Coval 的 TTS 基准测试中，Gradium 报告的首音频延迟中位数（P50）为 216 毫秒，比其替代模型快 170 毫秒。更有参考意义的是数据的离散程度：在 480 次运行中，第 75 百分位与第 25 百分位之间的四分位距仅为 30 毫秒，是所测五种模型中最紧凑的。Cartesia Sonic 3.6 的中位数为 454 毫秒，离散度为 165 毫秒（占其中位数的 36%），而呼叫者实际经历的是长尾分布而非中位数表现。
Gradium is not the fastest model on that chart. Inworld TTS 2 posts a 166 ms median; Fish Audio S2.1 Pro (291 ms) and ElevenLabs v3 Conversational (329 ms) trail Gradium. The claim being made is about joint position: the lowest hard-case failure rate at sub-250 ms first audio, with very little variance.
Gradium 并非该图表中最快的模型。Inworld TTS 2 的中位数为 166 毫秒；Fish Audio S2.1 Pro（291 毫秒）和 ElevenLabs v3 Conversational（329 毫秒）则落后于 Gradium。其主张的核心在于综合排名：在首音频延迟低于 250 毫秒的前提下，拥有最低的高难度场景失败率，且方差极小。
Getting started
入门指南
Existing users need do nothing. New teams install the Python SDK, point at the WebSocket TTS endpoint, and reuse existing voice IDs. Gradium is offering 1M credits for complete hard-case failure reports on its Discord.
现有用户无需执行任何操作。新团队安装 Python SDK，指向 WebSocket TTS 端点，并复用现有的语音 ID。Gradium 正在其 Discord 上提供 100 万积分，用于提交关于其完全硬案例失败的报告。
Key Takeaways
关键要点
- New Gradium TTS model is live and default as of August 31, 2026; no migration needed.
- 81.0% human-rated pass rate on 500 hard sentences, ahead of Cartesia, ElevenLabs, Fish Audio and Inworld.
- 216 ms P50 time to first audio on Coval, with a 30 ms interquartile spread over 480 runs.
- Reads phone numbers, emails, IBANs and reference codes with no text normalization required.
- Vendor-run benchmark, but the 500-sentence evaluation set is open on Hugging Face under CC BY 4.0.
- 新的 Gradium TTS 模型已于 2026 年 8 月 31 日上线并成为默认选项；无需进行迁移。
- 在 500 个高难度句子中，人工评分通过率高达 81.0%，领先于 Cartesia、ElevenLabs、Fish Audio 和 Inworld。
- 在 Coval 上，首音频生成时间 P50 为 216 毫秒，在 480 次运行中四分位距为 30 毫秒。
- 能够准确读取电话号码、电子邮件、IBAN 和参考代码，无需进行文本规范化。
- 虽然基准测试由供应商运行，但包含 500 个句子的评估数据集已在 Hugging Face 上以 CC BY 4.0 许可开源。
Check out the release post and the dataset. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well.
查看发布帖子和数据集。此外，欢迎在 Twitter 上关注我们，别忘了加入我们拥有 15 万+成员的 ML SubReddit，并订阅我们的新闻通讯。等等！你在 Telegram 上吗？现在你也可以加入我们的 Telegram 群组。
Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us
需要与我们合作推广你的 GitHub 仓库、Hugging Face 页面、产品发布或网络研讨会等吗？请联系我们
The post Gradium AI Releases New Default TTS Model: 81.0% Hard-Case Pass Rate at 216 ms Time-to-First-Audio appeared first on MarkTechPost.
MarkTechPost 上首先刊登的文章《Gradium AI 发布新的默认 TTS 模型：81.0% 的高难度案例通过率，首音频生成时间为 216 毫秒》
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力