原帖引用 · AI 翻译
阅读 Phonic-1 的技术报告
一家由 10 人组成的初创公司 Phonic 发布了 Phonic-1。该公司称，这是唯一能在来电者话还没说完时就执行查询和预订的语音模型。
> 在 Artificial Analysis 的语音到语音指数中以 87.1 分位居第 1，仅比 xAI 的 Grok Voice 3.0 高 0.2 分；完成了 79.9% 的客户服务任务，首次输出音频的耗时为 0.61 秒。
> Phonic-1 在来电者仍在说话时调用工具；演示中，它在对方说话期间完成了 5 次查询，并在对方停止说话后 1.1 秒作出回复，而 OpenAI 的 GPT-Live 耗时 15.7 秒。
> Phonic 报告称，其在 Cekura 的通话测试套件中得分 90.2，超过了得分 82.9% 的“转录—思考—说话”级联方案；后者在该测试中超过了所有上榜的语音到语音模型。不过，Cekura 的公开榜单将 Phonic v1 列为 68.3%。
> 他们完成了由 First Round 领投的 2750 万美元 A 轮融资，使总融资额达到 3150 万美元；Phonic 已在 25 个国家为 Ashby 的 AI 求职面试提供支持。
https://t.co/laxOzIhDwx
原文
A 10-person startup Phonic has released Phonic-1, which it says is the only voice model that can run lookups and bookings while the caller is still mid-sentence. > First on Artificial Analysis’s speech-to-speech index at 87.1, just 0.2 points above xAI’s Grok Voice 3.0, completing 79.9% of customer-service tasks with 0.61 seconds to first audio. > Phonic-1 calls tools while the caller is still talking; in the demo it finished 5 lookups mid-speech and replied 1.1 seconds after the caller stopped, versus 15.7 seconds for OpenAI’s GPT-Live. > Phonic reports 90.2 on Cekura’s call suite, beating a transcribe-think-speak cascade at 82.9% that outscores every ranked speech-to-speech model there, though Cekura’s public board lists Phonic v1 at 68.3%. > They raised a $27.5M Series A led by First Round, taking total funding to $31.5M, and Phonic already powers Ashby’s AI job interviews in 25 countries. https://t.co/laxOzIhDwx
推出 Phonic-1：来自语音到语音模型公司 @phonic_co、面向智能体任务的最先进语音到语音模型。
过去 1 年，我们取得了突破性进展，并完成了由 @firstround 领投、@Lux_Capital、@felicis 和 Definition 参与的 2750 万美元 A 轮融资。https://t.co/syYLXbQ4lL
原文
Introducing Phonic-1, the most advanced speech-to-speech model for agentic tasks — from @phonic_co, the speech-to-speech model company. We’ve had a breakout year and raised a $27.5M Series A led by @firstround with participation from @Lux_Capital, @felicis, and Definition. https://t.co/syYLXbQ4lL