學語言的人每天按幾十次返回鍵，而沒有任何一個 app 在記錄那一下。 Echo 把它接住：不問問題、不打斷聆聽，只是把你重聽的那十五秒留下來，隔天再拿回來給你。 Language learners hit rewind dozens of times a day, and no app records that tap. Echo keeps the fifteen seconds you replayed — no questions, no interruption — and hands them back tomorrow.
回答 9 題 · 約 90 秒 TestFlight 開放時我會寄一次邀請給你 Answer 9 questions · about 90 seconds One TestFlight invite when the beta opens
Mirror · AI Speech Coach 把你的回答，用你自己的克隆聲音講回來。
4 週 502 註冊、零廣告、onboarding 完成率 90.2%、191 位啟用者跑了 556 場模擬面試。
所以「克隆聲音 + 逐句串流 TTS」這條技術鏈不是投影片，是上架的東西。
但要講精確：我自己那份 N=12 的受試者內實驗裡，主觀偏好在
helpful / like-me / easiest / comfortable 四個維度全勝，
客觀指標卻沒有分離。所以自聲買到的是「願意開口」，不是療效——
Echo 的每一次驗收都回到真實原音。
Mirror · AI Speech Coach replays your spoken answer in your own cloned voice.
502 signups in four weeks, zero ad spend, 90.2% onboarding completion, 556 mock-interview
sessions from 191 activated users. So the cloned-voice + per-sentence streaming TTS chain
isn’t a slide — it’s shipped.
To be precise about what it proves: in my own N=12 within-subject study, subjective preference
swept all four dimensions (helpful / like-me / easiest / comfortable), but
the objective measures did not separate. The cloned voice buys willingness to speak,
not efficacy — so Echo always grades against the original audio.