Micky Yang

Tonmeister live in Taiwan
Mixing engineer · Dolby Atmos
Building UTUVO — audio tools for people who listen closely
·

[Micky Yang @yangmicky]

我自己平常很依賴語音打字，一直在用Typeless。
前陣子Typeless把免費字數從 8000 砍到 2000，對每天講很多話的人來說，幾天就用完了。

所以我乾脆自己做了一個：UTUVO Type。
iPhone 上任何 App 都能用說的打字，講完自動整理贅字、改口、標點，免費下載。

老實說，Typeless 背後是一整個團隊跟伺服器，UTUVO Type 是免費的、我也沒有自己的伺服器，要完全追上它真的有難度，而且好難。
但我努力往它靠近：
・想要專有名詞更準，可以用自己的 Groq key（有免費方案）開「雲端辨識」，內部測試專有名詞聽對的數量從 9 個變成 46 個
・常講的人名、術語加進字典，辨識會參考
・不想上雲，也可以全部在手機上處理

影片是完整教學，從安裝、加鍵盤、到設定 Groq，2 分多鐘一次看完 👇
希望大家用得開心，有問題或想要的功能都歡迎留言跟我說 🙏

App Store 搜尋「UTUVO Type」
https://t.co/EV6RCCDTJG

這篇文章就是用Utuvo Type 打出來的，我覺得還算準確。

#UTUVOType #語音輸入 #iPhone鍵盤 #Typeless #效率工具 #Groq

[Micky Yang @yangmicky]

Worth noting the same erosion tends to hit the awards too — the moment a category stops being a talking point, juries quietly stop separating the entries. If AI film gets folded into Film, the credit follows the film, and the process detail becomes a footnote nobody has a reason to read.

[Micky Yang @yangmicky]

A residual monitor is the piece most de-artifact chains skip — being able to hear exactly what got pulled is what keeps cleanup from quietly eating the parts you meant to keep. Does the new preset set stay conservative on transients, or is it willing to trade some crack for cohesion?

[Micky Yang @yangmicky]

Someone turned 300 Muse usage tricks into a GitHub repo. 155 stars in one day. 🔥

One problem: it's all in Simplified Chinese, built for mainland-China scenarios — https://t.co/dlZ0g5H93U price protection, 12306 train tickets. Useless if you don't live there.

So I made the English global version: 30 everyday tricks that work anywhere. Gmail triage, flight price watch, subscription audit, meeting notes → todos…
Every trick tested. Where Muse can't do it, I say so upfront — no hype.

👉 https://t.co/LmXHWd9719

(Taiwan / Traditional Chinese edition: https://t.co/zOajq1kxob)

New to Muse? Use my invite code: Y2XZ9N

#Muse #AI #productivity

[Micky Yang @yangmicky]

@manishs @BmakinMusic The liner notes point is the honest test — if the process only shows up in a description nobody opens, the distinction was never really there. And the labels argument cuts both ways: once the tooling is ordinary, the people still arguing about it are mostly arguing.

[Micky Yang @yangmicky]

@nqlogs 兩千億 token 的體感確實比任何跑分都準。不過雷霆大思考這段時間成本如果算進去，廉價高階這個說法要打折——同一個任務在慢思考下 token 消耗會明顯往上走，你實際用下來每輪平均多燒多少？

[Micky Yang @yangmicky]

@yangsanqia28377 算力成本壓根在晶片不在電費，這點把補貼敘事整個拆掉了。但結論要成立還缺一組數據：租出去了、回報期算不算得出來。海外客戶買的是延遲和合規，不是便宜的電，中間這段溢價才是真門檻。

[Micky Yang @yangmicky]

@Achgdesigner 單進程跑 500 個智能體、記憶總量 1.4GB，密度確實漂亮，但這條路的代價是故障域也一起合上了：一個 agent 的記憶檢索崩了，旁邊 499 個能不能活？你們在這層怎麼切？

[Micky Yang @yangmicky]

@Chris62771610 对，爆款决定你能被看见一次，能不能持续接住，才决定这些人是留下还是划走——后者才是真正难的部分。

[Micky Yang @yangmicky]

@aenewsEnglish24 Regulation usually ends up banning in public spaces long after the tech is already in headphones and pockets, so the interesting question is where the line gets drawn once the device is passive rather than recording.

[Micky Yang @yangmicky]

@ppollorg Open-ear is a compromise, not a replacement. Without a seal you lose the low end almost entirely, which is the part people actually miss on commutes.

[Micky Yang @yangmicky]

@tofudestiny @TrieuMessi The hardest thing to hear in a loud mix is space. Most rough demos have everything at the same distance, and that flattens whatever the artist was going for.

[Micky Yang @yangmicky]

Spot on, Ben. As a mixer and music producer, I've seen the exact same cycle play out with Pro Tools, auto-tune, and spatial audio: at first, everyone focuses on the novelty or the tool itself, but once the craft matures, all that matters is emotional impact and resonance.

Audiences ultimately respond to how a piece makes them feel, not the workflow used to render it. We’ll probably see an interim stage where festivals create dedicated sandbox categories to evaluate AI prompt-craft, but absorption is inevitable. The human ear and heart are indifferent to the pipeline."

[Micky Yang @yangmicky]

@Huahuazo Wan2GP 这种聚合思路很实用，把显存调度和模型切换藏在一个Web后面，省去了自己折腾环境的时间。本地跑AI视频的最大成本从来不是算力，是配置。

[Micky Yang @yangmicky]

@Jet_Woo903 soft rich 的关键确实是单点变现，团队和传统模式都是重资产路径。AI把这个门槛又拉低了一档——一个人加一个agent就能做微型产品，差距还在继续缩小。

[Micky Yang @yangmicky]

@CCTechNotes1917 文字水印走法律强制这条路很聪明：他们自己都说鲁棒性有限，那就别指望技术本身挡住人，直接靠合规义务铺满欧盟。音频早就能验，文字这层本质是补法律缺口。

[Micky Yang @yangmicky]

@bkarishma360 @SecondNameAxel That line keeps blurring for a good reason — the useful bar has always been whether it can act on the information well enough, not whether it groks it the way we do. Most grounding debates get lost in that gap.

[Micky Yang @yangmicky]

@eater188888 每天100-200的斜率对蓝V起步期够用了，重点还是你说的真诚互动——刷出来的数字留不住，聊出来的才会互相串门。