Skype Translator hands-on: fairly accurate speech recognition, but translation needs work
Skype Translator Hands-On: Close But No Babel Fish — When Skype announced its real-time translation program back in May, most of us seized on the sci-fi-ness off it all—Star Trek's universal translator, Babel fish, etc.
Context & Ripple Effects
This hands-on lands weeks after Microsoft opened its real-time translation program to early testing, and the verdict splits the problem in two: the speech recognition layer is judged fairly accurate, while the translation itself lags. That split matters because it defines what each competitor has to fix first.
The arc that follows is rapid: Google counters within weeks with a Translate app update adding automatic transcription for popular languages, and Microsoft keeps widening access until the preview requires no sign-up and eventually ships to every Skype user on Windows.
First-order effects
- Microsoft now has a public quality baseline to beat before general release: accurate transcription but imperfect translation means the hard engineering work sits in the language models, not the audio pipeline.
- Early testers get a working preview, but the rough edges cap everyday use — this is a credibility test for Skype's flagship feature, not yet a finished product.
Second-order effects
- Google's fast-follow Translate announcement shows the move forced a competitive response on mobile, where automatic transcription targets exactly the weakness Gizmodo flagged in Microsoft's implementation.
- Microsoft's answer to the criticism is distribution, not just model tuning: expanding from invite-only to open preview and then baking translation into the Windows client removes friction faster than incremental accuracy gains alone would.
Third-order effects
- If the pattern holds, real-time translation shifts from an opt-in novelty to a default capability embedded in communication apps — the endgame visible when translation reaches all Skype for Windows users and later extends to multi-device, in-person conversations.
- Speech recognition and translation decouple into separate competitive layers, letting platform owners like Microsoft and Google iterate each independently and bundle whichever wins into their existing call products.
The trend: Real-time machine translation is moving from sci-fi demo to built-in feature in mainstream communication apps, with Microsoft and Google racing to widen access ahead of perfecting accuracy.