Pinch

The interesting thing about Pinch is that it solves two different problems with one wallet and one rate card. Dubbing is asynchronous and batch-shaped: you POST a video URL to https://api.startpinch.com/api/dubbing/jobs, Pinch transcribes it, translates with natural phrasing, synthesizes in the original speaker's voice and preserves the background audio, and you poll for a download link (output URLs are valid for 48 hours). Real-time is an open WebSocket at ws.startpinch.com where you stream microphone audio in and get either translated speech or translated captions back, chosen per session, with no SDK required. That second product is where the June 2026 Relay-1 model lives, covering 30+ languages on a single connection — a materially different engineering problem from dubbing a file, and Pinch's June 2026 benchmark is an unusual willingness to publish numbers on translation quality, intelligibility, naturalness, latency and speaker similarity rather than only word-level accuracy. Strengths start with the browser studio. You can inspect speaker identification, reassign or merge speakers, edit the transcript and the translation line by line side by side, see the estimated cost, and only then generate. Anyone who has paid a per-minute dubbing bill for a bad speaker map will recognize why that matters. Dubbing ships in 10 languages with no watermarks, up to 60 minutes per input, at $0.50/min of source media ($1.00/min with lip sync). The language shortfall versus competitors is real but the vendor says more are on the way and invites requests. The $5 signup credit (roughly 10 minutes of dubbing, no card required) is enough to judge one genuine file rather than a demo. Weaknesses are mostly about scale and platform. $0.50/min of source media compounds quickly across a back catalog — an hour of source is $30, and a 40-episode season of 45-minute episodes is north of $900. That's why the managed service exists: you send a link and target languages, Pinch fetches, dubs and puts a human reviewer on every file, quoted per project once the vendor knows volume and languages. The Desktop app, which translates your own mic into any meeting app, is macOS and early beta. There's a live translation sidecar for meetings, webinars and browser audio, and the docs list Claude and Zoom among the places it shows up, but you shouldn't treat Pinch as a meeting platform — it's a speech layer you route audio through. Where it fits: developers adding a localization or accessibility feature to a video, conferencing or language-learning product, because the REST dubbing API and the WebSocket translation API are the product rather than an afterthought. Creators and podcasters who care more about sounding like themselves in Spanish or Japanese than about covering 40 languages badly. Teams doing internal training or webinars across a handful of markets. Where it doesn't: subtitle-only work, where a captioning tool is cheaper; high-volume catalog localization on a tight