Can I vibecode ekto?

KINDA · weekend project

price $29.99/moyou'd save $359.88/yrbuild time a weekendcategory 🔊 voice aireplaced by 0 people

The pipeline is no longer exotic: capture mic audio, segment it with a voice activity detector, transcribe with Whisper, translate, speak it back with a local TTS voice. An agent can wire that into a working local app in a weekend and it will genuinely translate a conversation. What it will not do out of the gate is stay graceful for an hour: chunk boundaries clip words, speaker turns bleed together, latency creeps as the buffer grows, and the sentence by sentence pacing that makes these apps usable in real conversation is a tuning problem, not a coding problem. You also get no phone app, which is where voice translation actually happens. Fine for a desk setup and travel prep, unconvincing when you are holding it out to a stranger in a market.

the prompt

Build a local real-time voice translation app. No accounts, no cloud services, no telemetry.

Stack, non-negotiable:
- Python 3.11 + FastAPI, served with uvicorn on port 8000.
- One HTML page with vanilla JS, no framework, no build step.
- Audio in: browser getUserMedia, 16kHz mono, streamed to the server over a WebSocket in 250ms PCM chunks.
- Speech to text: faster-whisper (small model default, configurable via .env).
- Segmentation: silero-vad or webrtcvad to detect end of utterance. Do not translate on fixed timers, translate on detected utterance boundaries.
- Translation: argostranslate with locally installed language pairs.
- Text to speech: piper, one voice per target language, downloaded on first run into ./models.

Behavior:
- User picks source and target language in a dropdown before starting.
- Press Start, speak, and on each detected utterance the server returns: original text, translated text, and a WAV of the translated speech. The page appends both lines to a running transcript and plays the audio.
- Show live latency per utterance in ms in the corner. Be honest, measure end of speech to audio ready.
- Handle overlap: if a new utterance arrives while audio is playing, queue it, never drop it.
- Long session hygiene: cap the in-memory transcript at 500 lines, reset the whisper buffer after every utterance, log RSS every 60 seconds.

Out of scope, do not build: mobile app, user accounts, cloud sync, speaker diarization, a two-phone conversation mode.

Deliverables: main.py, static/index.html, static/app.js, requirements.txt, .env.example (WHISPER_MODEL, DEVICE, COMPUTE_TYPE), scripts/download_models.py, and a README with exact run steps plus one paragraph on where this degrades in sessions over 20 minutes.

Run it, speak a test sentence in English with Spanish as target, and paste the measured latency into the README.

$ open in your agent (prompt prefilled, you press enter) or copy it raw · this prompt is generated from the build plan · improve it via PR

why people still pay

Because voice translation is judged entirely on the seconds between someone finishing a sentence and you hearing it, and on whether it still works on minute 40. A local build nails the demo and then frays: barge-in, background noise, two people talking over each other, the phone locking. Paying gets you a phone in your pocket that handles those cases without you adding VAD thresholds mid-conversation.

what you lose

xLong session reliability: memory growth, drifting segmentation and dropped turns after the first 20 minutes

xClean sentence by sentence pacing and turn detection, which is most of the perceived quality

xA mobile app, so no translating anything while standing up

xOffline or low-bandwidth behavior tuned for actual travel

xLatency budgets someone else already fought for: streaming partial results instead of waiting for a full segment

prior art · use these instead of building, if you'd rather

Nothing worth pointing at. That's why the prompt exists.

share on X ↗"I just replaced ekto ($29.99/mo) with one prompt"