Try it live — this is the real engine

Hear it. Read it. Instantly.

Real-time speech to text in 40+ languages — tap the mic and start talking. Your words are transcribed as you speak, by the same engine you'd run on your own hardware.

Tap to talk

Your words will appear here…

RUNS IN YOUR BROWSER · AUDIO SENT SECURELY OVER HTTPS

FIG. 01 — a live demo, not a video. Words appear as you speak.

40+world languages

12Indian languages

Livewords appear as you speak

27×faster on CPU

What it does

Everything you need from speech.

Accurate, real-time and private — built to drop into calls, meetings, apps and phone lines.

Real-time streaming

Words appear as they're spoken, with automatic end-of-utterance and backchannel detection so you know when a thought is finished.

40+ languages, auto-detected

One model, dozens of locales including strong Indian-language coverage. Pick a language or let it detect automatically.

Word-level timestamps

Every word comes back timed to the millisecond with a confidence score — ready for captions, search and compliance.

Phone & telephony ready

Handles 8 kHz call audio and noisy lines — perfect for contact centres, IVRs and call transcription.

Fast & tiny

Up to 27× faster than common alternatives, with about half the memory — runs comfortably on an ordinary CPU.

Private & on-device

Audio never has to leave your hardware. Ideal for healthcare, finance and any regulated workload.

40+ world languages

EnglishSpanishFrenchGermanItalianPortugueseDutchRussianArabicMandarinJapaneseKoreanVietnameseThaiTurkishPolishSwedish+ more

12 Indian languages

HindiBengaliTamilTeluguMarathiGujaratiKannadaMalayalamPunjabiOdiaAssameseUrdu

Build with it

Speech in, speech out. Both streaming.

OpenAI-compatible endpoints for real-time transcription and text-to-speech — one private key covers both.

Speech-to-text, in real time

Stream audio in and transcripts stream back over SSE as the words are spoken — 40+ languages, auto-detected, with word-level timestamps and confidence.

Text-to-speech

POST text to /audio/speech and natural 24 kHz audio streams back — English plus 12 Indian languages, female or male voice.

What you get

Streaming STT (SSE partials) + streaming TTS (audio)
Per-word start/end & confidence
language=auto or any supported locale
Self-hosted option — runs fully on-prem

Speech-to-text — stream audio in, transcript outcurl -N https://genie.superchat.in/g/<id>/v1/audio/transcriptions?language=auto \ -H "Authorization: Bearer $GENIE_KEY" \ -H "Content-Type: audio/webm" \ --data-binary @speech.webm # -> data: {"transcript":"Let's ship on Friday","lang":"en"}

Text-to-speech — text in, 24kHz audio outcurl https://genie.superchat.in/g/<id>/v1/audio/speech \ -H "Authorization: Bearer $GENIE_KEY" \ -H "Content-Type: application/json" \ -o reply.wav \ -d '{"model":"genie-voice","voice":"female", "language":"en","input":"Hello from Genie."}'

Python — text-to-speechimport requests r = requests.post("https://genie.superchat.in/g/<id>/v1/audio/speech", headers={"Authorization": "Bearer $GENIE_KEY"}, json={"model":"genie-voice", "voice":"female", "language":"hi", "input":"Namaste, main Genie hoon."}) open("reply.wav", "wb").write(r.content)