Built for production AI applications.
Speech-to-text for uploads and live streams, live captions and live translation in 14 languages, text-to-speech with 34 voices, text tools, chat, coding, vision, document OCR, image generation, sound effects, music and speaker diarization — all with one API key.
Speech-to-Text (Hindi, Hinglish & more)
Upload a recording and get the transcript. Hindi comes back in Devanagari with English technical words in Latin script; other languages are supported too. Speech can also be turned into English text, and word timestamps are available.
Fast File Transcription
Fast transcription of recordings and videos up to 100 MB in English, Spanish, French, German, Portuguese and Russian, cut at pauses, with English numbers written as digits. Speech detection returns the start and end time of each stretch of speech.
Realtime Speech-to-Text
Stream audio over a WebSocket with OpenAI-realtime-style events and receive partial and final transcripts as people speak, in 14 languages. Send a commit to finalize at once.
Live Captions & Live Translation
Stream a speaker's audio over a WebSocket and receive partial and final captions in the spoken language; a final arrives about 1 s after the speaker pauses. Send each final to the translation API to show it in any of 14 languages as people talk — numbers, IP addresses and commands kept intact.
Text-to-Speech (TTS)
34 voices with OpenAI-style requests in English, Spanish, French, German, Brazilian Portuguese and Hindi: delia, maeve, freya, easton, russell, edwin (English); noa, marcos (Spanish); noemie, augustin (French); nele, anton (German); manuela, davi (Portuguese); and 20 Hindi voices with ahana as the default, which can also speak Indian-accented English. Or design a voice: describe accent, age, tone and role, and each request invents a matching speaker in any of 14 languages.
Translation & Text Tools
Translate text between English, Hindi, Spanish, French, Chinese, German, Arabic, Brazilian Portuguese, Russian, Japanese, Korean, Italian, Dutch and Vietnamese, and Tamil, with meaning, numbers, IP addresses, commands and placeholders kept intact — streamed for live captions. Turn spoken numbers into digits in English and Hindi ("open port four forty three" → "open port 443"), create embeddings and classify text.
Chat, Coding & Vision
Always-on chat (chat) and coding (coding) with OpenAI-style requests, tool calls and streaming; plus vision, loaded on first use.
Document OCR
Turn PDFs and scanned or photographed pages into markdown, page by page: headings, lists and tables (as HTML), with optional block positions for each page. Tested on English, Hindi and Chinese documents. Always on; runs as a job you poll.
Image Generation & Editing
Create pictures from a text prompt, or edit an uploaded picture with a prompt. Sizes from 256 to 1024 px per side; the same prompt, size and seed give the same picture. The default image model is always on; a second image model loads on demand.
Sound Effects, Music & Songs
Generate sound effects or instrumental music clips from a text description, 1 to 120 seconds, returned straight away as a WAV file. For songs with vocals, send a style and tagged lyrics and download the finished song as M4A or FLAC.
Speaker Diarization
Find who spoke when in a recording, for up to 8 speakers. Returns the duration, the speakers and start/end segments per speaker, as JSON or RTTM.
Paused & Not Offered
Speech-to-speech and streaming TTS (sound-stream), the voice studio routes, cloning from your own clip and the old Indic voice host are paused (voice design works on the text-to-speech route). Voices outside English, Spanish, French, German, Brazilian Portuguese and Hindi are not available; live captions are not translated.