Speech-to-Text API in India

Many teams searching for a speech-to-text API in India actually need the whole voice-to-insight pipeline: audio in, clean and useful text out. It's worth being upfront about what unoblox is today — a text-in, text-out gateway (plus native /v1/embeddings and /v1/rerank) billed in rupees — rather than a speech-to-text engine itself. Pair your existing transcription engine with unoblox's chat models for the "understanding" half of that pipeline: cleanup, translation, and summarisation, all billed per million tokens in rupees.

What unoblox does today — and doesn't

unoblox does not host a dedicated audio-transcription model right now; the catalog is chat-completion models plus embeddings and reranking. If your product needs raw audio converted to text, that step happens in a separate speech-to-text engine you bring. What unoblox is genuinely good for is everything downstream of that transcript: fixing punctuation and disfluencies, labelling speaker turns you've already detected, translating a transcript, and summarising a call into notes or action items.

A practical pipeline

1. Your speech-to-text engine converts audio to a raw transcript
2. unoblox cleans up punctuation, filler words, and obvious transcription errors
3. unoblox translates or summarises the cleaned transcript as needed
4. unoblox extracts action items or a structured call summary for your CRM or ticketing tool

Models for the text layer

Task	Suggested model	₹ input / ₹ output (per 1M tokens)
Bulk transcript cleanup	deepseek-ai/deepseek-v4-flash	₹9.07 / ₹18.14
Multilingual transcript translation	qwen/qwen3-235b-a22b-instruct-2507	₹9.07 / ₹55.44
Meeting notes and action items	openai/gpt-5-mini	₹25.2 / ₹201.6
Prototyping the pipeline	qwen/qwen3-1.7b	FREE (₹0)

Sample call: transcript to clean notes

curl https://api.unoblox.ai/v1/chat/completions \
  -H "Authorization: Bearer ub-gw-xxxxxxxxxxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-5-mini",
    "messages": [{"role": "user", "content": "Clean up this raw call transcript and list three action items."}]
  }'

Billing, and what to check before you build

Every model here is priced per million tokens in rupees with a monthly GST invoice — no international card, one key, one endpoint. Because unoblox's catalog changes over time, check /models before you architect a product around a specific capability; if native audio transcription is added, it will show up there rather than in a claim made on this page.

Frequently asked questions

Does unoblox convert speech to text directly? Not today. unoblox is a text (plus embeddings and rerank) gateway — bring your own speech-to-text engine for the audio step, and use unoblox for cleanup, translation, and summarisation of the resulting transcript.

Why use unoblox at all if it doesn't do transcription? Because the "understanding" step after transcription — cleanup, translation, summarisation, action-item extraction — is real LLM work, and unoblox lets you run it through one rupee-billed endpoint across several model choices.

Can it translate a transcript into another Indian language? General multilingual models can attempt this with varying quality; test your specific language pair before relying on it in production.

Is there a free way to test the pipeline? Yes — qwen/qwen3-1.7b costs ₹0, enough to validate your cleanup and summarisation prompts.

Is pricing in rupees only? Yes, every model is priced and billed in rupees per million tokens.

Will unoblox add native transcription later? We're not going to promise a roadmap date here — check /models for the current, accurate list of what's available.

Get started in rupees → https://unoblox.ai/sign-in