Sonex Labs

Pāṇini TTS API

One API. 250+ languages. Streaming under 200 ms.

Text in, audio out. No orchestration, no language tiers, no quality drop outside English.

Start building freeRead the docs

250+

Languages, one model

< 200ms

Time to first byte

120

Platform voices

Quickstart

Live in minutes, not days.

One API call, streamed audio back. Same code in every language, leave language on auto and the engine detects and switches mid-sentence, Hinglish included.

synthesize.py
1import requests
2
3response = requests.post(
4    "https://api.sonexlabs.com/v1/speech/stream",
5    headers={"Authorization": "Bearer YOUR_SECRET_TOKEN"},
6    json={
7        "text": "नमस्ते, मैं आपकी कैसे मदद कर सकती हूँ?",
8        "voice_id": "lkwlapu6ab",
9    },
10    stream=True,
11)
12
13# first audio bytes arrive in under 200 ms
POST/v1/speechSynthesize full audio in one response.
POST/v1/speech/streamChunked streaming synthesis, sentence by sentence.
GET/v1/voicesList platform preset voices and your clones.
POST/v1/voices/cloneClone a voice from a file or public audio URL.
GET/v1/balanceWallet balance, credits and character expiry.

Authenticate with a bearer token or the X-Api-Key header. Keep the key server-side only. Testing tier limits: 5 requests/second burst, 120 requests/minute sustained, with a Retry-After header on 429.

Products

Two building blocks, one platform.

Pāṇini text to speech

Speech that code switches, emotes and breathes.

250+ languages, 120 platform voices, inline non-verbal tags, custom dictionaries and abbreviation handling, cloning from ten seconds of audio.

Explore text to speech

Sonex Flows

Orchestration for voice agents, without the glue code.

Build call flows the way you would in n8n: reusable API, MCP and database blocks, branching logic, telephony in and out, and live handoff to a human. Wire it to Pipecat or LiveKit, or run the whole pipeline on Flows.

Included with every account

Capabilities

Everything you need for production voice.

250

250+ languages, one endpoint

Bhojpuri, Garhwali, Marwari, Saraiki, Yoruba, Arabic, Mandarin and 200+ more. Single endpoint, single price, no separate models per language.

150

Sub-200 ms time to first byte

Audio starts arriving in under 200 ms from request. Built for real-time applications, no buffering, no batch waiting.

10s

Voice cloning in under ten seconds

Send a ten-second sample. The cloned voice speaks all 250+ languages with the original accent carried over, not flattened.

≋

Streamed generation

Audio streams sentence by sentence as it generates, so playback starts before the full response is synthesised.

Ah!

Non-verbal expressions

Render laugh, sigh, hm, aha, gasp and oops inline in your text. No separate API calls.

8k

Telephony-grade audio

Trained on real phone-call conditions, not studio recordings resampled after the fact.

Integrations

Works with how you already build.

Reusable API, MCP and database blocks connect to anything, not just the names below.

PipecatLiveKitSonex FlowsExotelTwilioVonageSoonPlivoSoonOpenAI (BYOK)Anthropic (BYOK)Gemini (BYOK)HubSpotSalesforcen8nZapierMake.comPostgreSQLWhatsAppRazorpayStripe

Building with Pipecat?

Install pipecat-sonex or the LiveKit plugin and swap in Pāṇini with two lines of config. Sub-200 ms TTFB, 250+ languages, no other changes to your pipeline.

Pricing

Simple pricing. No surprises.

Every language billed at the same rate. BYOK on LLM with zero markup. Free tier: 10,000 characters and one voice clone, no card required.

Pāṇini TTS API$0.0313 / min
End-to-end voice calling, bundled$0.055 / min
Start building free

FAQ

Common questions from developers.

The TTS API ($0.0313/min) is a pure text-to-speech primitive. Send text, get audio back. You own everything else in the pipeline. The voice calling plan ($0.055/min) is the full stack: STT, LLM via BYOK, Pāṇini TTS, telephony and workflow orchestration bundled into one.

Yes. Full BYOK. Bring your own OpenAI, Anthropic or Gemini contract and pay your contracted rates directly. Sonex adds zero markup on LLM usage.

Send a 10-second sample of any speaker in any language. Pāṇini extracts the voice characteristics and applies them across all 250+ supported languages. The cloned voice speaks Marathi with the same timbre and accent as the original Hindi speaker. No per-language re-cloning required.

Yes. Streaming is the default output mode across all 250+ languages. There is no batch mode requirement for non-English languages.

Burst is 5 requests per second, sustained is 120 requests per minute; both return 429 with a Retry-After header. Build retry-with-backoff keyed on that header. Ahead of go-live your workload migrates to the scaled inference tier with a separate pre-production sign-off.

For production deployments, high concurrency requirements, or private VPC and on-premises hosting, talk to us directly. Enterprise pricing and support structures are handled separately from self-serve.

Start building in the next ten minutes.

Free tier, no card, 10,000 characters included. Every language from day one.

Start building freeRead the docs