audivra docs
Start free

Or jump from this list on a smaller screen.

This is a page from the Audivra documentation. Developer exports — index: http://127.0.0.1:3000/docs/llms.txt, full corpus: http://127.0.0.1:3000/docs/llms-full.txt, OpenAPI: http://127.0.0.1:3000/docs/openapi.json, Postman: http://127.0.0.1:3000/docs/postman.json. Press ⌘K to search docs.

Audivra Documentation

API-first voice and digital human infrastructure — text-to-speech, lip-synced video, voice cloning, and conversational agents.

Audivra provides AI voice and avatar infrastructure for creators and developers. Generate speech, talking-head video, and agent responses through a unified REST API, a no-code dashboard, or a self-hosted open-source stack on your own GPU.

At a glance

TopicAudivra
Lip-sync modelsmusetalk, latentsync, liveportrait, sadtalker, emo, codeformer
Max video durationFree 3 min · paid up to 30 min
Credits1 / audio character · 10 / video second
Generate rate100 req/min on paid plans
BatchJSONL or JSON · max 50 / 200 / 500 by plan
SDKsTypeScript (Node 18+) and Python 3.10+
Free trial12,000 credits/mo · no credit card · 3 min media cap
Sample media/api/docs/samples/example-video.mp4 · /api/docs/samples/example-audio.wav

Developer exports

Machine-readable docs and API specs for Cursor, IDE plugins, Postman, Insomnia, and client codegen — kept in sync via verify:docs.

llms.txt

Doc index for AI agents and IDE plugins

http://127.0.0.1:3000/docs/llms.txt

Download Markdown

llms-full.txt

Complete documentation corpus in a single Markdown file

http://127.0.0.1:3000/docs/llms-full.txt

Download Markdown

openapi.json

OpenAPI 3.0 — Postman, Insomnia, and client codegen

http://127.0.0.1:3000/docs/openapi.json

Download JSON

postman.json

Importable Postman Collection v2.1

http://127.0.0.1:3000/docs/postman.json

Download JSON

View export reference · Step-by-step guides: Guides · Try in browser: TTS · Generate · Model comparison: Meet the models · Voice library: Browse voices · OpenAPI reference

Search documentation

Client-side search over nav pages, section headings, guide steps, and API endpoints. Press ⌘K anywhere in docs.

Try it

Interactive API playgrounds embedded in docs — session auth or pasted sk_live_ key, live audio preview, and async job status.

Text to speech

Hear WAV or Opus output in the browser — no curl required.

Open playground →

Unified generate

Submit async avatar jobs and watch live pipeline status + video preview.

Open playground →

Speech to text

Upload a clip and read the transcript in the docs.

Open playground →

Music

Score beds — play a sample on the homepage playground, then mint a key.

Open playground →

Sound effects

Short cues — hear one on the homepage, then call POST /api/v1/sfx.

Open playground →

Meet the models

Pick the right engine for latency, language coverage, VRAM footprint, and plan tier — Kokoro for regional TTS, F5/XTTS for cloning, MuseTalk for lip-sync, or local Ollama on your own GPU.

Compare all models

Popular picks

The four engines developers compare most when choosing Audivra for production.

Kokoro-82M

kokoro-82m

Apache 2.0

Latency
Ultra-fast
VRAM
Low (T4)
Languages
en-US, sw-KE, ha-NG +8 more
Max text
5,000 chars
Tier
All plans

Full specs →

F5-TTS

f5-tts

MIT

Latency
Streaming
VRAM
Med
Languages
en-US, Clone any language (Starter+)
Max text
5,000 chars
Tier
All plans

Full specs →

MuseTalk

musetalk

Video lip-sync

Latency
Moderate
VRAM
Med
Languages
Language-agnostic (audio-driven)
Max text
5,000 chars
Tier
All plans

Full specs →

Ollama — Qwen2.5 3B

ollama-qwen

Fast local LLM

Latency
Local (<500 chars)
VRAM
~2.5 GB
Languages
Multilingual
Max text
500 chars
Tier
Self-hosted

Full specs →

Speech (TTS)

Open-weight text-to-speech engines for API and dashboard playback.

Kokoro-82M

kokoro-82m

Apache 2.0

Latency
Ultra-fast
VRAM
Low (T4)
Languages
en-US, sw-KE, ha-NG +8 more
Max text
5,000 chars
Tier
All plans

Full specs →

Kokoro

kokoro

Apache 2.0

Latency
Fast
VRAM
Low
Languages
en-US, en-GB
Max text
5,000 chars
Tier
All plans

Full specs →

F5-TTS

f5-tts

MIT

Latency
Streaming
VRAM
Med
Languages
en-US, Clone any language (Starter+)
Max text
5,000 chars
Tier
All plans

Full specs →

XTTS v2

xtts-v2

Coqui

Latency
Clone
VRAM
Med
Languages
ja-JP, 17+ clone languages
Max text
5,000 chars
Tier
Starter+

Full specs →

CosyVoice 2

cosyvoice-2

Apache 2.0

Latency
Expressive
VRAM
High
Languages
fr-FR
Max text
5,000 chars
Tier
Creator Pro+

Full specs →

Video & lip-sync

Talking-head and avatar pipelines — billed per video second.

MuseTalk

musetalk

Video lip-sync

Latency
Moderate
VRAM
Med
Languages
Language-agnostic (audio-driven)
Max text
5,000 chars
Tier
All plans

Full specs →

LatentSync

latentsync

Fast lip-sync

Latency
Fast
VRAM
Low
Languages
Language-agnostic (audio-driven)
Max text
5,000 chars
Tier
All plans

Full specs →

LivePortrait

liveportrait

Photo expressions

Latency
Moderate
VRAM
Med
Languages
Language-agnostic (audio-driven)
Max text
5,000 chars
Tier
Starter+

Full specs →

SadTalker

sadtalker

Audio-driven portrait

Latency
Moderate
VRAM
Med
Languages
Language-agnostic (audio-driven)
Max text
5,000 chars
Tier
All plans

Full specs →

EMO

emo

Expressive generation

Latency
Slow
VRAM
High
Languages
Language-agnostic (audio-driven)
Max text
5,000 chars
Tier
Creator Pro+

Full specs →

CodeFormer

codeformer

Face enhancement

Latency
Fast
VRAM
Low
Languages
N/A (post-processing)
Max text
5,000 chars
Tier
Starter+

Full specs →

Local / self-hosted

Fast-path routing on your hardware before cloud GPU offload.

Ollama — Qwen2.5 3B

ollama-qwen

Fast local LLM

Latency
Local (<500 chars)
VRAM
~2.5 GB
Languages
Multilingual
Max text
500 chars
Tier
Self-hosted

Full specs →

Ollama — Llama 3.2 3B

ollama-llama

Fast local LLM

Latency
Local (<500 chars)
VRAM
~2.5 GB
Languages
Multilingual
Max text
500 chars
Tier
Self-hosted

Full specs →

Piper TTS

piper-tts

CPU-friendly TTS

Latency
Ultra-fast
VRAM
CPU
Languages
en-US, Regional packs
Max text
500 chars
Tier
Self-hosted

Full specs →

Browse voices

Play samples and copy voice_id for Swahili, Hausa, Yoruba, Hindi, and more — the fastest path from docs to your first API call.

Browse all voices →
African regional
sw-KE

Zuri

Swahili — WAXAL / OpenBibleTTS

zuri-sw
African regional
ha-NG

Umar

Hausa regional voice

umar-ha
African regional
yo-NG

Ezinne

Yoruba regional voice

ezinne-yo
South Asian
hi-IN

Swara

Hindi — low VRAM Kokoro-82M

swara-hi

Guides

Step-by-step tutorials for creators, developers, and self-hosters — reference tells what; guides tell how.

Creators
~5 min

First talking avatar in 5 minutes

Create a lip-synced talking-head video from text — dashboard or API — and download the MP4 when the job completes.

Start guide →
Creators
~8 min

Dub a talking head

Upload source speech, translate into a target language, and lip-sync a talking-head with a catalog voice.

Start guide →
Developers
~10 min

Stream Opus TTS on low bandwidth

Deliver chunked Opus audio for mobile networks in the Global South — 48 kbps streams instead of multi-megabyte WAV files.

Start guide →
Self-hosters
~30 min

Self-host on an RTX 4090

Run Ollama, Kokoro-82M, and MuseTalk on a local RTX 4090 — route short prompts on-box before cloud GPU offload.

Start guide →
Developers
~20 min

Accept M-Pesa via Paystack webhook

Let Kenyan users pay with M-Pesa through Paystack — checkout redirect, webhook verification, and plan upgrade on charge.success.

Start guide →
Creators
~15 min

Professional voice clone

Train a higher-quality clone from 5–25 samples covering 2–30 minutes of speech.

Start guide →
Creators
~10 min

Share a community voice

Publish a ready clone to the marketplace and let others use the listing as voice_id.

Start guide →
Creators
~4 min

Media content tips

Prepare video and audio so lip-sync looks natural — faces, noise, and duration mismatch.

Start guide →
Developers
~3 min

Media formats

Supported containers, how to pass URL vs upload vs asset_id, and plan size caps.

Start guide →
Developers
~3 min

Generation times

What to expect from async lipsync jobs and how to poll or wait on webhooks.

Start guide →

Choose your path

Audivra Studio

No-code dashboard for creators — playground, avatar studio, voice library, keys, and billing.

First talking avatar guide

Audivra API

REST + WebSocket API with hashed keys, async jobs, and metered credits for embedding in your app.

API quickstart

Self-hosted stack

Run Ollama, Kokoro-82M, and MuseTalk on local or RTX hardware with fast-path routing before cloud GPU.

RTX 4090 guide

Global South payments

Global South users favor Mobile Money (M-Pesa, MTN, Airtel), instant bank rails (Pix, UPI), and e-wallets — not credit cards alone.

M-Pesa via Paystack

Enterprise & private cloud

VPC isolation, on-prem GPU clusters, air-gapped inference, and SLA-backed uptime for B2B teams.

Enterprise deployment

Voices

Preset library, clones, remixes, designed voices, and community listings — each has a voice_id for TTS and generate.

Avatars

avatar_id selects a talking-head preset. Lip-sync models include MuseTalk, LivePortrait, and SadTalker.

Models

Compare Kokoro, F5-TTS, MuseTalk, and local Ollama — latency, languages, character limits, VRAM, and tier gates.

Credits

1 audio character = 1 credit · 1 video second = 10 credits · 1 STT second = 5 credits. Estimate a mix, then checkout on Pix, UPI, or M-Pesa — see /docs/concepts/credits.

Pricing

Video-first plans, resolution gates, Studio/Team, Private GPU, and studio grants — all from PRICING_TIERS.

How Audivra works

Audio and video on GPUs takes 2–30 seconds. The API returns a job ID immediately and processes in the background via Celery/Redis or Modal serverless GPUs — preventing HTTP timeouts.

SaaS (Next.js)Dashboard, playground, API keys, billing, developer console
Gateway (FastAPI :8080)Auth, rate limits, Celery queue, Stripe meters
Inference (:8090 / Modal)Kokoro TTS → MuseTalk lip-sync → FFmpeg MP4
DataSupabase Postgres + RLS, Redis, S3/R2 CDN

Browse by capability

Text to speech

Generate lifelike speech — try WAV or Opus streaming in the browser on the API page.

Speech to text

Transcribe uploads or a public audio URL — billed per second.

Music

Score beds under talking-heads — billed per second from the same prepaid pool.

Sound effects

Short cues muxed with lip-sync — billed per clip.

Dubbing

STT, translate, TTS, and lip-sync as one async job.

Voice library

Three voice types — presets, clones, and prompt-based design.

Talking avatars

Async lip-synced MP4 from text — poll jobs or subscribe to SSE events.

Unified generate

Text→avatar or video+audio lipsync (URL, upload, asset_id) with sync_mode.

Media assets

Reusable library IDs for generate and batch.

Voice cloning

Instant or professional multi-sample clones on paid plans.

Voice remix

Prompt-conditioned variants of clones you own.

Community marketplace

Publish clones and license community voice_id listings.

Voice design

Describe a voice in text — preview and save on paid plans.

Voice agents

Turn-based conversational agents with TTS output.

Webhooks

Stripe, Paystack, Flutterwave, and dLocal — payload schemas, signatures, retries, and idempotency.

Errors & limits

Central HTTP error codes and quota reference — 401/402/429/502, credits, routing, rate limits.

Low-bandwidth streaming

Opus/AAC 48k chunks for mobile networks in the Global South.

Self-hosted inference

Route short prompts to local Ollama and Kokoro before cloud GPU.

Enterprise deployment

Private VPC, on-prem GPU, air-gapped inference, and SLA tiers for B2B.

Pricing

Talking-head minutes, resolution gates, local-rail checkout, Private GPU, and studio grants.

Changelog & API v1

Release history and stability promise for /api/v1/* integrations.

SDKs

Thin TypeScript + Python clients for TTS, unified generate, voices, and jobs.

Quick examples

curl -X POST https://audivra.dev/api/v1/tts \
  -H "Authorization: Bearer sk_live_…" \
  -H "Content-Type: application/json" \
  -d '{"text":"Hello from Audivra","voice_id":"aurora","model":"kokoro"}' \
  --output speech.wav