Audra four engines guide: Essential, Plus, Pro, Conversations — Audra

Audra exposes four public engines on one OpenAI-compatible API — Essential, Plus, Pro, and Conversations. Each engine maps to a model id (audra-essential, audra-plus, audra-pro, audra-conversations) with a fixed billing multiplier on the same $25/1M prepaid base rate.

Essential (1.0×) — default batch narration

audra-essential is the entry tier: 72 voices across 10 languages (English US & UK, Japanese, Mandarin Chinese, Spanish, Portuguese Brazil, Hindi, French, Italian, German, and Dutch), 50k sync chars per request, −16 LUFS MP3 exports, and QA-gated billing. Pick this for podcasts, courses, YouTube narration, and cron-driven MP3 jobs. Optional language (BCP-47) on POST /v2/speech is inferred from slug when omitted. Preview voices on /voices — june, arthur, and grant are common starting points.

Plus (1.1×) — multilingual CPU

audra-plus adds 28 multilingual CPU presets and light cloning workflows at 1.1× the base rate (~$27.50/1M at prepaid). Use for extended multilingual presets and cloning-lite without full GPU spend. Voice catalog expands beyond Essential — filter by engine on /compare.

Pro (1.25×) — GPU MOSS quality

audra-pro runs GPU MOSS inference with instruct control and podcast-oriented delivery at 1.25× (~$31.25/1M). Best for premium ads, character reads, and audiobook-style hosts where Essential presets feel too flat.

Conversations (1.75×) — multi-speaker shows

audra-conversations renders 2–5 speakers in one request at 1.75× (~$43.75/1M) — co-hosted podcasts, interviews, and drama without stitching separate files. Not a realtime agent stack; still batch HTTP with mastered MP3 output.

How to choose

Start on Essential, shadow-test another engine on staging scripts, then flip model id in production. Legacy ids (audra-core, audra-studio) still route — see /changelog.