Podcast TTS API: batch narration, loudness, and per-episode costs — Audra

Podcast pipelines render the same shape every week: ingest a script, POST to a TTS API, drop a mastered MP3 into a feed. The API choice matters on three axes — per-episode cost, loudness consistency, and how much audio post-processing your workers run.

delivery_profile and podcast_v4

POST /v2/speech accepts delivery_profile for show-ready output. podcast_v4 applies podcast-oriented pacing and −16 LUFS loudness normalization on MP3 exports — so Apple Podcasts and Spotify targets are met without a DAW pass. Legacy profiles remain for backward compatibility; new shows should start on podcast_v4.

{
  "model": "audra-essential",
  "voice": "grant",
  "text": "Episode narration body…",
  "format": "mp3",
  "delivery_profile": "podcast_v4"
}

Cost per 20-minute episode

A narration-heavy 20-minute episode is roughly 12,000–18,000 characters (~150 words/min). At 15,000 chars and prepaid $25/1M, Audra costs about $0.38 per episode. ElevenLabs list (~$120/1M) lands near $1.80 for the same script — before subscriptions or unused credits.

Voice presets for shows

Map each show ID to a slug voice once — grant for documentary, ellis for UK hosts, cole for news reads. Preview all 150 voices on /voices with your actual intro copy before locking production presets.

Batch 50k characters per sync request for long episodes; chunk server-side only when scripts exceed that limit. Commercial rights unlock on first paid pack — header X-Audra-Commercial-Use: true on production renders.