Voice & TTS API Pricing

Understanding TTS pricing units: characters vs. minutes

Text-to-speech APIs split into two billing camps: per-character (you pay for input text length) and per-minute (you pay for output audio duration). These aren't easily comparable without a conversion factor. The practical rule is that 1,000 characters of typical English prose produces roughly 60–65 seconds of spoken audio at a natural reading pace — so a per-character rate of $0.00027 per character translates to about $0.27 per minute of audio output. When comparing vendors, always normalize to per-minute or per-thousand-characters before drawing conclusions. The headline number is almost always in the units most favorable to that vendor's pricing structure.

Free tiers and the commercial use trap

Nearly every TTS provider offers a free tier, but free tier output is typically prohibited for commercial use — meaning you can't legally publish it in a monetized video, podcast, or product. ElevenLabs' free tier also adds a watermark attribution requirement. If you're building anything intended to generate revenue, even indirectly, you need a paid plan before generating production audio. The entry paid tier (typically $5–11/month) is cheap enough that there's no good reason to risk the terms-of-service violation for anything beyond testing and evaluation.

Model quality tiers and the cost-quality tradeoff

ElevenLabs offers multiple model tiers — Multilingual v2, Flash, Turbo — at roughly 2x price steps between them. The cheaper Flash/Turbo models cost about half as much per character and have sub-second latency, which makes them genuinely useful for real-time applications like voice agents and conversational UIs. For pre-recorded content — podcasts, explainer videos, narration — the quality gap between Flash and Multilingual v2 is real but often acceptable, especially for supporting voices (not the primary host voice). The cost-effective strategy for high-volume production is to render all secondary audio with Flash and reserve the premium model only for the output the listener is most likely to scrutinize.

Voice cloning: what's actually included at each plan tier

Instant voice cloning from a short audio sample is a mid-tier feature — it's not available on the lowest paid plan. Professional voice cloning, which produces higher-quality results from a longer training set, requires a higher tier still. If voice cloning is a core requirement for your project rather than a nice-to-have, check the specific plan gate before subscribing, because the cheapest plan that allows API access may not be the cheapest plan that allows cloning. Budget for the tier that includes the specific capability you need, not just general API access.