Every provider, ranked cheapest first
| Provider / voice | Per 1M chars | Per 1k chars | Your monthly cost |
|---|
Why per-character pricing trips people up
You think in minutes of audio; the bill is in characters of text. That mismatch hides the real cost. A rule of thumb bridges it: people speak around 150 words a minute, a word averages about 6 characters with spaces, so a minute of speech is roughly 900 characters — and a million characters is about 18 hours of audio. This calculator converts whichever unit you have into characters and prices it across every major provider, so "we generate 40 hours of narration a month" becomes a number you can actually compare.
The market splits into tiers. Standard voices (AWS Polly Standard, Google Standard) are the cheapest at roughly $4 per million characters and sound fine for IVR and alerts. Neural voices (OpenAI TTS, Polly Neural, Google Neural2, Azure Neural) sit around $15–16 per million and are the default for most products today. Voice-clone and emotive engines like ElevenLabs lead on realism but cost several times more and are priced by subscription tier, so the per-character rate drops as you commit to a bigger plan. Pick the cheapest tier that clears your quality bar — the gap between standard and premium is often 25× for audio many listeners can't tell apart.
Building a voice product end to end? Speech is only one leg. Pair this with the speech-to-text calculator for the listening side and the AI voice agent calculator to add the LLM in the middle and get the full per-minute cost of a talking bot.
How to use it
1. Enter your monthly volume and choose whether it's in characters, words or minutes of audio.
2. The calculator converts everything to characters (the billing unit) for you.
3. Read the cheapest option and a representative neural option at the top.
4. Compare every provider in the ranked table, per 1k characters and per month.
Common mistakes
Estimating in minutes and forgetting the conversion. Always price in characters — that's what you're charged for. Paying for premium voices everywhere. Use a cheap standard voice for system messages and reserve the expensive clone for marketing or front-of-house. Ignoring free tiers. Most providers give a monthly character allowance that covers small projects entirely. Re-generating unchanged audio. Cache generated speech for static phrases instead of paying to synthesise them again.
FAQ
Is ElevenLabs really more expensive than OpenAI TTS?
Per character, usually yes — but it depends on your ElevenLabs tier, and many teams accept the premium for its more natural, emotive and cloneable voices. For neutral narration OpenAI, Polly Neural or Google Neural are far cheaper.
Do spaces and punctuation count as characters?
Generally yes — providers bill the raw input string, including spaces and punctuation. SSML tags may or may not count depending on the provider.
How many characters is an hour of audio?
About 54,000 characters at a natural pace (~150 words/min × ~6 chars/word × 60 min). So roughly 18.5 hours fit in a million characters.
Which is cheapest for high volume?
Standard voices from AWS Polly or Google at about $4 per million characters. If you need neural quality, the ~$15–16 cluster is your floor; negotiate enterprise rates above a few million characters a month.
Estimate only. TTS prices are reference figures and vary by voice, plan and region — verify in each provider's pricing page.