I Needed a Cheap TTS API. So I Built One.

Last year I was building a side project — nothing fancy, just an app that reads articles aloud. I tried a few TTS options and eventually landed on ElevenLabs because the voice quality was genuinely good. Sentences didn't sound robotic. Pauses landed in the right places. It felt natural.

Then I looked at the billing page.

At $0.165–$0.22 per 1,000 characters, my little app was going to cost me $300–$500/month at moderate usage. Before I had a single paying user. That wasn't going to work.

I looked at the alternatives. Google Cloud and Amazon Polly are cheap, but you can tell. The voices are fine for utilities — navigation prompts, system alerts — but anything you actually want people to listen to sounds off. OpenAI's TTS is decent but pricey for what you get, and they cap you at six voices.

So I built LeanVox. This is the thing I wish had existed.

What makes ElevenLabs expensive, exactly?

It's not a scam — ElevenLabs charges what the market will bear and puts serious money into model R&D. But their pricing is structured for businesses, not developers building something to see if it sticks.

Here's the math. Say you're generating 5 million characters a month — roughly 550 minutes of audio, a solid amount for a real app:

That's not a small difference. $25 vs $835 for the same amount of audio. Even if you're generous and say LeanVox is 80% as good as ElevenLabs (it's closer than that, but let's say), you'd need to be generating a lot of revenue before the gap justifies itself.

What LeanVox actually is

Two tiers, no subscriptions:

• Standard ($0.005/1K chars) — Natural speech, ~200ms latency, 10 languages. This covers most use cases.
• Pro ($0.01/1K chars) — Voice cloning from a 10-second audio clip, emotion tags, voice design from text descriptions.

You buy credits. Purchased credits never expire. Free signup credits ($0.50) expire after 30 days. That's it.

The emotion tags thing is worth calling out specifically. You can write speech like this:

"Are you serious? [gasp] I can't believe you did that. [laugh]
That's actually... kind of genius. [sigh] I hate you."

And it actually delivers those emotions rather than just reading the words flatly. It's the kind of thing you don't realize you want until you hear it.

Feature rundown

How to use it

It's a standard REST API. You POST text, get back a URL to an audio file. Here's the whole thing:

curl

curl -X POST https://api.leanvox.com/v1/tts/generate \
  -H "Authorization: Bearer lv_live_YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Hello! [laugh] This was easier than I expected.",
    "model": "pro",
    "voice": "af_heart",
    "language": "en"
  }'

You get back:

{
  "audio_url": "https://cdn.leanvox.com/audio/abc123.mp3",
  "duration_seconds": 3.1,
  "characters_billed": 47
}

Python

import requests

r = requests.post(
    "https://api.leanvox.com/v1/tts/generate",
    headers={"Authorization": "Bearer lv_live_YOUR_API_KEY"},
    json={"text": "Hello from Python!", "model": "standard", "voice": "af_heart", "language": "en"}
)
print(r.json()["audio_url"])

Or just use the CLI

brew install leanvox/tap/lvox
lvox auth signup          # free $0.50 credit, no CC
lvox gen "Hello world!" -o hello.mp3

That last command — from zero to audio file — takes about 30 seconds. I timed it.

Is it as good as ElevenLabs?

Honest answer: close, but not identical. ElevenLabs has been doing this longer and their voice catalog is deeper. If you're doing premium audiobook production or building a product where voice quality is the main selling point, ElevenLabs is still worth considering.

For most apps though? The gap is smaller than the price difference. Try both. We give you $0.50 free — generate a few hundred words, compare, decide.

Try it

Sign up at leanvox.com — $0.50 in free credits, no credit card required. That's 100,000 characters at Standard, or 50,000 at Pro with voice cloning. Enough to know if it works for what you're building.

LeanVox is in public beta. We're expanding the voice catalog and adding languages. Feedback welcome.