I looked up what nine speech APIs charge per million characters. Here is what repeated lines really cost.

eyeguide-dev — Thu, 29 Oct 2026 03:17:00 +0000

I make Speakvora, a speech API with offline voice bundles, so read this with that in mind. I tried to keep it fair.

I wanted to know what text to speech costs at the price-list level, so I read the pricing pages of nine providers on October 8, 2026. All numbers below are list prices per million characters. They change, so check them before you decide.

Provider and model	Per 1M characters	At 10M characters
Amazon Polly Standard, Google Cloud Standard	$4	$40
OpenAI tts-1, Deepgram Aura-1, Fish Audio, Inworld Flash	$15	$150
OpenAI tts-1-hd, Rime Mist v3	$30	$300
ElevenLabs Flash / Turbo	$40	$400
ElevenLabs v2 and v3	$80	$800

At 10 million characters a month that is $40 at the cheap end and $800 at the expensive one. These models do not sound the same, so this is cost only.

The part price lists hide

Most apps say the same lines again and again: greetings, confirmations, error prompts. If you generate each line once and keep the file, you are not paying per play at all. A million characters is very roughly 15 to 20 hours of speech, so even at ElevenLabs v3 prices, 300 distinct 60-character prompts is 18,000 characters, about a dollar and a half.

Where per-character billing really hurts is when the text is different every time, or when your code calls the API on every request instead of caching. If that is you, fix the caching before you switch providers.

When pay-once makes sense

The other case is where you need audio to work with no internet, or you want a fixed price for a standard set of lines for one industry. That is what our one-time bundles are for (audio from $14). It is the only part where I would say we sell something the big providers do not.

I put the full table, with the source link and the date I read each page, here: https://speakvora.com/cheapest-text-to-speech-api

What I would not claim

Our voices are AI-generated and we are smaller than the big names, with fewer models and languages. Listen to samples before you pick on price.

What do you pay per million characters today, and do you cache the audio?

Add spoken status messages to a web app without exposing your API key

eyeguide-dev — Thu, 08 Oct 2026 11:48:18 +0000

Spoken confirmations ("Your order is ready for pickup") are easy to add, and easy to get wrong by shipping an API key in front-end code. Here is a small pattern that keeps the key on your server.

I work on Speakvora, a speech API. It is in early access (free allowance first, then pay as you go), and this post is mainly about the pattern; the same shape works with any text-to-speech API that returns an audio link.

The idea

The browser never talks to the speech API. It asks your server to speak a sentence; your server calls the API with the key and sends back only the audio URL.

1. Server route (Node 18+, Express)

import express from "express";
const app = express(); app.use(express.json());

app.post("/api/speak", async (req, res) => {
  const text = String(req.body.text || "").slice(0, 500);
  const r = await fetch("https://speakvora.com/api/v1/speak", {
    method: "POST",
    headers: { "x-api-key": process.env.SPEAKVORA_KEY, "content-type": "application/json" },
    body: JSON.stringify({ text, voice: "af_heart", lang: "en" }),
  });
  const j = await r.json();
  res.status(r.status).json(r.ok ? { url: j.url } : { error: j.error });
});
app.listen(3000);

2. Browser

async function say(text) {
  const r = await fetch("/api/speak", { method: "POST", headers: { "content-type": "application/json" }, body: JSON.stringify({ text }) });
  const j = await r.json();
  if (!r.ok) return console.error(j.error);
  new Audio(j.url).play(); // must follow a user gesture such as a click
}
document.querySelector("#go").onclick = () => say("Welcome back.");

3. The errors worth handling

402 free_allowance_used: the free allowance is spent.
429 rate_limited: wait retry_after_seconds and retry.
404 not_in_library: that voice only plays pre-recorded library lines; pick a live voice.

Things I would do next

Show a loading state. Phrases already in the library come back almost instantly; a brand-new sentence is generated and takes longer.
Keep the returned URL and reuse it rather than asking for the same sentence again.
Put a length limit and a rate limit on your own /api/speak route, because it spends your quota.

Full version with the Python and cURL variants: https://speakvora.com/guides/add-speech-to-a-web-app?utm_source=devto&utm_medium=article&utm_campaign=tutorial-add-speech

Free key (no card): https://speakvora.com/dashboard. Questions: support@eyeguide.ai

DEV Community: eyeguide-dev