TTS & Speech-to-Text API Cost Calculator

TTS and speech-to-text API cost calculator

This calculator turns a monthly volume into a monthly bill for every major speech API. For text to speech, enter characters, words or minutes of audio per month and it prices ElevenLabs (pay-as-you-go and every plan), OpenAI, Google Cloud Text-to-Speech, Amazon Polly, Azure AI Speech and Deepgram Aura. For speech to text, enter audio hours or minutes and it prices ElevenLabs Scribe, OpenAI Whisper and the gpt-4o transcribe models, Deepgram Nova-3, AssemblyAI, Google Speech-to-Text, Amazon Transcribe and Azure. The table is sorted cheapest first, the cheapest option is called out, and the URL updates as you type so you can share the exact comparison.

Prices are list prices in US dollars for US regions, taken from each provider’s own pricing page on October 8, 2026, with a link to each source under the table. Where a provider bills by token and publishes no per-character or per-minute figure (OpenAI’s gpt-4o-mini-tts and Google’s Gemini TTS), it is left out rather than estimated.

Text-to-speech prices used

Speech-to-text prices used

Characters, words and minutes

Every text-to-speech API on this page bills by the character you send, and a character includes spaces, punctuation and newlines. Google and Polly also count SSML markup (Google excludes only <mark> tags), so a heavily tagged script costs more than its spoken text suggests. If you think in words, the calculator assumes 6 characters per English word including the space after it. If you think in finished audio, it assumes about 900 characters per minute, which is 150 spoken words a minute. Use the “count characters from your text” box to measure a real script instead: paste it, set how many times a month you generate something that size, and the count is used directly. The text never leaves your browser.

Speech-to-text is billed by audio duration, not by the length of the transcript. Two details change real bills: AssemblyAI and Google bill each audio channel separately, so a two-channel call recording costs twice as much; and AssemblyAI’s streaming product bills the whole WebSocket session including silence. Amazon Transcribe bills per second with no minimum, and Google rounds each request up to the nearest second.

Worked examples

• 100,000 characters a month (a few long blog posts read aloud): free on Google’s Standard, WaveNet, Neural2 and Chirp 3: HD voices, $0.40 on Polly Standard, $1.50 on OpenAI tts-1, $7.20 on ElevenLabs pay-as-you-go with the high-quality models.
• 1 million characters: still free on Google’s free allowances, $4 on Polly Standard, $15 on OpenAI tts-1 or Azure Neural, $30 on OpenAI tts-1-hd, and $79.20 on ElevenLabs v4 or Multilingual v2 ($39.20 on Flash).
• 10 million characters: $24 on Google Standard or WaveNet, $40 on Polly Standard, $150 on OpenAI tts-1, $270 on Google Chirp 3: HD, about $800 on ElevenLabs’ high-quality models.
• 100 hours of audio transcribed: $15 on AssemblyAI Universal-2, $18 on gpt-4o-mini-transcribe or Azure batch, about $21 on ElevenLabs Scribe v2, $25.80 on Deepgram Nova-3, $36 on Whisper or Amazon Transcribe, $100 on Azure real-time.

The pattern holds at every volume: the cheapest voices are not the ones people choose ElevenLabs for. Price is one column; listen to samples of the voices you would actually ship before deciding on it.

ElevenLabs plans versus pay-as-you-go

ElevenLabs bills API usage in dollars at the same per-1,000-character rate on every plan, and each paid plan’s fee is prepaid usage at that rate: Starter at $6 includes 75,000 Multilingual characters, Creator at $22 includes 275,000, Pro at $99 includes 1,238,000, Scale at $299 includes 3,738,000 and Business at $990 includes 12,375,000. Flash and v4 Turbo cost half as much per character, so each plan covers twice as many characters on those models. For pure API spend, the cheapest choice is whichever plan’s allowance you will actually use up, and the calculator shows every plan so you can see where each one stops being wasted money. Choose a plan for its non-API features (seats, voice cloning, commercial licence) rather than for a discount it does not give.

Calling a text-to-speech API from code

Both SDKs below read the API key from an environment variable (ELEVENLABS_API_KEY and OPENAI_API_KEY). The ElevenLabs voice ID is the “George” voice used in the ElevenLabs quickstart; swap in any voice ID from your voice library.

ElevenLabs, Python (pip install elevenlabs). convert returns an iterator of audio byte chunks:

import os
from elevenlabs.client import ElevenLabs
client = ElevenLabs(api_key=os.getenv("ELEVENLABS_API_KEY"))
audio = client.text_to_speech.convert(
text="Your order has shipped.",
voice_id="JBFqnCBsd6RMkjVDRZzb",
model_id="eleven_multilingual_v2",
output_format="mp3_44100_128",
)
with open("speech.mp3", "wb") as f:
for chunk in audio:
f.write(chunk)

ElevenLabs, TypeScript (npm install @elevenlabs/elevenlabs-js). textToSpeech.convert returns a web ReadableStream:

import { createWriteStream } from "node:fs";
import { Readable } from "node:stream";
import { pipeline } from "node:stream/promises";
import { ElevenLabsClient } from "@elevenlabs/elevenlabs-js";
const client = new ElevenLabsClient(); // reads ELEVENLABS_API_KEY
const audio = await client.textToSpeech.convert("JBFqnCBsd6RMkjVDRZzb", {
text: "Your order has shipped.",
modelId: "eleven_multilingual_v2",
outputFormat: "mp3_44100_128",
});
await pipeline(Readable.fromWeb(audio), createWriteStream("speech.mp3"));

OpenAI, Python (pip install openai):

from openai import OpenAI
client = OpenAI()
with client.audio.speech.with_streaming_response.create(
model="tts-1",
voice="coral",
input="Your order has shipped.",
) as response:
response.stream_to_file("speech.mp3")

OpenAI, TypeScript (npm install openai):

import fs from "node:fs";
import OpenAI from "openai";
const openai = new OpenAI();
const mp3 = await openai.audio.speech.create({
model: "tts-1",
voice: "coral",
input: "Your order has shipped.",
});
await fs.promises.writeFile("speech.mp3", Buffer.from(await mp3.arrayBuffer()));

Model IDs map to the price rows above: eleven_multilingual_v2, eleven_v3 and eleven_v4 bill at $0.08 per 1,000 characters, eleven_flash_v2_5 and eleven_v4_turbo at $0.04. On OpenAI, tts-1 is $15 per million characters and tts-1-hd is $30; gpt-4o-mini-tts accepts an instructions field for tone but bills per audio token. OpenAI’s usage policy requires telling listeners the voice is AI-generated.

What the calculator does not include

• Promotions with an end date, one-off signup credits (Deepgram $200, AssemblyAI $50, AWS Free Tier credits) and 12-month trial allowances. Always-free monthly allowances are subtracted only when the checkbox is on.
• Volume discounts, committed-use contracts, Azure commitment tiers and enterprise pricing.
• Add-ons billed on top of transcription, such as speaker diarization, PII redaction or key-term prompting.
• Tax, and the separate charges for storing audio in a cloud bucket before transcribing it.

Disclosure: the “Try ElevenLabs” links are referral links, and InventiveHQ may earn a commission if you subscribe. The ranking is by price only and every provider is priced the same way.