AI Voices for Your Business
Which text-to-speech engine to use, and what it actually costs. No hype -- just the honest numbers for a phone greeting, an explainer video, or a branded assistant.
The 90-second answer
Most small businesses fall into one of three buckets. Pick the one that matches how much you'll actually use it.
Free and private, on your own hardware
Kokoro or Piper, self-hosted. No per-word bill, no data leaving your building. Best if you want a phone greeting, IVR menu, or simple notifications and don't need a "branded" cloned voice.
Who it's for: anyone comfortable running a small local service (or having us run it for you).
$0 / monthHosted, pay-as-you-go, occasional use
Breeze TTS 2's free/hosted tiers, OpenAI's TTS API, or Google Cloud's free monthly allowance. No server to run; you pay only for what you generate, and most small-business volumes stay inside the free tier.
Who it's for: businesses generating a modest, steady stream of short audio -- greetings, alerts, short explainer clips.
$0–$10 / monthPremium cloned or branded voice, heavy use
ElevenLabs, Breeze Pro, or a hosted API tier -- a distinctive voice cloned from a real person (with consent), dozens of voice slots, and enough monthly volume for regular content production.
Who it's for: a business that wants one recognizable "brand voice" across a lot of audio -- ads, IVR, video narration, a virtual assistant.
$20–$100+ / monthComparison
| Engine | Quality tier | Languages | Voice cloning | License | Cost | Runs where |
|---|---|---|---|---|---|---|
| Kokoro-82M | Good, fast | 9 languages, 54 voices | No | Apache-2.0, free, commercial OK | $0 | Your hardware, or ours |
| Piper | Good, fast | Many languages/voices | No | MIT, free, commercial OK | $0 | Your hardware (CPU is fine), or ours |
| Breeze TTS 2 — hosted (Free) | Very high (#1 open-weight on the Artificial Analysis speech arena) | English, Chinese (weights); more languages in the hosted app | Yes, from ~20 sec of audio | Commercial use OK on hosted plans | $0 (400 credits/day, 5 voice slots) | Vendor cloud (breezeblue.ai) |
| Breeze TTS 2 — hosted (Starter/Creator/Pro) | Very high | Same as above | Yes | Commercial use OK | $9 / $19 / $99 per month | Vendor cloud |
| Breeze TTS 2 — self-hosted weights | Very high | English, Chinese | Yes | Research & Non-Commercial License | Free for non-commercial use; commercial self-hosting needs written permission from RESONIA (no published price) | Your hardware |
| OpenAI TTS API (tts-1) | Good | Multiple | No | Vendor API terms | $15 / 1M characters | OpenAI cloud |
| OpenAI TTS API (tts-1-hd) | Higher fidelity | Multiple | No | Vendor API terms | $30 / 1M characters | OpenAI cloud |
| Google Cloud TTS — Standard / WaveNet | Good | 40+ languages | No | Vendor API terms | $4 / 1M characters (4M free/month) | Google Cloud |
| Google Cloud TTS — Neural2 | High | 40+ languages | No | Vendor API terms | $16 / 1M characters | Google Cloud |
| Google Cloud TTS — Chirp 3: HD / Studio | Highest | Fewer languages | Instant custom voice option | Vendor API terms | $30 / 1M (Chirp 3 HD, 1M free/month), $160 / 1M (Studio) | Google Cloud |
| Azure AI Speech — Neural | Good | 140+ languages | No (see Custom Neural) | Vendor API terms | $15 / 1M characters (500K free/month) | Microsoft Azure |
| Azure AI Speech — Neural HD | High | Fewer languages | No | Vendor API terms | $22 / 1M characters | Microsoft Azure |
| Azure AI Speech — Custom Neural | High, brand-matched | Depends on training data | Yes (professional fine-tune) | Limited-access program | $6 / 1M characters synthesis, plus training/hosting fees | Microsoft Azure |
| Amazon Polly — Standard | Fair | 30+ languages | No | Vendor API terms | $4 / 1M characters (5M free/month) | AWS |
| Amazon Polly — Neural | Good | Fewer languages | No | Vendor API terms | $16 / 1M characters (1M free/month, first 12 months) | AWS |
| Amazon Polly — Generative / Long-Form | High / Highest naturalness | Fewer languages | No | Vendor API terms | $30 / 1M (Generative), $100 / 1M (Long-Form) | AWS |
| ElevenLabs — subscription (Starter/Creator/Pro) | Very high, most natural cloning | 30+ languages | Yes, from a short sample | Vendor API terms | $6 / $22 / $99 per month (see plan detail below) | ElevenLabs cloud |
| ElevenLabs — API pay-as-you-go | Very high | 30+ languages | Yes | Vendor API terms | $50 / 1M characters (Flash/Turbo), $100 / 1M (Multilingual v2/v3) | ElevenLabs cloud |
Prices fetched directly from each vendor's own pricing page or public pricing API on 2026-09-13 unless noted; see the source list at the bottom of this page. Free-tier allowances and prices change -- always confirm on the vendor's site before budgeting a large volume.
A worked example
Say your business publishes: a 2-minute phone greeting, a 5-minute explainer, and 20 short (about 15-second) notifications a month. Here's the math, using 150 words per minute of speech and about 6 characters per word (including spaces) -- both common planning assumptions for TTS billing.
| Option | First-month cost | Ongoing monthly cost |
|---|---|---|
| Kokoro / Piper, self-hosted | $0 | $0 |
| Google Cloud (Standard/WaveNet, 4M free/month) | $0 | $0 |
| Amazon Polly Standard (5M free/month) | $0 | $0 |
| Azure Neural (500K free/month) | $0 | $0 |
| Breeze TTS 2, Free hosted plan (400 credits/day ≈ 12,000/month) | $0 | $0 |
| OpenAI tts-1 ($15/1M chars, no free tier) | $0.16 | $0.07 |
| ElevenLabs Free plan (10,000 credits/month) | Covers the notifications but is 740 characters short of the full first-month total — the Starter plan or a top-up covers it | $0 (4,440 < 10,000) |
| ElevenLabs Starter ($6/month, 30,000 credits) | $6 (included in plan) | $6 (included in plan) |
| ElevenLabs API pay-as-you-go, Flash/Turbo ($50/1M chars) | $0.54 | $0.22 |
The honest takeaway: at this volume, almost every hosted vendor's free tier covers you completely, and the paid options cost pennies. The real decision is quality and whether you want a cloned/branded voice (Option C), not the character bill.
What "non-commercial" actually means
License, in plain English
Apache-2.0 (Kokoro) and MIT (Piper): free to use in your business, no restrictions, no fees, ever. This is the safest choice if you want zero licensing risk.
Breeze TTS 2's open-weight download carries a Research and Non-Commercial License. You can run it for testing, demos, and personal projects, but using it to generate audio for a real business (a phone system, ads, a paid product) is commercial use and requires written authorization from RESONIA, the company behind it -- there's no published self-hosting price, so you'd need to ask them directly. Their hosted plans (Free through Pro) explicitly allow commercial use -- that's the compliant way to use Breeze's voices in a business today.
Vendor API terms (OpenAI, Google, Azure, Amazon, ElevenLabs): these are cloud services you pay to use, and commercial use is allowed under their standard terms -- read the specific usage policy for the voice model if you plan to clone a real person's voice, since consent and disclosure rules differ.
Bottom line: if you download open-weight AI voice model files to run yourself, check whether the license says "non-commercial" before you put it in front of customers. If it does, either stay on the vendor's paid hosted plan (which grants a commercial license) or ask the vendor for a commercial self-hosting agreement.
Hear the difference
A 10-second sample generated on our own self-hosted engine, produced today for this page:
Kokoro-82M (voice: af_heart)
Free, open-source, self-hosted — recorded live for this page, 2026-09-13.
Premium cloned voices
ElevenLabs and Breeze TTS 2 both publish their own voice demos and voice libraries -- worth a listen before you commit to a plan.
What Pro IT Hosting does
We run these engines on your own hardware or ours, set up and tune the voice you pick, wire it into your phone system, website, or app, and keep it current as the vendors update their models and pricing. You tell us the outcome you want -- a phone greeting, an IVR menu, notifications, a branded assistant -- and we handle the engineering and the licensing check.
Want an AI voice set up for your business?
Tell us what you need it for and we'll recommend the option that actually fits your volume -- honestly, even if that means the free one.
Contact Us →Sources (fetched 2026-09-13): breezeblue.ai/pricing · elevenlabs.io/pricing and elevenlabs.io/pricing/api · OpenAI TTS pricing per public pricing aggregators (OpenAI's own pricing page blocks automated fetches; confirm at openai.com/api/pricing) · cloud.google.com/text-to-speech/pricing · Microsoft's public Azure Retail Prices API (see azure.microsoft.com speech-services pricing) · aws.amazon.com/polly/pricing. Kokoro-82M (Apache-2.0) and Piper (MIT) are open-source projects with no vendor pricing page.