Text to speech API pricing is deliberately confusing: characters versus words versus minutes, standard versus neural versus premium tiers, free allowances with asterisks, and overage rates buried in docs. Choosing wrong means either overpaying for years or hitting a surprise bill after launch.
This guide compares text to speech API pricing across the major providers with a framework for honest comparison — so you can find the cheapest fit for your actual volume, not the marketing headline.
Key Takeaways
- APIs bill in three units: characters, words, or audio minutes — normalize before comparing.
- Voice quality tiers (standard/neural/premium) change prices 2-4x; compare at the tier you will actually use.
- Free tiers cover experimentation; production always needs the paid math.
- Hidden costs: storage, bandwidth, engineering time, and support tiers add 20-50% to sticker prices.
- Model 12-month cost at projected volume including growth — the cheapest starter is rarely the cheapest at scale.
How TTS APIs Charge: The Three Models
Every provider uses one of these:
1. Per character (most common). Google Cloud TTS, Azure AI Speech, Amazon Polly bill per character processed, with neural/premium tiers costing more. English averages ~5.5 characters per word including spaces — so 1 million characters ≈ 180,000 words ≈ 20 narration hours.
2. Per word or credit. Some voice-AI APIs use credits where 1 credit ≈ 1 character, or bill per word. Credits obscure the math — always convert to characters.
3. Per audio minute. A few providers bill per minute of generated audio. Simple to estimate (you know your audio length), but punishes slow/pause-heavy content.
The normalization rule: convert every quote to cost per 1 million characters of neural-quality output. That single number makes providers comparable. Anything else is marketing fog.
Watch the tier trap: providers advertise the standard-tier price prominently while everyone uses neural. Always price the tier you will ship.
Provider Pricing Compared
Directional comparison (verify current pricing — it changes):
| Provider | Billing Unit | Free Tier | Price Posture | Best Value When |
|---|---|---|---|---|
| Google Cloud TTS | Per character | Monthly allowance | Mid; neural premium | Already on GCP |
| Azure AI Speech | Per character | Monthly allowance | Mid; neural premium | Already on Azure / commits |
| Amazon Polly | Per character | 12-month intro tier | Mid; neural premium | Already on AWS |
| ElevenLabs | Credits (~chars) | Small free tier | Premium | Quality > cost |
| Smaller TTS APIs | Per character | Generous | Low-mid | Budget-first builds |
| Open-source (self-host) | Compute only | Free software | Infra cost | High volume, engineering team |
Cloud giants cluster similarly on price — choose on ecosystem, not pennies. Voice-AI specialists charge premiums justified (or not) by voice quality. Budget APIs undercut 30-60% with acceptable quality for utility audio.
Self-hosting (Coqui, Piper): software free, but GPU costs, maintenance, and voice quality gaps make it viable only at high volume with engineering capacity. Most teams underestimate the ops burden.
The Hidden Costs Nobody Quotes
Sticker price versus real price:
- Storage and bandwidth. Generated audio has to live somewhere and reach users — S3 + CDN costs scale with success.
- Engineering time. Integration, SSML tuning, pronunciation lexicons, monitoring, and incident response. A "cheap" API with poor docs costs senior-engineer weeks.
- Regeneration waste. Every retry, test, and revision bills characters. Budget 20-30% overhead above net output.
- Support tiers. Production SLAs often require paid support plans — factor them in.
- Lock-in exit costs. Migrating voice IDs, lexicons, and tuned SSML later costs real project time; price the second provider choice too.
- Success tax. Metered billing means viral growth spikes your bill — model a 10x traffic month before committing.
Add 20-50% to any provider's headline quote for the honest number. Then compare.
Choosing: A Decision Worksheet
Fill this in for each finalist:
1. Monthly volume: characters (net) × 1.25 (waste factor) = billable.
2. Quality tier needed: standard / neural / premium — price only this tier.
3. Provider quote at that volume: $_/month.
4. Hidden costs: storage $ + engineering $ + support $ = $.
5. 12-month total at current volume: $_.
6. 12-month total at 3x volume: $_ (kills "cheap now, expensive later" traps).
7. Exit difficulty: low / medium / high (custom voices and lexicons raise this).
Decision rules: if quality needs are modest, the cheapest adequate API wins — do not pay neural premiums for IVR prompts. If voice is the product (content, brand), pay for quality; the API is a fraction of content economics. And re-run this worksheet annually — pricing moves fast in this market.
Frequently Asked Questions
How is text to speech API pricing calculated?
Most APIs bill per character processed, with neural/premium voice tiers costing 2-4x standard. Some use credits or per-minute billing. One million characters is roughly 180,000 words or 20 hours of narration — use that to normalize providers.
What is the cheapest TTS API?
It depends on volume and quality tier. Budget TTS APIs undercut cloud giants 30-60% for utility audio; self-hosted open-source costs only infrastructure at high volume. For neural-quality output at moderate volumes, compare normalized per-million-character costs — the ranking changes with volume.
Are TTS API free tiers enough for production?
Rarely. Free tiers cover experimentation and tiny workloads — typically a few hundred thousand characters monthly at most, sometimes time-limited. Any real production workload needs paid math; model your actual monthly characters including 20-30% regeneration waste.
Why did my TTS bill spike unexpectedly?
Common causes: regeneration loops retrying failed requests, test environments hitting production endpoints, a viral content spike (metered billing taxes success), or tier confusion (neural billed instead of standard). Set billing alerts at 50%, 100%, and 200% of expected spend.
Should I choose per-character or per-minute API pricing?
Per-character suits variable content (you pay for input); per-minute suits predictable audio output (you pay for output). For pause-heavy or slow content, per-character is usually cheaper; for dense fast narration, compare carefully. Normalize both to cost per finished hour.
Related Guides
- Text to Speech API Guide for Developers
- AI Voice Credits Explained
- ElevenLabs Alternative: Compare Top AI Voice Tools
Ready to put this into practice? Try SynVoice free — generate studio-quality voiceovers in minutes, with a generous free tier and no credit card required.