Choosing a voice API is harder than it looks: every provider mixes a different set of services and quotes prices in different units, and the comparisons tend to be written by the providers themselves. This one too —we are one of them, Uttera— so we wrote it with one rule: where someone is better or cheaper than us, we say so. It is the only way it is useful to you.
Prices as of 9 October 2026, shown in euros (≈ €0.92/$) so they compare like-for-like. They change often: always check each provider's own page before deciding.
Before price comes which services each API offers, because almost nobody offers all of them:
| Service | Uttera | ElevenLabs | OpenAI | Deepgram | AssemblyAI | Gladia | |
|---|---|---|---|---|---|---|---|
| Standard voice (TTS) | ✅ | ✅ | ✅ | ✅ | — | ✅ | — |
| Voice cloning | ✅ | ✅ | — | — | — | ✅¹ | — |
| Transcription (STT) | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
| Diarization (who speaks) | ✅ | ✅ | — | ✅ | ✅ | ✅ | ✅ |
| Translation | ✅ | ✅ | ✅ | — | — | ✅¹ | ✅ |
| Voice analysis (emotion/profile) | ✅ | — | — | — | ✅² | — | — |
| Summarization | ✅ | — | ✅¹ | ✅ | ✅ | ✅¹ | ✅ |
| Entities (NER) | ✅ | ✅ | — | ✅ | ✅ | ✅¹ | ✅ |
| PII redaction | ✅ | — | — | ✅ | ✅ | ✅¹ | ✅ |
| Chapters | ✅ | — | — | — | ✅ | — | ✅ |
| Sound effects | ✅ | ✅ | — | — | — | — | — |
| Music | ✅ | ✅ | — | — | — | — | — |
| Signed PDF report | ✅ | — | — | — | — | — | — |
| Verifiable report (evidence graph) | ✅ | — | — | — | — | — | — |
| Native MCP server | ✅ | ✅ | — | — | — | — | — |
| OpenAI dialect (drop-in) | ✅ | — | ✅ | — | — | — | — |
| Asynchronous delivery + webhooks | ✅ | ✅ | — | ✅ | ✅ | ✅ | ✅ |
| Asterisk integration⁴ | ✅ | — | — | — | — | — | — |
| Telegram bot | ✅ | — | — | — | — | — | — |
| Slack app | ✅ | — | — | — | — | — | — |
| EU data residency | ✅ | — | — | — | — | ✅¹ | ✅ |
| Self-hostable | ✅ | — | — | ✅¹ | — | — | — |
| Does not train on your data (by default) | ✅ | — | ✅ | — | — | ✅ | — |
| Signature / watermark in the audio | ✅ | — | — | — | — | ✅³ | — |
| 100% solar-powered⁵ | ✅ | — | — | — | — | — | — |
1 as a separate product or tier · 2 text-based, not acoustic · 3 SynthID · 4 FreePBX via Asterisk (first-party dialplan) · 5 100% self-powered, measured at the utility meter; check it live. Table as of date; verify with each provider.
What nobody else has is the sovereignty combination: not training on your data and being self-hostable and EU residency, all at once. OpenAI and Google also do not train by default, but you cannot host them yourself; Deepgram can be self-hosted, but it trains by default unless you opt out. Real peace of mind is not a policy you have to trust: it is the audio never leaving your machine. And there is one more thing no one else offers: a signed, verifiable PDF report of the transcription, with an AI-generated summary up front to grasp in seconds what the recording was about, the analysis in plain words, and the full transcript below for the exact detail —and that a third party can check on their own. And the newest piece: that report can carry an evidence graph —each statement anchored to its quote, its speaker and its minute in the audio, with a stamp of whether the quote is literally in the recording—, so it can be checked sentence by sentence, not blindly. It is veracity by anchoring: we do not say whether what was said is true —that is for the listener to judge—, but we do say where each sentence comes from. No one else gives you that.
Where we are not unique, and we say so: asynchronous delivery with webhooks for long recordings is also offered by the batch-STT specialists (AssemblyAI, Deepgram, Gladia, ElevenLabs, Google). We added it to match them on long jobs —with the result encrypted and the webhook signed with your own key— without giving up voice, hearing, intelligence, sound, music and a signed report in the same account. The only one in the table without this deferred mode is OpenAI, whose audio API is synchronous.
The short read: AssemblyAI and Gladia are pure STT (no speech synthesis), ElevenLabs is voice and cloning (its STT is secondary), OpenAI and Google do everything but in separate products, and Uttera brings voice, listening, intelligence, sound, music and a signed report under one account, with the MCP connector and sovereignty as the difference.
| Provider | €/hour |
|---|---|
| Uttera (Pro) | 0.14 |
| AssemblyAI (Universal-2) | 0.14 |
| ElevenLabs (Scribe v2) | 0.20 |
| Deepgram (Nova-3, batch) | 0.24 |
| OpenAI (gpt-4o-transcribe) | 0.33 |
| Gladia (Starter) | 0.56 |
| Google (Chirp) | 0.88 |
We are now among the cheapest on the market for transcription, level with AssemblyAI's cheapest model (their Universal-2; their recommended model, Universal-3.5, costs more, ~€0.19/h). And the real difference shows the moment you need more than text: our transcription ships with acoustic intelligence —emotion, profile, diarization— in the same balance, whereas on AssemblyAI diarization, entities or summarization are billed separately, so for any real production use their cost climbs well above the base price. At the same bare price, we include what they charge as add-ons.
| Provider | €/1M chars |
|---|---|
| Uttera standard (Pro) | 3.2 |
| Google Standard | 3.7 |
| OpenAI tts-1 | 13.8 |
| Deepgram Aura-2 | 27.6 |
| Google Chirp3-HD | 27.6 |
| ElevenLabs | 46 – 92 |
Our standard voice is the cheapest on the market from the Professional plan up, below even the hyperscalers' basic voices. If you need volumes of preset speech, nothing is cheaper.
Cloning is a different league —it competes with ElevenLabs, not with preset voices— so it gets its own table:
| Provider | €/1M chars |
|---|---|
| Uttera cloning (Pro) | 45 |
| ElevenLabs | 46 – 92 |
| OpenAI · Deepgram · AssemblyAI · Gladia | no API voice cloning |
| Google (Custom Voice) | custom, enterprise tier |
Here we are not the cheap option: we are the quality just below ElevenLabs, with high-fidelity voice cloning and, on top, EU residency and no training on your data —things the leader does not give you.
Our pitch is not "cheapest at everything" —we are not, on cloning— but one API that listens, speaks, translates and analyses, with OpenAI's dialect so you migrate without touching code, EU residency, no training on your data, and an MCP server so your agent uses it on its own. If you want to see it, here is what it does and the plans.
Third-party prices are public and as-of-date; ours are what the API actually charges, published
at /v1/plans. Any figure ages: check at the source.
Anything to add or correct? Write to support@uttera.ai. If you correct us, we edit the post and credit you.
Uttera