# 6 best text to speech APIs for conversational AI in 2026, compared

By Samuel Oram-Jones

A conversational AI that never laughs, hesitates, or drops its voice sounds like an audiobook narrator reading a support script, and the language model behind it can't fix that because the problem sits in delivery. Each of these six APIs lets a developer shape the laugh, the whisper, or the sigh through tags, markup, or a prompt, and they differ on how, on price, and on whether a cloned voice is on offer at all.

Bland Speech comes first because Bland Speech v3 ranked first on realism in Design Arena's blind test of 4 August 2026 and documents its performance tags in full. Three rivals charge $15 per million characters against Bland Speech's $40, and for some teams that settles it.

## What are the best text to speech APIs for conversational AI?

The best text to speech API for conversational AI is Bland Speech, followed by Grok TTS, Microsoft Azure, Google, ElevenLabs, and Inworld, as of November 2026. Bland Speech v3 ranked first of 13 AI voices in Design Arena's Audio Realism Bench results of 4 August 2026, and it takes performance tags such as [laughs], [sighs], and [say warmly] inline.

- **Bland Speech:** Best overall for conversational AI that should sound like a person
- **Grok TTS:** Best for inline speech tags at $15 per million characters
- **Microsoft Azure:** Best for MAI-Voice models and SSML speaking styles on Azure
- **Google:** Best for Gemini-TTS style prompts and tags on Google Cloud
- **ElevenLabs:** Best for audio tags and 90+ languages on Eleven v4
- **Inworld:** Best for plain-language steering and 50 non-verbal sounds

| API | Delivery controls | Price per million characters | Voice cloning | Audio Realism Bench, 4 August 2026 |
|---|---|---|---|---|
| Bland Speech | Performance tags, expressiveness and stability dials, timed pauses | $40 | Instant from about 10 seconds | Ranked 1st of 13, 1365 |
| Grok TTS | Inline and wrapping speech tags | $15 | Instant from a clip up to 120 seconds | Ranked 3rd, 1156 |
| Microsoft Azure | SSML speaking styles, MAI-Voice | $15 for Neural | Personal voice, limited access | MAI-Voice-2 ranked 2nd, 1224 |
| Google | Gemini-TTS tags and style prompts | $30 for Chirp 3: HD, Gemini by token | Chirp 3 instant custom voice closed to new requests | Gemini 2.5 Pro TTS ranked 5th, 1109 |
| ElevenLabs | Audio tags on Eleven v4 | $40 to $80, by model | Instant from 10 seconds on v4 | Eleven v3 ranked 12th, 938 |
| Inworld | Plain-language steering, 50 non-verbal sounds | $15 to $25, pay as you go | Instant from 3 seconds | TTS-1.5 Max ranked 13th, 763 |

<details><summary>What these rows mean</summary>

- **Delivery controls:** What each company says shapes how a line is spoken. Inworld's steering works on TTS-2 only, and Bland's tags work on its current-generation voices.
- **Voice cloning:** The fastest clone each company offers from a short sample, with access limits noted where they apply.
- **Audio Realism Bench, 4 August 2026:** Design Arena's blind listening test of AI voices, with each score's place among 13 AI voices. Azure's MAI-Voice-2.1 models and Inworld's Realtime TTS-2 weren't in those results.

</details>

Bland Speech is the top pick because it combines the realism result with a documented tag set and cloning from a 10-second sample, and Grok TTS is the pick for a team that cares more about a $15 rate.

## How should a team compare text to speech for conversational AI?

The six APIs here were compared on delivery controls, voice cloning, price per character, and how human each voice sounded in Design Arena's blind test. Samuel Oram-Jones took each rival fact from the company's own pricing and documentation pages.

### Can the voice laugh, whisper, and hesitate on cue?

A performance tag is a short instruction in square brackets that sits in the text right before the words it shapes. Bland publishes tags for emotion, such as [say warmly] and [say nervously], and for vocalizations, such as [laughs], [sighs], and [clears throat], and says the model was trained on that exact set. Grok TTS takes tags such as [laugh] and [sigh] plus wrapping tags such as whisper and loud, Google's Gemini-TTS takes bracketed tags in four groups, ElevenLabs takes audio tags such as [whispering] on Eleven v4, and Inworld renders 50 non-verbal sounds. Azure drives delivery with SSML speaking styles, a markup standard where a style degree runs from 0.01 to 2.

### Does a prompt steer the delivery instead of a tag?

Some APIs take a plain-language description of the delivery. Inworld's TTS-2 performs a described tone sent as an instruction, Google's Gemini-TTS uses a style prompt as the main driver of emotional tone, and Azure's HD voices infer emotion from the text. OpenAI's gpt-4o-mini-tts takes an instructions prompt too, but it shuts down on 6 January 2027 with the rest of OpenAI's speech endpoint models.

### Can the voice be cloned, and under what limits?

Cloning access varies more than anything else here. Google's Chirp 3 instant custom voice is closed to new access requests, and Azure's personal voice is a limited access feature. Grok TTS clones from a reference clip in the console, with the API gated to Enterprise, and ElevenLabs, Inworld, and Bland Speech clone from a short sample on standard accounts.

## How do the 6 best text to speech APIs for conversational AI compare?

These six differ most in how they shape delivery and who can clone a voice.

### 1. Bland Speech: Best overall for conversational AI that should sound like a person

**What does it do?** [Bland Speech](/speech) is the text to speech model inside Bland, which runs its own model and speech. Tags such as [laughs], [say warmly], and [voice lowered] go inline, expressiveness sets how much intonation varies, stability sets how much the same input varies between renders, and a pause is written as <|1.5|> for 1.5 seconds. The studio has a director feature, Norm, that writes and performs a single-speaker scene.

**Who is it for?** Bland Speech suits a team building an assistant, companion, or support voice that should sound like a person from the first sentence.

<div class="cmp-proscons cmp-proscons-own"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>Bland Speech v3 ranked first of 13 AI voices in Audio Realism Bench results of 4 August 2026</li><li>A published set of emotion, intensity, and vocalization tags</li><li>Instant clones from about 10 seconds, private to the customer's organization</li><li>One standard rate, $0.04 per 1,000 characters, with no monthly plan</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>More than a weekend prototype needs, because Bland Speech is built for production calls</li></ul></div></div>

**Pricing:** $0.04 per 1,000 characters, which is $40 per million, pay as you go, with 50,000 characters free once for a new account.

### 2. Grok TTS: Best for inline speech tags at $15 per million characters

**What does it do?** Grok TTS is the text to speech API from xAI, now named SpaceXSI, and it reached general availability in March 2026. It has 28 built-in voices, tags such as [laugh], [chuckle], and [sigh], wrapping tags such as whisper and sing-song, and a WebSocket that takes text as deltas.

**Who is it for?** Grok TTS suits a team that wants expressive tags at a low per-character rate.

<div class="cmp-proscons" data-rival="grok-tts"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>$15 per million characters, with no plan tiers</li><li>Pronunciation replacements by respelling or IPA</li><li>Grok TTS ranked 3rd of 13 AI voices in Audio Realism Bench results of 4 August 2026</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>Cloning through the API is gated to Enterprise</li><li>Custom voices are available in the United States only, excluding Illinois</li></ul></div></div>

**Pricing:** $15 per million characters, pay as you go, with rate limits by spend tier and custom voice creation in the console free for up to 30 voices.

### 3. Microsoft Azure: Best for MAI-Voice models and SSML speaking styles on Azure

**What does it do?** Azure Speech offers neural voices, HD voices that infer emotion from text, and MAI-Voice-2.1 and MAI-Voice-2.1-Flash, which are in public preview without an SLA. A speaking style such as whispering or ecstatic is set through SSML.

**Who is it for?** Azure suits a team already on Microsoft's cloud that wants speaking styles and a Microsoft-built model in one account.

<div class="cmp-proscons" data-rival="azure"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>Microsoft's MAI-Voice-2 ranked 2nd of 13 AI voices in Audio Realism Bench results of 4 August 2026</li><li>Neural voices at $15 per million characters</li><li>A HIPAA BAA through Microsoft's Product Terms</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>MAI-Voice is in public preview with no SLA</li><li>Custom and personal voices are limited access, by registration for approved use cases</li></ul></div></div>

**Pricing:** $15 per million characters for Neural and Neural HD Flash voices, with 0.5 million characters free a month on the free tier. MAI-Voice-2.1 wasn't in the 4 August results.

### 4. Google: Best for Gemini-TTS style prompts and tags on Google Cloud

**What does it do?** Google Cloud Text-to-Speech sells Chirp 3: HD voices and Gemini-TTS models that take bracketed tags, such as [sigh], [whispering], and [short pause], plus a style prompt. Chirp 3: HD has no emotion control, only speaking rate and pause tags.

**Who is it for?** Google suits a team on Google Cloud that wants prompt-driven delivery and per-token billing.

<div class="cmp-proscons" data-rival="google"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>Gemini 2.5 Pro TTS ranked 5th of 13 AI voices in Audio Realism Bench results of 4 August 2026</li><li>Single-speaker and multi-speaker dialogue on Gemini 3.1 Flash TTS</li><li>The first million Chirp 3: HD characters free each month</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>Vocalized-adjective tags can be spoken aloud, a side effect Google warns about</li><li>Gemini-TTS has no free usage</li></ul></div></div>

**Pricing:** Gemini 2.5 Pro TTS costs $1.00 per million text tokens in and $20.00 per million audio tokens out, and audio runs 25 tokens a second, so a minute of speech is about $0.03 in audio tokens. Chirp 3: HD costs $30 per million characters.

### 5. ElevenLabs: Best for audio tags and 90+ languages on Eleven v4

**What does it do?** ElevenLabs' Eleven v4, launched on 28 September 2026, takes audio tags such as [whispering], [shouting], and [laughing] and speaks 90+ languages. Eleven v3 and v4 don't take SSML break tags, so pauses come from punctuation and tags.

**Who is it for?** ElevenLabs suits a team that wants expressive tags across many languages and a plan-based way to clone voices.

<div class="cmp-proscons" data-rival="elevenlabs"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>Audio tags on Eleven v4</li><li>90+ languages on the Eleven v4 family</li><li>Instant clones from 10 seconds, with voice captcha verification for professional clones</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>Eleven v3 ranked 12th of 13 AI voices in Audio Realism Bench results of 4 August 2026</li><li>Eleven v4 costs $0.08 per 1,000 characters</li></ul></div></div>

**Pricing:** $0.08 per 1,000 characters for Eleven v4, v3, and Multilingual v2, and $0.04 for v4 Turbo, v3 Conversational, and Flash. Eleven v4 came out after the 4 August results.

### 6. Inworld: Best for plain-language steering and 50 non-verbal sounds

**What does it do?** Inworld's Realtime TTS-2 performs a delivery described in plain words, and both TTS-2 and TTS-2 Flash render 50 non-verbal sounds such as [laugh] and [sigh]. Inworld aims at consumer-facing applications and names companions, education, and games among its use cases.

**Who is it for?** Inworld suits a team building a companion or consumer assistant that wants steering and low rates.

<div class="cmp-proscons" data-rival="inworld"><div class="cmp-proscons-pros"><p><strong>Pros</strong></p><ul><li>TTS-2 Flash at $15 per million characters</li><li>Instant clones from as little as 3 seconds</li><li>Word and character timestamps, plus visemes for lip-sync</li></ul></div><div class="cmp-proscons-cons"><p><strong>Cons</strong></p><ul><li>TTS-1.5 Max ranked 13th of 13 AI voices in Audio Realism Bench results of 4 August 2026</li><li>Steering works on TTS-2 only</li></ul></div></div>

**Pricing:** On-Demand is $0 a month with TTS-2 at $25 per million characters and Flash at $15, and paid plans lower both. Realtime TTS-2 wasn't in the 4 August results.

## How much does text to speech for conversational AI cost?

Prices run from $15 per million characters at Grok TTS, Azure, and Inworld's Flash to $40 at Bland Speech and $80 for Eleven v4, with Google's Gemini models billing by audio token instead. Bland Speech is not the low-cost option on this list, and a team weighing it against Grok TTS is weighing a $25 per million difference against the realism result.

Enterprise pricing is custom, set with Bland's sales team.

## Which text to speech API sounds most human in conversational AI?

Bland Speech v3 was picked as the most human of these voices in Design Arena's Audio Realism Bench results of 4 August 2026, ranking first of 13 AI voices at 1365, ahead of Microsoft's MAI-Voice-2 at 1224, xAI's Grok TTS at 1156, Google's Gemini 2.5 Pro TTS at 1109, ElevenLabs' Eleven v3 at 938, and Inworld's TTS-1.5 Max at 763, which ranked 2nd, 3rd, 5th, 12th, and 13th. Real human recordings scored 1500.

OpenAI's GPT Realtime 2, a speech-to-speech model on its Realtime API, ranked 4th at 1119. Design Arena's test covers explainers and conversations as well as phone agents, and it leaves out acting-heavy speech such as game characters and companion voices, a limit for a team building those. [Design Arena's published results](https://notes.designarena.ai/audio-realism-benchmark-measuring-realism-in-audio-models/) carry the full method. Bland's [comparison of text to speech APIs](/compare/text-to-speech), its [best lists](/compare/text-to-speech/best-lists), and [best text to speech with performance tags](/blog/best-tts-performance-tags) cover the same companies by other jobs.

## Frequently asked questions

### What is the best text to speech API for conversational AI?

Bland Speech is the best text to speech API for conversational AI in 2026. Bland Speech v3 ranked first of 13 AI voices in Design Arena's Audio Realism Bench results of 4 August 2026, and it takes inline tags for emotion and vocalizations.

### Which text to speech APIs support laughs and sighs?

Bland Speech takes [laughs] and [sighs], Grok TTS takes [laugh] and [sigh], Google's Gemini-TTS takes [sigh] and [laughing], ElevenLabs' Eleven v4 takes [laughing], and Inworld renders [laugh] and [sigh] among 50 non-verbal sounds.

### Can Bland Speech express emotion?

Yes. Bland publishes emotion tags such as [say warmly], [say nervously], and [say playfully], intensity tags such as [very loud] and [voice lowered], and vocalizations such as [gasps] and [clears throat]. They work on Bland's current-generation voices only.

### Can Bland Speech clone a voice for a conversational agent?

Yes. An instant clone needs one clean sample of about 10 seconds, and a professional clone is fine-tuned on 30 minutes or more of audio and goes through Bland's consent and rights review. Every clone is private to the customer's organization, and the customer confirms the voice is theirs to use.

### Is OpenAI's text to speech API good for conversational AI?

OpenAI's gpt-4o-mini-tts takes an instructions prompt for tone, but it, tts-1, and tts-1-hd shut down on 6 January 2027. OpenAI points customers to gpt-realtime-2.1-mini on its Realtime API, and a team staying on a speech endpoint can move to Bland Speech by changing a base URL, a key, and a model name.

### What does a minute of conversational speech cost?

At $0.04 per 1,000 characters and about 750 to 1,000 characters a minute, Bland Speech costs about $0.03 to $0.04 a minute. Gemini 2.5 Pro TTS costs about $0.03 a minute in audio tokens before text input, and Grok TTS at $15 per million characters costs about $0.01 to $0.015.
