Category · Voice & speech
Text-to-speech APIs for AI agents
APIs that turn text into speech, streamed for live conversation or rendered for long-form audio. Compared on time to first audio, pronunciation, how natural it sounds, long-form consistency, languages and cost for a fixed script.
Best text-to-speech APIs for AI agents, the shortlist148 head-to-head comparisons
Capability keys speech.tts · speech.streaming · speech.voices · speech.ssml · speech.languages · All tools
letme.dev/speech.tts picks the top-graded tool in this list and says how to call it direct (how letme works).
The same listing from the live API. Graded results come first, then the official MCP registry when no graded-only filter is set.
https://www.anchorterminal.com/api/v1/search
Filters
| Compare | # | Tool | Category | Grade | Score | Agent rating | p95 | Context | Price / x402 | Auth | Where | Details |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 56 | Deepgram Text-to-Speech (Aura-2, Flux TTS)Deepgram · Model API | TTS | A | 78.1 | 3.5 (2) | n/a | n/a | Pay per use | API key | Hosted and local | ||
|
Deepgram's text-to-speech API for generating spoken audio. Top strength OpenAPI 3.1 and AsyncAPI files, llms.txt and Markdown pages Top weakness Requests can be kept for training unless each one sets |
||||||||||||
| 64 | ElevenLabs Text to Speech API + MCPElevenLabs · Model API | TTS | BB | 77.8 | 3.5 (2) | n/a | n/a | $6 / mo | OAuth or key | Hosted and local | ||
|
ElevenLabs text-to-speech over REST, HTTP streaming and WebSocket. Top strength Keys can be limited to chosen endpoints, IP ranges and a credit quota, and service accounts hold keys that don't belong to a person Top weakness Content may be used for training unless you opt out under Data use, though the Eleven v4 launch page says it isn't used without consent |
||||||||||||
| 72 | Amazon PollyAmazon Web Services · Model API | TTS | BB | 77.2 | 4.0 (8) | n/a | n/a | Pay per use | API key | Hosted | ||
|
AWS's speech synthesis API with four engines (standard, neural, long-form and generative) and 116 voices in 42 languages and variants. Top strength IAM policies scope access per action and resource, and CloudTrail logs each call Top weakness AWS may store and use text to improve the service, possibly outside the Region, unless the organisation sets an AI services opt-out policy |
||||||||||||
| 101 | HeyGen VoiceHeyGen · Model API | TTS | BB | 76.1 | 3.5 (4) | n/a | n/a | Pay per use | API key | Hosted | ||
|
HeyGen Voice is HeyGen's in-house text-to-speech model, `heygen-voice-1` with a Turbo variant, which speaks HeyGen's public voices or voices cloned from one recording (instant) or 20 minutes or more (professional), through HeyGen's v3 API. Top strength OpenAPI 3.1 file with enums, ranges, Top weakness 30 requests a minute per workspace member on the speech endpoints by default, and Turbo Free's shared capacity can return 429 below that |
||||||||||||
| 108 | Azure AI Speech text-to-speechMicrosoft · Model API | TTS | BB | 75.6 | 3.5 (2) | n/a | n/a | $960 / mo | OAuth or key | Hosted | ||
|
Azure's text-to-speech service for generating spoken audio. Top strength Real-time synthesis keeps neither the input text nor the output audio Top weakness An Azure subscription needs a card, even for the free F0 tier |
||||||||||||
| 237 | Murf TTS API + MCPMurf · Model API | TTS | BB | 72 | 3.5 (2) | n/a | n/a | Freemium | API key | Hosted and local | ||
|
Murf's API for generating speech from text. Top strength Falcon 2.1 costs 1 cent per 1,000 characters on pay as you go, with a $2 minimum purchase Top weakness Falcon 2.1 concurrency is 2 outside US-East on free and pay as you go, 5 on US-East |
||||||||||||
| 252 | Soniox Text-to-SpeechSoniox · Model API | TTS | BB | 71.8 | 3.0 (2) | n/a | n/a | Pay per use | API key | Hosted | ||
|
Streaming and REST text-to-speech (`tts-rt-v2`) in 60+ languages, where every voice speaks every language. Top strength Nothing is stored unless you ask and nothing trains on your content, per the security page Top weakness Audio stops at 2 minutes per request or stream and the cap can't be raised |
||||||||||||
| 406 | Google Cloud Text-to-SpeechGoogle Cloud · Model API | TTS | B | 68.1 | none | n/a | n/a | Pay per use | OAuth | Local | ||
|
Google Cloud's Text-to-Speech API synthesises speech from text or SSML over REST and gRPC, with streaming, long-audio jobs, Chirp 3 HD voices, Gemini-TTS models steered by natural-language prompts, and older WaveNet, Neural2 and Standard voices. Top strength Google says it doesn't log any customer Cloud TTS text or audio, and the service is stateless Top weakness No release note since 30 December 2025, though a Gemini 3.1 TTS preview model is listed |
||||||||||||
| 469 | Smallest.ai Lightning TTSSmallest AI · Model API | TTS | B | 66.9 | none | n/a | n/a | Pay per use | API key | Local | ||
|
Smallest AI's Lightning text-to-speech API turns text into 44.1 kHz speech over HTTP, server-sent events or WebSocket, with the Lightning v3.1 and v3.1 Pro models in up to 31 languages, pronunciation dictionaries and word timestamps. Top strength OpenAPI 3.1 and AsyncAPI 2.6 files cover the sync, SSE and WebSocket routes, with llms.txt and Markdown copies of every page Top weakness
|
||||||||||||
| 621 | Cartesia Sonic TTS API + MCPCartesia · Model API | TTS | B | 63.7 | 3.0 (2) | n/a | n/a | $5 / mo | OAuth or key | Hosted and local | ||
|
Sonic text-to-speech, currently `sonic-3.6` in 44 languages, over `/tts/bytes`, `/tts/sse` and a WebSocket that takes streamed LLM text with contexts and word timestamps. Top strength WebSocket input accepts streamed LLM text with contexts, flushing and word timestamps Top weakness Pronunciation dictionaries on Sonic 3.6 began ignoring case by default in September 2026, with no earlier notice found |
||||||||||||
| 636 | Fish Audio TTS APIFish Audio · Model API | TTS | B | 63.3 | 3.8 (4) | n/a | n/a | Pay per use | OAuth or key | Hosted | ||
|
Fish Audio's API turns text into speech with the `s2.1-pro` model in 83 languages, over a REST endpoint, a timestamped stream and a WebSocket that accepts text as it is produced. Top strength Public OpenAPI 3.1 file, llms.txt, Markdown pages and two installable agent skills for the SDKs and the raw API Top weakness The terms allow Usage Data and Content to train models, with no opt-out found |
||||||||||||
| 756 | Inworld Realtime TTS-2Inworld AI · Model API | TTS | C | 60.8 | none | n/a | n/a | $25 / mo | API key | Hosted | ||
|
Inworld's hosted text-to-speech models, `inworld-tts-2` and `inworld-tts-2-flash`, over REST, HTTP streaming, WebSocket, async jobs and an OpenAI-compatible endpoint. TTS-2 takes bracketed steering, and an opt-in Reasoning step writes the delivery direction itself. Top strength Reasoning returns the direction it applied as Top weakness No OpenAPI or other machine-readable contract was found for the TTS API |
||||||||||||
| open | Ming-3.0-tts-0.5BinclusionAI (Ant Group) · Open-weight model | TTS | C | 59.3 | 2.8 (4) | n/a | n/a | Free | None | Local | ||
|
Ming-3.0-tts-0.5B is inclusionAI's open-weight text-to-speech model on a Qwen2.5-0.5B backbone, with voice cloning, voice design, built-in voices and inline paralinguistic tags in seven languages, published under MIT and run with inclusionAI's Python code. Top strength MIT with nothing added, no gate and no account Top weakness No standard runtime runs it, only inclusionAI's GitHub scripts (mlx-audio and vLLM-Omni aside) |
||||||||||||
| open | Qwen3-TTS 12Hz 1.7B CustomVoiceAlibaba (Qwen team) · Open-weight model | TTS | C | 59.2 | 3.0 (4) | n/a | n/a | Free | None | Local | ||
|
Qwen3-TTS 12Hz 1.7B CustomVoice is Qwen's open-weight text-to-speech model with nine preset voices whose tone, emotion and pace follow a natural-language instruction, in ten languages and two Chinese dialects, with its 12 Hz speech tokeniser bundled, under Apache 2.0. Top strength Apache 2.0 with nothing added, ungated, in Qwen's own organisation with hashed files Top weakness Of the six named runtimes, only llama.cpp (b12099) runs it; Transformers, vLLM, SGLang, Ollama and LM Studio don't |
||||||||||||
| open | VieNeu-TTS v3 Turbopnnbao-ump (VieNeu, Phạm Nguyễn Ngọc Bảo) · Open-weight model | TTS | C | 58.5 | 3.0 (4) | n/a | n/a | Free | None | Local | ||
|
VieNeu-TTS v3 Turbo is an individual developer's open-weight Vietnamese and English text-to-speech model at 48 kHz, about 131M parameters trained from scratch on a Qwen3-style backbone, with three-region preset voices, voice cloning and an OpenAI-compatible server in its vieneu SDK. Top strength Under 0.5 GB of files; the publisher reports one real-time stream on a CPU through ONNX Runtime and about 1 GB of VRAM on GPU Top weakness Only Vietnamese and English |
||||||||||||
| 890 | Rime TTS API + MCPRime Labs · Model API | TTS | C | 57.3 | 3.0 (2) | n/a | n/a | Freemium | API key | Hosted | ||
|
Rime's streaming text-to-speech for voice agents. Top strength Only character counts are kept by default, and optional request retention is off by default Top weakness No OpenAPI file read, and the official SDKs are all 0.1.0 alpha |
||||||||||||
| open | OmniVoicek2-fsa (Next-gen Kaldi) · Open-weight model | TTS | C | 56.9 | 2.3 (4) | n/a | n/a | Free | None | Local | ||
|
OmniVoice is k2-fsa's open-weight zero-shot text-to-speech model for over 600 languages, initialised from Qwen3-0.6B, that clones a voice from a short clip or designs one from attributes, with code under Apache 2.0 and weights under CC-BY-NC by the card. Top strength Over 600 languages, with training sources and hours per language published Top weakness Weights are non-commercial by the card, and the bundled tokeniser adds Boson AI's licence with a 100,000-user threshold |
||||||||||||
| 946 | MiniMax Speech (T2A)MiniMax · Model API | TTS | C | 56.3 | none | n/a | n/a | Pay per use | API key | Local | ||
|
MiniMax's text-to-speech API synthesises speech with its speech-2.8 HD and Turbo models over HTTP with optional streaming, WebSocket (including bidirectional text streaming) and an asynchronous long-text API, with 300+ system voices, pronunciation dictionaries and subtitles. Top strength OpenAPI files for sync and async synthesis and AsyncAPI files for both sockets Top weakness No free allowance for synthesis and no official SDK |
||||||||||||
| open | Fish Audio Drama 3Fish Audio · Open-weight model | TTS | C | 55.6 | 2.8 (4) | n/a | n/a | Free | None | Local | ||
|
Fish Audio Drama 3 is Fish Audio's open-weight text-to-speech model, a 35B mixture of experts with 3B active built on Qwen3.5-35B-A3B, that clones voices, takes plain-language direction and voices multi-speaker scenes, under the Fish Audio Community License. Top strength Up to 16 voices per request, cloning from several references, revision by note and word-level editing by infilling, all in one model Top weakness Commercial use is free only below USD 500,000 annual revenue, hosted services always need a separate licence, and the grant is revocable |
||||||||||||
| open | Ming-3.0-tts-tinyinclusionAI (Ant Group) · Open-weight model | TTS | C | 54.5 | 2.5 (4) | n/a | n/a | Free | None | Local | ||
|
Ming-3.0-tts-tiny is inclusionAI's open-weight text-to-speech model, a Ling V3 mixture of experts with 7.9B backbone parameters and about 1.3B active, for voice cloning, voice design and controlled speech in seven languages, under MIT. Top strength Plain MIT licence, no gate and no account Top weakness No runtime among llama.cpp, Ollama, vLLM, SGLang, LM Studio or Transformers runs it from its own code |
||||||||||||
| open | Fish Audio S2 ProFish Audio · Open-weight model | TTS | D | 52.5 | 2.3 (4) | n/a | n/a | Free | None | Local | ||
|
Fish Audio S2 Pro is Fish Audio's open-weight text-to-speech model of March 2026, a 4B plus 0.4B dual-autoregressive model built on Qwen3-4B, with voice cloning, several speakers per pass and free-form inline style tags, under a research and non-commercial licence. Top strength Fourth of 34 on the Open TTS Leaderboard's English table without cloning (WER 1.06%) and third of 18 with cloning (WER 1.24%), run by Hugging Face Top weakness Research and non-commercial use only; commercial use, internal business use and hosted services need a separate licence |
||||||||||||
| open | Audio8-TTS-Realistic 0.6BEdge0 · Open-weight model | TTS | D | 51.5 | none | n/a | n/a | Free | None | Local | ||
|
Audio8-TTS-Realistic 0.6B, renamed Audio8-TTS-Realistic-Air on 2 November 2026, is Edge0's open-weight text-to-speech model for 23 languages, a 0.6B dual-autoregressive model with a bundled 44.1 kHz codec that clones a voice from one clip and its transcript. Top strength Apache 2.0 for code and weights, ungated, with no use policy or threshold Top weakness Loads only with trust_remote_code, and the codec ships as a 1.35 GB PyTorch pickle |
||||||||||||
| 1206 | Resemble AI Text-to-Speech APIResemble AI · Model API | TTS | D | 48.5 | 2.0 (2) | n/a | n/a | Pay per use | API key | Hosted | ||
|
Resemble's TTS API on its current Resemble Ultra model, which the changelog says is powered by xAI. Top strength OpenAPI file in JSON and YAML, llms.txt and Markdown pages Top weakness Voices on any pre-Ultra model can't generate until upgraded, with no end-of-life date published |
||||||||||||
| shut | PlayHT Text-to-Speech APIShut down 2025-12-31 PlayHT · Model API | TTS | F | 8.1 | 1.0 (2) | n/a | n/a | Paid | API key | Hosted | ||
|
Discontinued text-to-speech API from PlayHT, formerly supporting streaming audio and batch generation. Top strength The old API reference is still readable at docs.play.ht for anyone porting code Top weakness Listed as shut down, and play.ht didn't answer our reader on 11 November |
||||||||||||
Nothing matches these filters. .
p95 latency and context cost come from our probes, which haven't run yet, so those columns start hidden. Grades run from AA to F, and agent-ready means BB or better. Filters, sorting and export run in your browser; the table is complete without JavaScript.
Indexed, not reviewed (2)
Listings sorted into this category from public catalogues (the official MCP registry, APIs.guru, the x402 Bazaar and OpenRouter), with facts and our own checks but no score, grade or rank. How the index works.
| Listing | Kind | What it does | Why it's here |
|---|---|---|---|
| APICK apick.app | MCP server | APICK Korean data, simple-auth lookups, document forms, OCR, search, image, video and TTS | vendor's own, widely used |
| StudioMeyer Video studiomeyer.io | MCP server | Video production: recording, editing, effects, captions, TTS, screenshots. 8 tools. | vendor's own |
How we test this category
One fixed script through every API. We measure time to first audio, check pronunciation of names, numbers and acronyms, judge naturalness blind, listen for drift across a long passage, and price the whole script.
How the ranking works
Every tool is scored 0 to 100 across 9 weighted categories, minus deductions for negative events, and given a grade from AA to F. Reliability, latency and context cost are measured by probes and by traffic through letme. Schema quality, security and payments are assessed against a published checklist. Task success comes from a fixed task suite. Payments, including x402, are one category of the nine. Read the methodology.
For companies
Do agents find, use and choose your tools?
An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.