Text-to-speech APIs
A text-to-speech API takes text and streams back audio in a voice you pick or clone. Two things decide most choices: how fast the first audio arrives, which matters for a voice agent on a call, and how the read sounds over a long script, which matters for narration and video. ElevenLabs, Cartesia and Rime sell voice first. Deepgram sells it next to transcription, OpenAI next to its models. Google Cloud, Amazon Polly and Azure bill it like any other cloud service, with free characters every month. DeepInfra hosts the open Kokoro model for under a dollar per million characters. 9 tools, in the order they appear in the directory. This is not a ranking.
Grouped by job: AI Agent Infrastructure
Deepgram
Speech-to-text API: Nova-3 transcribes recorded files and live streams in 60+ languages, Flux adds turn detection for English voice agents, with diarization, keyterm prompting and redaction, billed per second. Aura-2 text-to-speech in 7 languages on the same key, and self-hosted containers on Enterprise.
ElevenLabs Text to Speech
ElevenLabs' text-to-speech API: Eleven v4 for expressive speech in 90+ languages with audio tags, v4 Turbo at about 100 ms for agents, and Flash for the lowest latency. Use any voice in your library by ID, clones included. Billed per 1,000 characters.
OpenAI Text to Speech
OpenAI's text-to-speech endpoint: gpt-4o-mini-tts takes prompts for tone, accent and pace, with 13 built-in voices, streaming and custom voices. tts-1, tts-1-hd and gpt-4o-mini-tts shut down on January 6, 2027; the replacement is gpt-realtime-2.1-mini on the Realtime API.
Cartesia
Text-to-speech API built for voice agents: Sonic 3.6 speaks 44 languages and reads confirmation codes without preprocessing, with instant voice cloning. Speech-to-text and voice agents from the same company.
Rime
Text-to-speech API for phone and voice agents: Mist v3 for the lowest latency, Coda v2 for the most natural read in 10 languages, streaming over HTTP and WebSockets, billed by the minute. English pronunciation control, speech-to-text too, and self-hosting on Enterprise.
Google Cloud Text-to-Speech
Google Cloud's text-to-speech API: Chirp 3 HD voices, Gemini-TTS models, instant custom voices, and older Neural2, WaveNet and Standard voices, billed per character with a monthly free allowance on most voice types.
Amazon Polly
AWS's text-to-speech service: Generative, Long-Form, Neural and Standard voices billed per character, with Speech Marks for lip sync and word highlighting, and free replay of audio you cache.
Azure Speech
Microsoft's speech service in Azure AI Foundry: neural and Neural HD text-to-speech voices, custom professional voices and long-form audio, next to speech-to-text and translation, with containers on commitment tiers.
DeepInfra
OpenAI-compatible API for open models such as Llama, DeepSeek and Qwen, billed per token with Standard, Priority and Flex tiers. Also open text-to-speech models such as Kokoro, billed per character, and dedicated GPUs for your own LLMs billed by the minute.
What it is, where it runs, what a million characters costs.
Straight from each vendor's own pages. Plans move often, so the check date is below.
| Tool | What it is | Runs on | Open source | Price |
|---|---|---|---|---|
| Deepgram | Aura-2 voices in English, Spanish, German, French, Dutch, Italian and Japanese, plus Flux TTS, on the same key as Deepgram's speech-to-text and voice agent API. | Deepgram's cloud, with an EU endpoint. Self-hosted containers on Enterprise. | No. | $200 free credit, no card. Per 1,000 characters: Aura-2 $0.030, Aura-1 $0.015, Flux TTS $0.045. |
| ElevenLabs Text to Speech | Eleven v4 for expressive speech in 90+ languages with audio tags, v4 Turbo at about 100 ms for agents, v3 for multi-speaker dialogue, Flash for the lowest latency. Any voice in your library by ID, clones included. | ElevenLabs' cloud. | No. | Per 1,000 characters: v4, v3 and v2 Multilingual $0.08; v4 Turbo, v3 Conversational and Flash $0.04. Free plan includes 10,000 characters a month; plans from $6/mo. |
| OpenAI Text to Speech | gpt-4o-mini-tts with prompts for tone, accent and pace, 13 built-in voices, streaming and custom voices. tts-1 and tts-1-hd are the older models. All three shut down January 6, 2027. | OpenAI's cloud. The replacement, gpt-realtime-2.1-mini, runs on the Realtime API, not the speech endpoint. | No. | gpt-4o-mini-tts: $0.60 per 1M text tokens in, $12 per 1M audio tokens out. tts-1 $15 and tts-1-hd $30 per 1M characters. |
| Cartesia | Sonic 3.6 in 44 languages, reading confirmation codes without preprocessing, with instant voice cloning from Pro. Speech-to-text and voice agents from the same company. | Cartesia's cloud. On-prem or your VPC through sales. | No. | 1 credit per character. Free: 20K credits a month, no commercial license. Pro $5/mo; Startup $49/mo for 1.25M; Scale $299/mo for 8M. Overage $65, $45 or $38 per 1M by tier. |
| Rime | Mist v3 for the lowest latency in 4 languages, Coda v2 for the most natural read in 10. HTTP and WebSocket streaming, English pronunciation control. Built for phone agents. | Rime's cloud. Cloud, VPC or on-prem on Enterprise. | No. | Billed per minute, about 1,000 characters each. Mist v3 $0.03, Coda v2 $0.05. Free $50 of usage a month, no card. Scale $100/mo takes 20% off. |
| Google Cloud Text-to-Speech | Chirp 3 HD voices, Gemini-TTS models, instant custom voices, plus older Neural2, WaveNet and Standard voices. | Google Cloud. | No. | Per 1M characters: Chirp 3 HD $30 after 1M free a month; Neural2 $16; WaveNet and Standard $4 after 4M free. Instant custom voice $60. Gemini-TTS billed per token. |
| Amazon Polly | Generative, Long-Form, Neural and Standard voices, with Speech Marks for lip sync and word highlighting. Replaying audio you cache costs nothing. | AWS, in your AWS account and regions. | No. | Per 1M characters: Standard $4, Neural $16, Generative $30, Long-Form $100. Free tier: 5M Standard characters a month; Neural, Generative and Long-Form free amounts for the first 12 months. |
| Azure Speech | Neural and Neural HD voices, custom professional voices and long-form audio, next to speech-to-text and translation in one Speech resource. | Azure regions. Containers on commitment tiers. | No. | Free tier: 0.5M characters a month. Per 1M characters (US East): Neural $15, Neural HD $22, custom professional voice $24, long audio $100. |
| DeepInfra | Hosted Kokoro-82M, a small open text-to-speech model with 54 voices in 8 languages, over DeepInfra's API. | DeepInfra's cloud. The model runs anywhere you can run Python. | Kokoro weights and code are Apache-2.0. DeepInfra is not. | Kokoro-82M $0.99 per 1M characters. |
Checked 2026-11-01. Sources: ElevenLabs API pricing, OpenAI pricing, OpenAI text-to-speech guide, Cartesia pricing, Cartesia Sonic 3.6 docs, Deepgram pricing, Deepgram voices and languages, Rime pricing, Google Cloud Text-to-Speech pricing, Amazon Polly pricing, Azure Speech pricing, Azure retail prices API, Kokoro-82M on DeepInfra, Kokoro on Hugging Face, Kokoro on GitHub.
Pick it if.
- ElevenLabs Text to SpeechYou want the widest choice of voices and languages, or to clone a voice and use it everywhere.
- OpenAI Text to SpeechYou need a voice on the OpenAI key you already have, and can move to the Realtime API before January 2027.
- CartesiaYou are building a voice agent and want low latency in many languages, with a $5 plan to ship on.
- DeepgramYou already transcribe with Deepgram and want the voice on the same key and bill.
- RimeYour agent is on the phone, and you want per-minute pricing and English pronunciation control.
- Google Cloud Text-to-SpeechYour stack runs on Google Cloud and a million free Chirp 3 HD characters a month covers your start.
- Amazon PollyYou are on AWS and want cheap Standard or Neural voices, with Speech Marks for highlighting or lip sync.
- Azure SpeechYou are on Azure, or need custom voices or containers on your own hardware.
- DeepInfraPrice per character matters most, and an open model you could later run yourself is enough.
Each listing.
The directory page, alternatives in the same category, and a head-to-head where we have one.
- DeepgramAlternatives
- ElevenLabs Text to SpeechAlternatives
- OpenAI Text to SpeechAlternatives
- CartesiaAlternatives
- RimeAlternatives
- Google Cloud Text-to-SpeechAlternatives
- Amazon PollyAlternatives
- Azure SpeechAlternatives
- DeepInfraAlternatives
Category pages list every tool in that category, including ones left off this page. Built one of these? Get it featured in its category.
The 2 biggest AI stories, every morning.
One short email: the two stories that matter, and why.
You're in.
The Daily AI Brief will land in your inbox soon. If you don't see it, check Promotions or Spam and move it to your main inbox.