Cartesia turns text into speech through an API, with a free plan (20K credits/month, no card required) that's enough to try it properly before paying. Pick it when you need a voice for an app or a short piece of audio; pick a human recording when a small number of messages matter enough to get exactly right.
Other good fits
Comparing hosted text-to-speech quality on your own script before committing to a paid tier.
+ Pros
Read pros
Cartesia's Sonic models convert written text into spoken audio through an API, with stock voices, instant voice cloning on paid plans, and a range of output formats. It's the speaking half of a voice product: it doesn't listen, recognise speech, or manage a conversation on its own.
Pick it when…
You want a spoken version of short, fixed application messages (an announcement, a confirmation, a read-aloud summary).
You want to compare hosted voices on your own script before deciding whether text-to-speech is good enough for the job.
You may later add streaming, where text arrives incrementally and speech starts before the whole sentence is written.
− Cons
Read cons
A language being listed as supported doesn't mean every name, number or abbreviation in your script will be pronounced the way you expect; listen to the actual output before shipping it. Hosted synthesis needs a network connection, and moving to a realtime, streaming setup adds buffering and playback work beyond a simple one-off file.
Pick something else when…
A small, fixed set of important messages need to be exactly right. A human recording still beats synthesis for a handful of critical lines.
You need offline generation with no network call. A local runtime such as F5-TTS keeps audio generation on your own machine.
You want speech as one node in a larger automation, not a standalone API call. Wire it into n8n instead of calling the API directly from your own code.
Avoid if
You're building a full conversational voice agent; Cartesia's text-to-speech is one piece of that, not the whole thing.
Connection
Call client.tts.generate with a model ID (for example sonic-3.6), a voice ID and a language, and choose an output format (container, encoding, sample rate). The API key authenticates the request; responses can be written straight to a file or streamed.
Text-to-speech and speech-to-text, no card required
Pro
$5/month
100,000
Adds commercial use license and instant voice cloning
Startup
$49/month
1,250,000
Adds professional voice cloning, organisations
Scale
$299/month
8,000,000
Priority support, high concurrency
Enterprise
Custom
Custom
Custom concurrency, compliance agreements, SSO
Standard text-to-speech is metered roughly per character; a managed voice agent (a full conversational product) is billed separately, per minute, so don't estimate agent costs from the text-to-speech rate. Workspace seats are unlimited on every plan. The running effort is watching credit use and deciding which voice and model version to pin for anything that needs consistent pronunciation.
Data handling
Text sent for synthesis is processed on Cartesia's hosted service; the resulting audio file, once downloaded, lives wherever you save it. Voice cloning (instant or professional, on paid plans) creates a reusable voice model tied to your account; check the current terms before cloning anyone's voice without clear consent.
First request
Documentation-based suggestion — not run here.
Generate a short spoken announcement for a fictional event using a stock voice: "Welcome to the neighbourhood seed exchange. Please place labelled packets on the table and leave the walkway clear." Save it as a WAV file and listen for missing words, odd pauses or mispronunciations, not just whether a file was produced.
Done when you've listened to the clip against the transcript and can say whether it's good enough to ship as-is, or needs another take.
Links & evidence
Documentation-based · not hands-on · Sep 23, 2026
Rests on Cartesia's pricing page, its Sonic model documentation and its generate-to-file example, read on 2026-09-23. No audio was generated for this review.
Stream synthesized speech for live voice assistants and interactive tutors.
Choose it when an application must start speaking with little delay.
Sonic-2 supports more than fifteen languages for developer workflows.