KittenTTS

KittenTTS is an open-source speech toolkit that can run models locally or through a hosted API. It is best for developers and teams adding voices, cloning, or transcription to products, especially when self-hosting matters. The combination is capable, but production users must account for separate weight licensing and usage-based API charges.

Overview

KittenTTS 2 is a CPU-capable 1.7B-parameter model, accompanied by lighter legacy models. Local use requires Linux, Windows, or macOS with Python 3.9 or later, so this is a development component rather than a ready-made consumer reader. Downloadable weights also make deployment in a VPC, on-premises environment, or device possible.

Zero-shot cloning works from a short audio sample without fine-tuning. The project describes 47 built-in voices across 20 languages, although the hosted Free plan narrows that to 36 voices and 10 languages. Emotion tags, inline vocal-event tags, and emphasis spans provide useful delivery controls; emotion remains beta and can vary by voice and sentence.

Key features

The hosted service exposes speech generation, voice cloning, and transcription endpoints through OpenAI-compatible HTTP requests, with streaming support. That lowers integration friction for teams already using compatible clients, while transcription adds an ASR path alongside TTS. On-device SDKs are offered for Python, Swift, React Native, Flutter, and Web; Kitten-TTS 2 SDKs are still forthcoming, so the newest model may require a different integration route.

Long scripts need sentence-aware chunking. A single whole-script pass can truncate or drift into repetition, making automated chunking an operational requirement for audiobooks, long-form narration, and other extended jobs.

The repository code uses Apache 2.0, but model weights have separate terms: KittenTTS 2 is under the Stellon Labs Community License. Review that license before commercial redistribution or deployment. The privacy policy says service inputs, including audio or voice recordings, and outputs are collected, with technical, physical, and organizational safeguards; data may be transferred to jurisdictions with different protection levels.

Pricing

KittenTTS is free and open source for local use, with no free trial because there is no paid local plan. Hosted costs depend on the following tiers.

• Free — free: 30 free minutes of speech and transcription, then $15 per 1M TTS characters and $1 per ASR audio hour. It includes 36 voices, 10 languages, five simultaneous ASR sessions, and two concurrent TTS streams. This suits experiments and small workloads, but the caps and post-quota rates limit sustained production use.
• Pro — $14.99 a month (billed $179.88 billed annually): $24.99 of monthly usage is included, with TTS at $0.65 per 1M characters and ASR at $0.10 per audio hour. It can be cancelled anytime and is the sensible tier for regular API traffic; annual billing is required for the stated $179.88 charge.
• Enterprise — custom pricing: custom usage and rates, API capacity, private deployment, invoicing, dedicated support, and onboarding. Choose it when governance, capacity, or deployment control outweighs the need for a published price.

Platforms

KittenTTS spans Android, iOS, Linux, macOS, Windows, web, API, and self-hosted deployments. Commercial use is supported, and WAV is the stated export format. Self-hosting is particularly relevant to organizations that cannot send voice data to a shared service.

Who it's for

It fits developers building voice agents, customer-support call automation, content-creation pipelines, and dedicated speech endpoints. Teams wanting local inference, OpenAI-compatible hosted calls, or zero-shot cloning have clear reasons to consider it. It is a poor fit for nontechnical users seeking a polished reading application, and long-form creators must accommodate chunking.

Pros and cons

• Pros: CPU-capable local model; zero-shot cloning without fine-tuning; broad stated voice and language coverage; hosted generation, cloning, and transcription; self-hosting in VPC, on-premises, or on-device environments; commercial integration assistance and enterprise support.
• Cons: KittenTTS 2 requires Python 3.9 or later; emotion control is beta and inconsistent; whole-script processing can truncate or repeat; model-weight licensing is separate from the Apache 2.0 code; hosted Free quotas and concurrency limits are restrictive.

Alternatives

Pick iSpeech instead if you want a freemium service with a free trial, broad platform coverage, and a low-cost Plus plan at $2.95 a month.

Vocalware is the better choice for paid, stream-count plans, starting at $99.50 a month for 40,000 streams.

Choose Voice RSS for a simpler freemium API with 350 daily requests on its free tier or 1,000 on Advanced at $5 a month.

Resemble AI suits teams wanting pay-as-you-go credits, one free team seat, and self-hosting options.

TTSReader is an alternative freemium text-to-speech application.

Adobe Firefly is a freemium Adobe platform with limited daily free generations and paid creative-generation tiers.

Pocket TTS is another free, open-source, CPU-based option with a Python API and CLI.

Voice.ai is a freemium service combining voice changing, text to speech, and audio tools with a 5,000-credit monthly Free plan.

Browse more options in Text-to-Speech Software.

Verdict

Choose KittenTTS if you are building speech features and value local or private deployment, voice cloning, and an API that also handles transcription. Look elsewhere if you need a turnkey consumer app, predictable long-script rendering without chunking, or model terms and hosted limits that require less scrutiny.