Neural text-to-speech. No cloud required.
6 languages, 37 voices, one Gradle line. On-device neural synthesis — your text never leaves the phone. No API key, no per-character fee, no internet. Start free, go Pro for streaming and extensibility.
// on-device AI · arm64-v8a · 24 kHz WAV · 82M params · 100% offline
Your text stays on the phone.
Cloud TTS APIs send your text to a server and charge per character. JokobeeTTS runs the neural model locally — nothing uploaded, nothing billed, works offline. Private by design, unlimited usage, zero recurring cost.
6 languages, production-grade normalization.
Each language has its own normalizer — currencies, dates, times, abbreviations, phone numbers, cardinal points — all spoken naturally. 400+ English loanwords automatically pronounced correctly in every romance language.
Everything you need to ship a TTS app.
The free tier is a complete, production-ready TTS library — not a crippled demo. 6 languages, 37 voices, smart normalization, loanwords engine, custom pronunciation. No limits, no watermarks.
Neural Synthesis
Kokoro-82M ONNX engine. Natural 24 kHz audio. 37 voices across 6 languages. Runs entirely on-device.
Smart Normalization
Currencies, dates, times, abbreviations, cardinal points, phone numbers — spoken naturally per locale.
Loanwords Engine
400+ English words automatically pronounced correctly in French, Spanish, Italian, Portuguese. "Package" = /ˈpækɪdʒ/, not /pakaʒ/.
Custom Pronunciation
IPA, ARPABET, and TSV dictionaries. One line to make your app say any word correctly.
Audio Stitching
Multi-sentence synthesis with configurable silence, crossfade, and peak normalization.
Model Manager
Download models and voices on first launch. Keep your APK light — heavy assets load from Cloudflare.
An extensible TTS platform for professional apps.
Pro unlocks the extensibility layer — streaming, accent adapters, terminology dictionaries, normalization adapters, voice import and blending, auto language detection. Open formats for your own data, encrypted formats for proprietary content.
Streaming
Phrase-by-phrase streaming with callback or Kotlin Flow. Audio starts before full synthesis completes. Essential for chatbots and voice assistants.
Auto Language Detection
Pass lang="auto" — the library detects the language. Per-segment in streaming: French then English uses the right pipeline for each.
Voice Import & Blending
Import custom .pt voice files. Blend 2-3 voices with weighted mixing to create unique vocal identities.
Accent Adapters
Transform phonemes for regional accents — Québécois French, Castilian Spanish, European Portuguese. JSON or encrypted .jkba.
Terminology Dictionaries
Medical, legal, technical dictionaries in JSON, TSV, IPA or ARPABET. Stack multiple with priority. Open or encrypted .jkbt.
Normalization Adapters
Customize text processing for regional formats, abbreviations, expressions. JSON rules or encrypted .jkbn.
4 extension points. One extensible pipeline.
Every word flows through a complete processing chain with 4 named extension points. The developer controls pronunciation at every level — from raw text to final phonemes.
Ship for free. Extend with Pro.
The free tier is a complete TTS library — not a trial. Pro unlocks the extensibility platform: streaming, adapters, dictionaries, voice import, auto-detection. One-time purchase, yours forever.
- 6 languages · 37 voices · neural synthesis
- Production-grade normalization per locale
- 402 loanwords (automatic, all romance languages)
- Custom pronunciation (IPA / ARPABET / TSV)
- Audio stitching (silence, crossfade, normalize)
- Model Manager (download on first launch)
- 24 kHz WAV · arm64-v8a · 100% offline
- Everything in Free
- Phrase-by-phrase streaming (callback + Kotlin Flow)
- Auto language detection (lang="auto", per-segment)
- Voice import + blending (custom .pt files)
- Accent adapters (JSON + encrypted .jkba)
- Normalization adapters (JSON + encrypted .jkbn)
- Terminology dictionaries (JSON/TSV + encrypted .jkbt)
- No subscription — no recurring fees
No NDK. No Python. Just add the dependency.
The library includes everything — neural model, G2P engines, normalizers, voice files. Add one line to your build file and synthesize.
Transcribe → Process → Speak. All on-device.
Pair with Whisper Android for speech-to-text and llama.cpp for LLM chat — a complete voice assistant pipeline running entirely on the phone. No cloud, no API, no limits.