Piper vs Kokoro: how on-device AI voices compare for reading books aloud

We ship both in Lectern, so here’s an honest comparison from the trenches of making them narrate entire novels on real devices.

Piper: the sprinter

Piper is a lightweight neural TTS engine designed for efficiency (it grew out of the open-source voice-assistant world). Its models are small, start quickly, and run comfortably on modest hardware.

• Voice quality: clearly better than classic system TTS — natural rhythm, decent intonation. You can still tell it’s synthetic in long sessions, but it’s pleasant.
• Speed & battery: excellent. Generates faster than real-time on most phones with minimal battery impact — fine for hours of listening.
• Languages: its biggest strength. Piper has voices for dozens of languages, which is how Lectern can match a voice to books in Arabic, Hindi, Chinese, Japanese and many European languages.

Kokoro: the actor

Kokoro is a newer, heavier model that prioritizes lifelike delivery. It’s the one that makes people say “wait, that’s running on my phone?”

• Voice quality: remarkably natural — smoother prosody, more human pacing, better handling of long sentences. For fiction, it’s the closest on-device gets to audiobook narration today.
• Speed & battery: heavier. On recent phones it keeps up fine; on older or budget devices Piper is the safer choice.
• Languages: a smaller (but growing) set than Piper.

Which should you pick?

There’s no wrong answer, which is why Lectern doesn’t make you choose upfront:

• Novels in a major language, modern phone → Kokoro for maximum immersion.
• Multilingual library, older device, or all-day listening → Piper.
• Language learning → Piper’s language breadth plus Lectern’s automatic language detection: open a book, get a native-sounding voice in that language, with every sentence highlighted as it’s spoken.

Either way, everything runs on-device: no account, no cloud upload, no per-month cost, and narration that works in airplane mode.

Why on-device beats the cloud for books

A book is 8–12 hours of audio. Cloud TTS at that scale means constant connectivity, real money (subscriptions or per-character API fees), and shipping the full text of everything you read to a third party. On-device narration costs nothing after you own the app, works offline, and keeps your library exactly where it belongs — with you. In Lectern that’s the whole design: one-time €2.49 on Google Play, no subscription, “no data collected” on the Play safety label.

Frequently asked