## Castwright v1.10.0 — three more languages, series memory you can see, and a sharper ear for quality

Quick recap for anyone new: Castwright turns a book into a full-cast audiobook — every character in its own voice, kept consistent across a series — rendered locally on your own machine. Source-available (FSL), solo project, in beta. Here’s what landed in v1.10.0.

Three new performed languages — Spanish, French & German. Five now (English, Russian, Spanish, French, German), and it detects the manuscript's language the moment you import it. The analyser reads each one, the cast and their descriptions come out in that language, and Qwen3-TTS designs and performs voices that speak it. A cast never crosses languages within a book.

Series memory, made visible. The voices you’ve designed, returning book after book, are now shown: a "your cast" chip, a consistency sparkline for a series, a reveal panel with the carried roster, and a shareable cast card you can export. On a real 8-book library, the carried cast came out at 47 characters, with no false voice changes.

A new ear for quality — acoustic voice-drift detection (opt-in). On top of the dead-air / duration / transcript checks, an opt-in speaker-fingerprint gate (ECAPA-TDNN) catches a line where a character's voice drifts out of character even when the words are right — drift the other checks can't hear.

The Quality (1.7B) tier. The everyday Kokoro/Qwen voices stay quick; when you want a book at its best, a larger Qwen3-TTS model is one click away — per character, or pinned across a whole cast. On my RTX 4070 Laptop it renders end-to-end (QA gates and re-records included) at a touch over real time.

Small human sounds, automatically. Gasps, sighs and laughter are now written into a line's delivery, detected by default, in the book's language.

Numbers, dates and currency spoken the way you'd read them aloud — "$1,200" → "one thousand two hundred dollars" — in every language.

Also:

• an LLM Script Review that proposes (and now reassigns) who-said-what, reaching across a chapter boundary to settle a tagless opening line;
• multi-GPU, so you can place each voice engine on the card you choose; and a manuscript-analysis pill that shows every analysis pass as it runs and holds Generate until it's done.

Where it’s still honest: solo project, beta; voice cloning ("your own voice") is still in development; no OCR for scanned/image-only PDFs yet; and I’ve only really tested the non-English languages in short pieces myself — if you read fiction in Spanish, French or German, I’d love to know where the analysis or QA slips.

Hear a sample (one narrator vs full cast): https://www.castwright.ai/#demo · Full release notes: https://github.com/dudarenok-maker/Castwright/wiki/Release-Notes-v1.10.0 · It's at castwright.ai. Feedback or a hand: hello@castwright.ai.