SayFable: Offline NLP-tagged Library with 68-Language Synthesis and Book Chat
Apple
Silero
SayFable is a personal iOS library app that processes books, articles, podcasts, and videos entirely on-device, offering transcription, synthesis, translation, and semantic search. It builds a columnar table of linguistic features for fast offline analysis, uses Apple's local Foundation Model for scene labeling and chat with citations, and supports 68 languages for synthesis.
SayFable is a personal media library for iOS that analyzes books, articles, podcasts, and videos entirely on the device, without cloud processing. It imports content from files or links, cleans articles from ads, and provides transcription, synthesis, translation, and semantic search. The app builds a columnar table of linguistic features for each book, enabling fast offline operations like scene navigation, importance sliding, playlist creation, and chat with citations. It uses Apple's local Foundation Model for scene labeling and question-answering with references, avoiding hallucinations. Speech synthesis supports 68 languages, with performance on iPhone 15 Pro Max achieving about 14x faster than real-time transcription and 9.7-10.3x faster synthesis. The app is in beta and free during this period, with a single object with full AI analysis remaining free forever after release.
- Abbreviations
- NLP = Natural Language Processing — Обработка естественного языка
- POS = Part-of-Speech — Часть речи
- NER = Named Entity Recognition — Распознавание именованных сущностей
- RAG = Retrieval-Augmented Generation — Генерация с дополнением извлечением
- RSS = Really Simple Syndication — Формат синдикации новостей
- FM = Foundation Model — Фундаментальная модель
- ONNX = Open Neural Network Exchange — Открытый формат обмена нейросетями
Source: Habr — хаб ML —
original