English · 简体中文 · 日本語

Side Story builds on work, ideas, bug reports, and testing shared by many people. Thank you to the original Alexandria community and to everyone helping improve the forks.

• Finrandojin: the original Alexandria Audiobook application and the downloadable Alexandria voice presets.

• Ayush Naphade: persona generation, speaker alias resolution, and contextual script review (upstream PR #42). His related project is Lily.

• Michii / on22s: the Alexandria Audiobook2 improvements carried into Side Story, including the upstream System Health Dashboard, cross-platform subprocess runner, and Voice Training Dataset Preparer (#45, #46, #47).

• REFLEXGAMING007: detailed feature proposals, reproduction cases, and repeated testing that shaped the forks' manual LLM workflow, snapshots, incremental personas, editable prompts, and character-state voices (manual generation, snapshots, incremental updates, prompt editing, state personas).

• Alchemist-Production: the original Web UI, Audio Editor, and chunk-management work integrated into upstream Alexandria (integration, editor).

• PierrunoYT: the original HTML-escaping security work incorporated by on22s in upstream PR #39.

• Robertoge65: the external-TTS missing-import fix (upstream PR #19); kishore-1404 for the CUDA VRAM-property correction (upstream PR #35).

• MNeMoNiCuZ: the Docker persona-prompt packaging fix, incorporated through upstream's later Docker update (original contribution and credit).

Community reports, ideas, and testing behind inherited fixes

• ElBiggus: early Windows/audio feedback and download and CPU-loading reports.
• 1BadAshB: batch-rendering and cancellation testing, and the stale clone-prompt cache diagnosis and fix verification.
• generallogin: model-cache loading and NVIDIA slowdown reports and testing.
• ryuhda: Korean LoRA testing that helped identify and validate the training/inference prefill fix.
• sdbrown67: the malformed LLM-output report behind safer entry filtering.
• WirelessRoth: German-language and prompt feedback that helped improve language controls and editable review prompts.
• damajor: Docker startup, configuration, and VRAM-cleanup reports.
• futuregadets: Drive-first storage and ngrok-free Colab suggestions adopted upstream.
• Tortoise17: the external voice-clone language report.
• starpause: the image-based EPUB chapter-heading discussion that prompted TOC-based heading recovery (AA2 implementation).

• cjdell: the AMD APU precision fix that informed Side Story's fp32 model-loading path (port).
• XinchaoGou: OpenAI reasoning-model request handling and environment-referenced API keys, plus ideas for reasoning controls, clone-reference validation, external TTS server pools, and rolling story context (ports, reasoning controls, reference validation, server pools).
• Darkkingwill: MP3 bitrate findings and ideas for cancellable merging, export progress, model selection, and remembered tabs, independently rebuilt in AA2 and carried into Side Story (bitrate, interface improvements).
• shangyuok: the idea of configurable TTS generation-token limits, implemented together with matching memory estimates (implementation).
• dudarenok-maker / Castwright: ideas for remembering a cast across books and checking rendered voice consistency, implemented independently in Alexandria (series memory, voice checks).
• Gaspard Michel, Elena V. Epure, Romain Hennequin, and Christophe Cerisara: the quotation-attribution research behind the michel prompt family, adapted into Alexandria's own prompts and surrounding-context handling (implementation).

Research influences and shared resources

Side Story also benefits from the research behind AA2. These credits include experiments and evaluation resources; they do not mean every experiment, dataset, or model is included in the Side Story runtime.

• Xiao Zhang / buddies: per-line instruction vocabulary and experiments separating stable voice identity from delivery instructions. The lexicon was ported into AA2's research tools; the resulting comparisons informed the voice-identity work (audit, voice changes).
• p0n1: punctuation priorities adapted for AA2's quotation-aware chunking experiment.
• zeropointnine: transcribe-back validation and truncation-detection ideas explored in AA2.
• Drew Thomasson / VoxNovel: the BookNLP baseline that prompted attribution comparisons; David Bamman and the BookNLP contributors for the underlying literary NLP work.
• nazdridoy: weighted voice-blending ideas explored in AA2; WhiskeyCoder for independent Qwen3-TTS chunk-length findings used as a comparison.
• Krishnapriya Vishnubhotla, Adam Hammond, and Graeme Hirst and the PDNC annotators; Sean Papay and Sebastian Padó for RiQuA; and the DraCor community: annotated literary resources used in AA2's attribution research.
• David K. Elson and Kathleen R. McKeown: quotation-attribution categories explored in AA2's syntactic-frame experiments; Frank Rudzicz and the BookNLP-OG coauthors for end-to-end attribution comparisons.
• The CosyVoice team: speech-model design ideas studied in AA2; CosyVoice is not a Side Story runtime dependency.
• LibriVox readers, Project Gutenberg and Aozora Bunko volunteers, and the creators of LJSpeech, Hi-Fi TTS, AISHELL-3, and the Kokoro Speech Dataset: human readings and source texts used for research comparisons.
• Jay Myers, Malk Williams, and the casting guidance recorded in Third-Party Notices: practical ideas about consistent character voices and contrast between characters who share scenes.

Further research sources, what was adapted, and resource-specific rights caveats are recorded in THIRD_PARTY_NOTICES.md.

Thanks to the maintainers and contributors behind these projects:

• The Alibaba Qwen team for Qwen3-TTS, which powers local speech generation, voice cloning, and voice design.
• PyTorch, Hugging Face Transformers, and PEFT for the model runtime and adapter training; SpeechBrain for speaker embeddings used by voice checks.
• FastAPI and the Python audio ecosystem, including pydub, SoundFile, librosa, and FFmpeg, for serving the app and working with audio; Gradio for the external TTS client.
• Bootstrap, Font Awesome, Marked, and DOMPurify for the bundled browser interface assets. Their pinned sources and license files are retained under app/static/vendor.
• Pinokio for the installation and launcher platform.
• For the optional Voice Tools: the llama.cpp, llama-cpp-python, OpenAI Whisper, whisper.cpp, Silero VAD, and pyannote.audio teams. These support local LLM serving, transcription, speech detection, and optional speaker diarization; their installation and model requirements are separate from the core workflow.

Credits do not imply endorsement. This section complements the license, third-party notices, and individual model, dataset, and dependency terms; it does not replace them.

Home · Installation · First-Audiobook

Credits checked against Side Story main 8c7a33c, its retained notices, and the linked upstream and AA2 contribution history.