Answers are grounded in this paper and cite their sources.
https://github.com/Borealiscodes/NonDual-Audio-Preprint-Studio/tree/main https://github.com/Borealiscodes/piper_tts Disclaimer on Full-Text Transposition & Analytical Baseline Scope: Please note that the audio files included in this deposit represent a four-turn analytical distillation rather than a complete word-for-word narration of the underlying preprint (DOI: 10.5281/zenodo.24393275) or its companion educational materials (such as the Bill Nye plain-language explainer tiles). This analytical baseline was built specifically to establish, stress-test, and verify phonetic locks, zero-flattery guardrails, and unhurried dual-voice cadence before transposing the full body of written preprints and educational texts into spoken-word audio—or pairing them with future human-narrated recordings. This preprint and accompanying empirical audio dataset chronicle six iterative engineering runs across NonDual-Audio-Preprint-Studio (github.com) and the piper_tts research fork (github.com) while building an accessible, non-jarring spoken-word companion to "Beyond Bolted-On Screen Readers: Narrative Phenomenology, the RNIB Roundhouse Sensory Dark, and Perceptual-Cognitive Unity (Rasa-Darśan)" (DOI: 10.5281/zenodo.24393275). Drawing on RNIB and London Roundhouse sensory-darkness storytelling collaborations, Michael Oliver's Social Model of Disability, and research-projection experiments pioneered by Flyxion (@standardgalactic), the software enforces deterministic Unicode phonetic locks ("Flyxion" -> "FLIK-shun", "rasa-darśan" -> "RAH-sah DAR-shahn", "Tōrō" -> "TOH-roh", "Bonshō" -> "BON-shoh"), zero-flattery script linting with mandatory explicit limitation quotas, and six-register two-speaker alternation over a <= 680 Hz low-pass filtered river/wind stillness bed. Included Empirical Audio Artifacts: 1. run4-cli-flag-literalism.wav — Demonstrates text-to-speech CLI flag literalism when unmatched wrapper flags cause Piper 1.8 to speak its own invocation arguments aloud over the <= 680 Hz stillness bed. 2. run5-single-voice-compressed.wav (0:55) — Demonstrates accurate phonetic rendering via plain stdin, yet reveals single-speaker collapse and breathless pacing on a single American VITS model (en_US-lessac-medium). 3. run6-dual-gb-embodied-success.wav (1:14) — Demonstrates verified dual-voice conversational synthesis pairing en_GB-alan-medium (Speaker A) and en_GB-alba-medium (Speaker B) at 1.30x length scale, 0.55s sentence silence, and 2,000 ms inter-speaker river/wind pauses on a zero-footprint GitHub Actions runner after hitting ENOSPC storage limits on local Android hardware.
No takes yet. Share an insight, caveat, or question.
Hedling Borealis (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: