Nedimo: An Evidence-Grounded Multilingual Pipeline for AI-Generated Sleep Stories

Georgios Makridis; Stefanos Kampouris

cs.AIOct 29, 2026

Nedimo: An Evidence-Grounded Multilingual Pipeline for AI-Generated Sleep Stories

Authors: Georgios Makridis, Stefanos Kampouris

Organizations: Department of Digital Systems, University of Piraeus, Piraeus, Greece

Abstract

Generative audio systems can produce large libraries of personalized bedtime content, but ordinary media objectives (retention, completion, and novelty) are poorly aligned with the goal of supporting a transition toward sleep. Direct evidence for narrated bedtime audio in adults is mixed, and a recent active-controlled trial did not support such audio as a standalone treatment for sleep disturbance. We translate this uncertain evidence into conservative engineering requirements and present a multilingual pipeline for generating and narrating low-arousal sleep stories. The implemented prototype separates deterministic planning from large language model generation, represents content and delivery parameters explicitly, shapes each script through a four-phase attention taper, validates structural and lexical constraints, performs a bounded error-aware retry, and quarantines unresolved outputs. Provider-agnostic text-to-speech synthesis is followed by paragraph-aware joining, two-pass loudness normalization, and a publication gate that re-measures the saved, encoded file against fixed loudness, loudness-range, and true-peak limits. The product data model records stop position, replay, early abandonment, and completion while treating these events as ambiguous behavioral proxies rather than clinical outcomes. We contribute an evidence-to-requirement mapping, a traceable guardrail architecture with explicit acceptance thresholds, technical evaluation criteria, and a prospective protocol for testing sleep-onset effects. No clinical efficacy is claimed; the system is positioned as a relaxation and attentional-support tool whose benefit and possible reliance must be evaluated empirically.

Figures & tables

Study	Design and sample	Intervention	Implication for the present system
Harmat et al. [ 1 ]	Controlled study, $n=94$ students	45 min/night for three weeks: music, audiobook, or control	Spoken content is not intrinsically sleep promoting; content and delivery require explicit design.
Kirk et al. [ 2 ]	Randomized within-person crossover, $n=40$	Sleepcasts, music, mindfulness, and no-intervention weeks	Supports feasibility and possible physiological effects, but subjective benefit was not specific to stories.
Economides et al. [ 3 ]	Multi-arm pilot, $n=300$ working adults	App-based stories or music over four weeks	Supports scalable delivery and retention; pilot findings require active-controlled confirmation.
Mermer and Arslan [ 4 ]	Quasi-experimental intensive-care study, $n=56$	One 30-min audiobook-plus-music session	Provides context-specific positive evidence but does not isolate narration or generalize to home use.
Vazzaz et al. [ 5 ]	Four-arm active-controlled RCT, $n=495$	Stories, sounds, skills, or digital control for four weeks	No superiority on the primary outcome; motivates conservative claims and controlled evaluation.
Berglund et al. [ 6 ]	Cross-sectional self-selected online survey, $n=1{,}136$ Swedish adults	Self-reported bedtime audiobook use	Describes respondents’ motives, familiarity, and perceived reliance; cannot establish prevalence or efficacy.

TABLE I: Direct and Closely Related Adult Bedtime-Audio Evidence

Evidence or risk	Design requirement	Implemented control
Engaging, pleasant imagery can redirect unwanted pre-sleep thought [ 10 ] .	Capture attention without threat, urgency, or an unresolved goal.	Concrete opening, continuous low-stakes movement, sensory palette, limited dialogue, and bounded novelty.
Bedtime-story trials are mixed and the strongest active-controlled result is null [ 5 ] .	Do not represent the system as treatment or guarantee sleep.	Claim boundary in prompts, interface copy, validation rules, paper, and evaluation plan.
Daily-listening survey respondents more often reported reusing a book and difficulty falling asleep without audio [ 6 ] .	Support familiarity without engagement-maximizing habit mechanics.	Recurring-world mode, autoplay off, no streak objective, and reliance-oriented telemetry.
Environmental sound can disturb sleep; masking evidence is inconsistent [ 13 , 14 ] .	Keep level and dynamics controlled and ambience optional.	Voice-first mix, two-pass loudness normalization, post-encode loudness, loudness-range, and true-peak gate, and timed fade.
LLMs violate schemas and numerical length constraints [ 18 , 19 ] .	Verify rather than trust generation instructions.	Parseable front matter, server-computed word and taper budgets, deterministic validators, one targeted retry, then quarantine.
TTS realization varies across voices, languages, and providers [ 20 , 21 ] .	Separate narrative intent from synthesis configuration and waveform QA.	Provider adapter, paragraph chunks, locale settings, pronunciation metadata, silent join, and post-synthesis measurement.

TABLE II: Evidence-to-Requirement Mapping

Lang.	Rendered (min)	Error	LUFS	TP (dBTP)	LRA (LU)
en	20.4	$+2.0\%$	$-22.22$	$-2.68$	4.4
de	20.6	$+2.9\%$	$-22.23$	$-2.81$	4.2
fr	17.4	$-13.0\%$	$-22.28$	$-2.90$	4.4
hi	12.4	$-38.2\%$	$-22.17$	$-2.98$	3.0
ja	15.2	$-24.1\%$	$-22.20$	$-2.98$	4.4

TABLE III: Pilot Measurement: Five Published Stories (20-min Target)

Audio Generation

Audio generation creates speech, music, and sound from text, media, symbolic controls, or reference audio, with growing focus on controllability and realism.

See all 137 papers

Feb 27, 2026

SleepLM: Natural-Language Intelligence for Human Sleep

We present SleepLM, a family of sleep-language foundation models that enable human sleep alignment, interpretation, and interaction with natural language. Despite the critical role of sleep, learning-based sleep analysis systems operate in closed label spaces (e.g., predefined stages or events) and fail to describe, query, or generalize to novel sleep phenomena. SleepLM bridges natural language and multimodal polysomnography, enabling language-grounded representations of sleep physiology. To support this alignment, we introduce a multilevel sleep caption generation pipeline that enables the curation of the first large-scale sleep-text dataset, comprising over 100K hours of data from more than 10,000 individuals. Furthermore, we present a unified pretraining objective that combines contrastive alignment, caption generation, and signal reconstruction to better capture physiological fidelity and cross-modal interactions. Extensive experiments on real-world sleep understanding tasks verify that SleepLM outperforms state-of-the-art in zero-shot and few-shot learning, cross-modal retrieval, and sleep captioning. Importantly, SleepLM also exhibits intriguing capabilities including language-guided event localization, targeted insight generation, and zero-shot generalization to unseen tasks. All code and data will be open-sourced.

Zongzhe Xu, Zitao Shuai, Eideen Mozaffari +3

University of California, Los Angeles

Multimodal Pretraining Cross-Modal Retrieval Language Modeling

Ask PDF

Jul 13, 2026

BackgroundMellow: A Multi-Modal Cohesive Framework for Narrative-Driven Rich Cinematic Soundscape Generation

Generating immersive, synchronized and cinematic audio for long-form textual narratives remains a significant challenge in multi-modal AI. While current Text-to-Audio (TTA) frameworks successfully synthesize isolated sound effects, they struggle with narrative cohesion, temporal alignment, and cinematic emotional depth. We present BackgroundMellow, a framework that treats story-to-audio generation as a precise orchestration and signal processing problem. This framework is enabled without ground-truth through a master-specialist agent architecture that decomposes text into precise and multi-layered audio cues, generates each category of sounds with suitable specialist model, and superimposes the soundscapes to create a unified and aligned audio segment. Our pipeline is built over Tango2 latent diffusion model for environmental synthesis alongside a novel Cinematic BGM Retriever mined from professional soundtracks. To automate the sound mixing process, we use an NLP based module that predicts precise audio parameters, like start time, duration, and relative loudness, based on the narrative timeline. We further empirically evaluate and show the efficacy of the proposed framework leveraging nearest-neighbor retrieval against a curated dataset of YouTube cinematic trailers to measure temporal synchronization, coverage, and spectral richness.

Ajitesh Jamulkar, Aritra Hazra

Department of Computer Science & Engineering, Indian Institute of Technology Kharagpur, India.

Text-to-Audio Generation Audio Editing Audio Generation

Ask PDF

Jun 17, 2026

Better Adherence, Richer Context: A Field Evaluation of LLM-Powered Conversational Voice Diaries for Sleep

Sleep diaries are central to behavioral sleep medicine and cognitive behavioral therapy for insomnia, yet daily completion is difficult to sustain, and static forms often provide limited context for interpreting night-to-night sleep variation. We designed an LLM-powered conversational voice diary that delivers clinically grounded morning and evening sleep diary questions through proactive smart-speaker prompts, structured conversational intake, and adaptive follow-up dialogue. We evaluated the system in a four-week between-subjects field study with 30 university students, comparing it with a text-based mobile diary using matched diary items, reporting windows, and reminder intervals. Compared with the text-based diary, the conversational voice diary showed higher adherence and elicited more detailed contextual self-report about routines, stressors, environmental conditions, and other sleep-related factors. Participants also described the voice diary as easier to integrate into daily routines, despite longer perceived completion time. However, voice-based conversational intake produced lower completeness for some structured diary fields, revealing a trade-off between expressive richness and structured precision. These findings show both the promise and the challenge of using LLM-powered conversational voice assistants for longitudinal health self-report.

Amama Mahmood, Bokyung Kim, Honghao Zhao +4

The Johns Hopkins University, USA · Department of Psychiatry and Behavioral Sciences, The Johns Hopkins University School of Medicine, USA

Healthcare Spoken Dialogue Systems Conversational Agents

Ask PDF

Abstract

Audio Generation

Explore similar work

SleepLM: Natural-Language Intelligence for Human Sleep

BackgroundMellow: A Multi-Modal Cohesive Framework for Narrative-Driven Rich Cinematic Soundscape Generation

Better Adherence, Richer Context: A Field Evaluation of LLM-Powered Conversational Voice Diaries for Sleep