EchoDepth Research & Whitepapers
Conference papers, technical architecture documentation and applied research from the Cavefish team. All papers are available to cite with attribution.
The Hemisphere Advantage: How Understanding Brain Lateralisation Transforms Customer Communication
The two cerebral hemispheres do not simply duplicate each other's work; they attend to the world in fundamentally different ways. The right hemisphere apprehends context, tone, imagery, novelty, and the felt body. The left hemisphere abstracts, analyses, sequences and codifies. This whitepaper sets out the neuroscience underpinning that idea (Sperry, McGilchrist, Damasio, Kahneman), the marketing-effectiveness evidence that validates it commercially (Binet & Field's IPA Databank analysis; Kantar's facial-coding programme), and introduces EchoDepth as the measurement layer that closes the loop between hemispheric theory and operational advertising measurement.
Emotional campaigns are roughly twice as likely to produce very large profit gains as rational ones (31% vs 16%) — IPA Databank
The optimal long-term budget allocation is 60% brand-building (right-brain led) and 40% activation (left-brain led)
Ads evoking strong emotions are 4× more likely to build long-term brand equity — Kantar
VAD (Valence-Arousal-Dominance) scoring captures the pre-conscious emotional response that self-report surveys miss
Core principle: right brain opens the door; left brain signs the contract — sequencing determines success
Feel to Know: Using Emotion to Measure Real Understanding
Traditional defence training assessment relies on post-training quizzes and self-reported comprehension, which fail to capture real-time understanding or identify struggling learners during critical safety instruction. This paper presents EchoDepth, a multimodal AI platform that measures trainee comprehension through continuous emotional analysis during training delivery. By detecting facial expressions via Action Units (AU) in contextual combinations, the system identifies confusion, confidence, hesitation and cognitive stress in real time, enabling dynamic content adjustment and verified attention monitoring.
Confusion signals emerge 3–5 minutes before trainees self-report difficulty or fail quiz questions
Contextual AU combinations — not single expressions — distinguish concentration from confusion
Verified engagement monitoring creates a richer audit record than completion certificates alone
Disengagement patterns provide a content quality signal that accelerates course improvement cycles
Case studies: HUMX (avatar-plus-emotion architecture) and Welsh Rugby Union (bidirectional communication analysis)
EchoDepth
EchoDepth is a multimodal deep learning framework designed to extract, interpret and generate emotional responses from textual, vocal and visual inputs. This whitepaper provides an in-depth overview of EchoDepth's architecture, training methodology and deployment strategy, covering the three-layered architecture combining pre-trained transformers, prosodic voice analysis and multimodal fusion models.
Three-layered architecture: NLP transformers + prosodic voice analysis + multimodal fusion
VAD (Valence-Arousal-Dominance) scoring across text, voice and visual modalities
92% accuracy on GoEmotions (text), 87% F1-score on RAVDESS (speech), 89% on CMU-MOSEI (multimodal)
Late fusion strategy with cross-modal attention mechanism and neural decision fusion
API-first deployment: RESTful API, low-latency response, edge AI compatibility
EchoPitch: Enhancing Sales Through Emotional Intelligence & Avatar-Based Training
Sales success demands more than product knowledge. Research from psychology and marketing shows that emotional intelligence, vocal tone and realistic practice are critical drivers of persuasion. This whitepaper summarises scientific evidence supporting EchoPitch, a training platform that uses AI-driven avatars to help sales professionals improve their emotional delivery, covering vocal tone research, avatar-based simulation outcomes and hybrid AI/human coaching models.
Salespeople with high EI and high emotional self-efficacy earn approximately 270% higher commissions
Focused vocal tone increased Kickstarter funding willingness by 30%; stressed tone decreased it by 26%
Tone conveys 84% of meaning in telephone conversations — dominant signal when visual cues are absent
AI coaching improves content retention by 50% after 48 hours vs human coaching alone
Personalised avatar doppelgangers increase motivation and self-belief for low-confidence learners
EchoDepth whitepapers and conference papers may be shared and cited with attribution to Cavefish and the named author. For commercial reproduction, bulk distribution or to request papers not available for direct download, contact hello@cavefish.ai.