ADVANCEMENTS IN AI AND BIOACOUSTIC SIGNAL PROCESSING - ATAL FDP Presentation 15.02.2025

I am grateful to be a Resource Person in the AICTE ATAL Academy-sponsored Faculty Development Program (FDP) on Advancements in AI and Bioacoustic Signal Processing on 15.02.2025. In the session, the evolution of Natural Language Processing (NLP) through the Transformer architecture, inspired by the groundbreaking research paper "Attention Is All You Need", was examined.

Using Jupyter Notebook, we demonstrated key NLP pre-processing techniques, including tokenization, stemming, and lemmatization, while also discussing the evolution of Large Language Models (LLMs) and the latest advancements, including the impact of DeepSeek at the time. A major highlight was the application of Generative AI for dental report summarization using LLMs and the demonstration of Agentic AI concepts using Langflow.

Building on these concepts, I presented three case studies:
1️⃣ Lung Acoustics for COPD diagnosis
2️⃣ Speech Acoustics for Dysarthria diagnosis
3️⃣ Heart Acoustics

These discussions emphasized how Generative AI and Agentic AI could transform the field of bioacoustic signal processing. Additionally, the impact of biases in healthcare data and their implications on AI-driven diagnostics were explored.

A heartfelt thank you to the organizers, the coordinator Ms. G. Nithya, AP/ECE, and participants for an engaging and insightful discussion and also to the Department of CSE(AI&ML), Sri Krishna College of Technology, Coimbatore for the constant encouragement and support. Looking forward to more such collaborative learning experiences!

2. Dr. John Amose
Assistant Professor (Sr.G)
Department of CSE (AI & ML)
Sri Krishna College of Technology (SKCT), Coimbatore
Mob: [phone]
linkedin.com/in/johnamose/
AICTE Training and Learning (ATAL) Academy sponsored
Faculty Development Program on
Transforming Healthcare through Wearable Technologies and AI
ADVANCEMENTS IN AI AND
BIOACOUSTIC SIGNAL PROCESSING
15.02.2025 at 3:30pm

3. Advancements in AI -NVIDIA1.
Perception AIa.
Generative AIb.
Agentic AIc.
Physical AId.
Roboticse.
Project Digitsf.
Foundational Technology2.
Natural Language Processinga.
Transformer Architectureb.
Bioacoustic Signal Processing3.
Case Study 1 - Lung Acoustica.
Case Study 2 - Speech Acousticb.
Case Study 3 - Heart Acousticc.
Wearable Technology & Responsible AI4.

4. Advancements in AI -
NVIDIA
SECTION 1

6. Consumer Electronics Show (CES) - Las Vegas 2025 - 07 Jan 2025
Highlights - NVIDIA CEO Jensen Huang Keynote at CES 2025

7. Consumer Electronics Show (CES) - Las Vegas 2025
Highlights - NVIDIA - Generative AI
DENTAL REPORT

8. Consumer Electronics Show (CES) - Las Vegas 2025
Highlights - NVIDIA - Agentic AI

9. Consumer Electronics Show (CES) - Las Vegas 2025
Highlights - NVIDIA - Physical AI

10. Consumer Electronics Show (CES) - Las Vegas 2025
Highlights - NVIDIA - Autonomous Vehicles

11. Consumer Electronics Show (CES) - Las Vegas 2025
Highlights - NVIDIA - Robotics

12. Consumer Electronics Show (CES) - Las Vegas 2025
Highlights - NVIDIA - Project Digits

13. Foundational
Technologies
SECTION 2

14. NATURAL LANGUAGE PROCESSING (NLP)
Text Pre-processing
Tokenization
Stemming
Lamnatization
Bag of Words
Vector Embedding
Word2vec
Cosine similarity (Semantic)
Large Language Models (LLMs)
Transformer

15. This is mathematically implemented using three main components:
Query (Q) – Represents the current processing element (e.g., a
word in a sentence).
1.
Key (K) – Represents all elements in the input that can be
attended to.
2.
Value (V) – Contains the actual information from the input.3.
Scaled Dot-Product Attention (Used in Transformers)
Each Query-Q computes a similarity score with all Keys-K,
determining which inputs are important. This is done as:
How Does Attention Work?
Attention assigns different weights to different
input elements, prioritizing the most important
ones.
TRANSFORMERS – ATTENTION IS ALL YOU NEED

16. Your paragraph text
TRANSFORMER
ARCHITECTURE

17. Bioacoustic
Signal
Processing
SECTION 3

18. BIOACOUSTIC SIGNAL
PROCESSING
Case Study (Pulmonologist)
Lung Acoustics for Respiratory Care
Implementation of
Perception AI
Generative AI
Agentic AI

19. CASE STUDY 1
LUNG ACOUSTICS - COPD DIAGNOSIS
Analog Stethoscopes
Digitization

20. Digital Stethoscopes

21. Indian - Digital Stethoscopes

22. Multi-Feature
Medical Devices

23. Wearable Respiratory Monitor - Strados Lab

24. Wearable
Respiratory
Monitor
Strados Lab

26. LUNG ACOUSTICS

27. Chest Sounds
Lung Sounds
Spectrogram MFCC
Time-Frequency Representation
Denoised Signal
Bandpass Filter [20-2000 Hz]
Downsampling to 4000Hz

28. •MFCC 0 [Energy Parameter] has a high correlation and redundancy, with average absolute correlations
around 0.49.
•Wheeze - Spectral shape are better captured by the higher-order MFCCs (spectral details of the sound)
•COPD tends to produce more consistent sound patterns due to chronic obstruction,overall energy level
might not fluctuate much over time
MFCC Coefficient Analysis - Correlation

29. Hyperparameter Tuning - Heatmap of GridSeachCV
Mean Test Scores

34. Precision Score
Recall Score
F1 Score

35. Langflow is a low-code tool for developers that makes it easier to build powerful
AI agents and workflows that can use any API, model, or database.

36. GENERATIVE AI
Data Augmentation for Lung Sound Classification
AI-Powered Lung Sound Interpretation
Explainable AI (XAI) for Lung Sound Predictions
AI-Generated Personalized Reports
Generative AI Chatbot for Medical Queries
Implementation Plan: Agentic AI for Bio-Acoustic Analysis
We will create an AI agent that can:
Analyze lung sound spectrograms using a trained CNN model.
Answer medical queries using LLMs.
Retrieve medical research papers based on user input.
Provide diagnostic insights based on lung sound classification.
AGENTIC AI

37. BIOACOUSTIC SIGNAL
PROCESSING
Case Study (Neurologist & Speech Pathologist)
Speech Acoustics- Dysarthria
Implementation of
Perception AI
Generative AI
Agentic AI

39. Thoppil MG, Kumar CS, Kumar A, Amose J. Speech signal analysis and pattern recognition in
diagnosis of dysarthria. Ann Indian Acad Neurol 2017;20:352-7 [SCI]

40. Comparison of pitch of normal speech with
spastic speech. Demonstrates F0 jitter
Comparison of pitch of normal speech with ataxic speech.
Demonstrates F0 break
Comparison of pitch of normal speech with
extrapyramidal speech. Demonstrates F0
monotonicity
Demonstrates that the formant range (F1 and F2) decreases as severity
of speech increases. Comparison of formant range of normal speech
with severe dysarthria

41. GENERATIVE AI
AI-Based Speech Reconstruction
Personalized Voice Cloning
AI-Powered Speech Prediction
Implementation Plan: Agentic AI for Bio-Acoustic Analysis
We will create an AI agent that can:
Autonomous Speech Severity Analysis Agent
Self-Improving Speech Therapy Agent with feedback on exercise
Intelligent Speech Prediction Agent for speech correction.
AGENTIC AI

42. BIOACOUSTIC SIGNAL
PROCESSING
Case Study (Cardiologist)
Heart Acoustics
Implementation of
Perception AI
Generative AI
Agentic AI

43. Acoustics Lab Research Associate
Andrew McDonald

44. GENERATIVE AI
Data Augmentation for Heart Sound Classification
AI-Powered Heart Sound Interpretation
Explainable AI (XAI) for Heart Sound Predictions
AI-Generated Personalized Reports
Generative AI Chatbot for Medical Queries
Implementation Plan: Agentic AI for Bio-Acoustic Analysis
We will create an AI agent that can:
Analyze Heart sound spectrograms using a trained CNN model.
Answer medical queries using LLMs.
Retrieve medical research papers based on user input.
Provide diagnostic insights based on heart sound classification.
AGENTIC AI

45. Wearable
Technologies
&
Responsible AI
SECTION 4

46. RESPONSIBLE AI FOR
WEARABLE HEALTHCARE APPLICATIONS
DATA & BIASES
Demographic
Unbalanced
Language
Political bias
PRIVACY & DATA SECURITY
HUMAN-AI COLLABORATION
TRANSPARENCY & EXPLAINABILITY
ROBUSTNESS & RELIABILITY
Data Augmentation and Generative
Challenges in Acceptance

47. What has changed?
Technology
Rise of Open-Source AI Challenging
Proprietary Models
Lower AI Deployment Costs Increasing
Accessibility
Specialized & Multimodal AI Reducing
Dependence on General Models
Wearable Technologies
Edge AI & On-Device Processing
Lower Costs & Faster Innovation
Enhanced Multimodal Integration
Geopolitical
Data is the new oil, a war on global dominance

48. CONTACT ME
[phone]
[email protected]
linkedin.com/in/johnamose
SKCT, Coimbatore

49. FOR YOUR ‘ATTENTION’
linkedin.com/in/johnamose
THANK YOU!