Mbelys
IoT-Based Goat Vocal Analysis for Real-Time Stress & Reproductive Activity Detection
Mbelys combines IoT audio sensing, machine learning classification, Google Cloud Platform infrastructure, and a mobile application to monitor goat vocalizations in real time, transforming biological acoustic signals into actionable livestock health insights.
Manual inspection misses critical reproductive & health windows.
Traditional livestock management relies on periodic physical inspection. In goat farming, identifying estrus (Masa Subur) and acute stress signals requires constant observation. Farmers often miss narrow fertile windows or early injury distress, leading to lower breeding success rates and delayed medical care.
24/7 Continuous acoustic sensing with instant cloud AI alerts.
Mbelys places an ESP32-S3 digital microphone node directly in goat pens. Captured vocalizations stream directly into a Google Cloud pipeline where a 2-stage Convolutional Neural Network (CNN) model classifies sound patterns into 8 specific physical states—delivering real-time alerts and smart recommendations straight to the farmer's smartphone.
MY ROLE — BUILDING THE SYSTEM BEHIND MBELYS
I owned the hardware engineering and end-to-end cloud pipeline, bridging physical acoustic sensing, GCP infrastructure, machine learning inference, and database triggers for the mobile app.
IoT ENGINEERING
Built and handled the complete IoT hardware device used to capture high-fidelity goat vocalization data in farm environments.
- Microcontroller: ESP32-S3 N16R8 module configuration
- Acoustic Sensor: INMP441 I2S digital omnidirectional mic
- Hardware Assembly: Enclosure wiring & power management
- Data Acquisition: 22kHz 16-bit PCM raw audio sampling
- IoT Connectivity: Wi-Fi transmission & field testing
CLOUD ENGINEERING
Designed and maintained the cloud pipeline connecting the IoT device, machine learning system, and mobile application.
- Cloud Microservices: Google Cloud Run container hosting
- Object Storage: Google Cloud Storage audio buckets
- Database Sync: Firebase Realtime & Firestore integration
- Data Pipeline: Secure IoT-to-Cloud streaming architecture
- Ops & Maintenance: GCP infrastructure monitoring & IAM
ML & CLOUD INTEGRATION
Integrated the machine learning inference pipeline into the cloud environment and connected its outputs to the mobile app.
- Model Hosting: Containerized CNN model endpoint on Cloud Run
- Payload Parser: Structured JSON prediction payload builder
- Event Triggers: GCS bucket event triggers on file upload
- Firestore Writes: Automated real-time result database updates
- App Pipeline: Feeding smart recommendations to Android client
THE DEVICE
An ESP32-S3-based IoT device equipped with an INMP441 digital microphone captures goat vocalizations in farm environments and sends collected audio data through the cloud pipeline for analysis.
REAL-WORLD FARM TESTING
Hardware calibration and acoustic signal verification conducted directly at goat farm facilities, testing device durability, Wi-Fi streaming stability, and cloud pipeline sync under real environmental conditions.
On-site testing at goat pen facilities, debugging the ESP32-S3 audio buffer transmission to Google Cloud Storage.
FROM GOAT VOCALIZATION TO ACTIONABLE INSIGHT
End-to-end data pipeline connecting hardware audio capture, cloud object storage, machine learning inference container, Firestore database, and mobile user interface.
My Integration Ownership: Steps 03, 04, and 05 represent the core IoT-to-Cloud architecture I engineered and deployed to ensure low-latency audio transmission and automated ML prediction sync.
2-Stage CNN Audio Classification
The acoustic classification engine uses a Convolutional Neural Network (CNN) trained on spectrogram representations of goat bleats to detect overall physical state and specific stress conditions.
Cloud Deployment Note: The CNN model was containerized and hosted on GCP Cloud Run. I built the API wrapper that receives audio buffers from Google Cloud Storage and outputs formatted JSON predictions to Firestore.
Primary Health & Heat Detection
Fine-Grained Condition Diagnosis
THE DATA PIPELINE IN ACTION
Raw audio captured by the IoT device is transferred to the cloud, stored in Google Cloud Storage buckets, processed via Cloud Run, and synchronized live in Firestore database collections.
goat-audio-bucket/raw/).MBELYS MOBILE APPLICATION
Android application providing livestock farmers with real-time pen status alerts, heat detection notifications, and environmental condition recommendations.
Daftar Kandang Overview
Lists all monitored goat pens with real-time health badges (e.g. "Masa Subur" detected) synced instantly from Firestore.
Pen Management & Pairing
Allows farmers to register new pens, configure location parameters, and pair ESP32-S3 IoT microphone hardware.
Real-Time Condition & Insights
Displays detected vocal state ("Masa Subur terdeteksi"), current pen temperature/humidity, and AI recommendation advice.
TRANSFORMING LIVESTOCK MONITORING
Traditional Manual Monitoring
- Manual physical observation requiring full-time presence
- Time-consuming pen inspections across large farms
- High rate of missed estrus/heat windows ("Masa Subur")
- Delayed injury or social isolation detection
- Subjective inspection with no historical data records
Cloud & IoT Automated Intelligence
- 24/7 automated continuous acoustic monitoring
- Instant cloud-connected alerts streamed directly to smartphones
- 85% – 88% accuracy in heat and stress classification
- Actionable smart recommendations for immediate farm interventions
- Scalable GCP infrastructure supporting multiple pens simultaneously