Vocal Well

Detect voice disorders early — using just your phone.

Comment

🎙️ VocalWell: Detect voice disorders in 10 seconds using just your voice and AI — no hospitals, no appointments, just health.

🌟 Inspiration

We come from families where teachers, elderly relatives, and even neighbours have struggled with vocal issues — hoarseness, strain, even loss of voice. But no one ever took it seriously until it got worse.

That’s when we realised — voice disorders affect over 500 million people globally, and most of them don’t even know it until it’s too late. In India, especially in rural areas, access to ENT specialists is extremely limited, and vocal health is one of the most ignored areas of public healthcare.

What moved us most was knowing that women, teachers, singers, and the elderly are at even greater risk, and there’s almost zero early diagnosis infrastructure.

We were inspired to build VocalWell — to create something simple, private, and useful that anyone, anywhere could use to detect early signs of vocal issues using just their voice.

🎤 What it does

VocalWell is a web app that allows users to:

  • Record or upload a 10-second voice clip
  • Our AI model then analyzes features like pitch, jitter, shimmer, MFCCs
  • It flags early signs of voice disorders — vocal strain, tremors, nodules, etc.
  • Gives an instant voice health score with visual graphs and a personalised report
  • Offers support via a multilingual AI helpline that provides:
    • Voice therapy advice
    • Mental health support
    • Nearest ENT suggestions (region-wise)
  • All user data is encrypted and stored on IPFS — users control access

The whole process takes less than a minute, and can be done from any phone — no doctor, no login, no cost.

🔧 How we built it

We built VocalWell in 48 hours using:

  • AI/ML Stack: TensorFlow, Keras, Librosa, SciPy, NumPy
  • Custom deep learning model using BiLSTM + CNN trained on open datasets of pathological voices
  • Voice features extracted include MFCCs, shimmer, jitter, pitch, etc.
  • Frontend: Next.js + TailwindCSS + Framer Motion
  • Backend: Flask (hosted on PythonAnywhere)
  • Storage: IPFS (via NFT.storage) to ensure privacy and decentralisation
  • AI Agent Helpline: Built with Langchain + Open-source LLMs, supports 10+ languages (Hindi, Punjabi, Tamil, Arabic, English, and more)

We also used Google Colab for model training and prototyping.

🚧 Challenges we ran into

  • Training an accurate model with limited and unbalanced voice datasets
  • Ensuring the AI is both inclusive and medically relevant without clinical data access
  • Multilingual voice guidance was tricky — getting tone and intent right in regional languages
  • Handling audio noise and low-quality recordings, especially on cheap phones
  • Integrating IPFS smoothly with Python backend
  • Making the whole app usable and accessible in under a minute — especially for rural users

🏅 Accomplishments that we're proud of

  • We built our own deep learning model from scratch — no pre-trained shortcuts
  • Completed a fully working MVP in 48 hours (frontend + backend + AI + IPFS)
  • Included multilingual mental health + therapy guidance, not just voice detection
  • Designed it to work with low-quality microphones and slow internet
  • Created a product that is private, accessible, and built for real people, not just demo judges

📚 What we learned

  • How to combine healthcare and AI meaningfully without clinical access
  • The power of privacy-first Web3 tools like IPFS for sensitive data like voice
  • How impactful even a small tool can be when built with empathy
  • The sheer scale of vocal health problems — and how ignored this space is
  • How many people (especially women and teachers) suffer silently due to lack of awareness

🚀 What's next for VocalWell

  • Add detection for stress, early Parkinson’s, respiratory issues
  • Launch a WhatsApp/IVR version for rural India
  • Improve our dataset and model accuracy with field data
  • Partner with public health organisations, NGOs, and schools
  • Launch in local clinics as a screening tool
  • Enable auto-scheduling with local ENT specialists
  • Explore tie-ins with speech therapy programs

We also plan to make VocalWell open-source for non-profits and integrate it with India's Ayushman Bharat mission.

🌍 Real-World Need + SDG Alignment

Why VocalWell matters:

  • Voice disorders are often ignored in public health, yet impact communication, livelihood, and mental well-being
  • Lack of ENT doctors in rural areas (only 1 ENT for ~1 lakh people in some districts)
  • Early diagnosis is supported by WHO, UN, and voice health bodies as essential for preventing chronic conditions

Aligned with UN SDGs:

  • Goal 3: Good Health and Well-being
  • Goal 9: Industry, Innovation and Infrastructure
  • Goal 10: Reduced Inequalities

📊 Market Opportunity

  • Global voice disorder diagnosis market expected to grow to $3.2B by 2030
  • India has over 18 million teachers, a core user base
  • 2.5 million professional singers and voice artists globally
  • 70%+ smartphone penetration in rural India — growing rapidly

There is a clear market for voice-based wellness tools, especially mobile-first, AI-powered, and privacy-friendly solutions.

🫂 Closing Thought

VocalWell isn’t just an app — it’s a step towards giving every voice the chance to be heard, protected, and cared for.

Built With

  • css
  • flask
  • framer-motion
  • google-colab
  • html
  • ipfs
  • keras
  • langchain
  • librosa
  • next.js
  • nft.storage
  • numpy
  • open-source-llms
  • python
  • pythonanywhere
  • scipy
  • tailwind-css
  • tensorflow
Share this project:

Updates