Contrastive Speech Analytics & Temporal Flaw Grounding

• 

  Live Dashboard: Interactive speech comparison with acoustic feature visualizations and candidate intervals for further review

Inspiration

I wanted to explore how AI could make speech comparison more interpretable by identifying where two recordings differ acoustically. My goal was to build a tool that helps users locate potential speech differences instead of relying only on overall results.

What it does

Contrastive Speech Analytics & Temporal Flaw Grounding compares reference and participant speech recordings using features such as RMS energy, zero-crossing rate, and estimated pitch. It visualizes acoustic patterns on a normalized timeline and highlights candidate intervals for further review.

How we built it

I built the project using Python, Librosa, NumPy, Pandas, Matplotlib, and Streamlit. I organized the code into separate modules for feature extraction, comparison, and candidate detection, then integrated them into an interactive dashboard where users can upload recordings and explore the visual results.

Challenges we ran into

One major challenge was comparing recordings with different durations. I addressed this by mapping their features onto a normalized timeline. I also worked through audio-format compatibility, debugging visualization issues, and deploying the dashboard successfully.

Accomplishments that we're proud of

I'm proud that I developed this project independently and deployed a working interactive dashboard. I successfully integrated audio feature extraction, comparative visualizations, and candidate interval detection into one application.

What we learned

I gained practical experience in audio analysis, modular Python programming, data visualization, debugging, and deployment. I also learned that acoustic differences do not automatically indicate speech flaws, making careful interpretation and human review essential.

What's next for Contrastive Speech Analytics & Temporal Flaw Grounding

Next, I aim to improve Contrastive Speech Analytics by adding word-level alignment, more precise speech-flaw detection, and support for diverse accents and speaking styles. I also plan to evaluate its performance using labeled recordings and enhance the dashboard to provide clearer, more reliable insights.

Built With

• acousticanalysis
• ai
• audioanalytics
• audioprocessing
• contrastiveanalysis
• datascience
• datavisualization
• featureextraction
• github
• hackathon
• librosa
• machine-learning
• matplotlib
• numpy
• opensource
• pandas
• python
• pythondevelopment
• signalprocessing
• speechanalysis
• speechtechnology
• streamlit
• timeseriesanalysis
• webdashboard