RhetorIQ — Turn Speech into Measurable Insights
Inspiration
Effective communication is an essential skill for students, professionals, and public speakers. However, improving the way we speak is difficult when feedback is limited to subjective opinions such as "speak more clearly," "slow down," or "use better pauses."
We wanted to explore a more measurable approach to speech practice. This inspired us to build RhetorIQ, an AI-powered speech analytics application that helps users understand their speaking delivery through acoustic measurements and reference-based comparisons.
Our goal is to make speech practice more structured by turning audio recordings into meaningful, timestamped feedback that users can review and learn from.
What It Does
RhetorIQ analyzes speech recordings and presents measurable insights into delivery characteristics, including:
- Pacing: Understand speech rate and variations in speaking speed.
- Pauses: Examine pauses and their timing throughout a recording.
- Pitch expression: Measure relative pitch variation.
- Vocal energy: Analyze changes in acoustic energy.
- Spectral characteristics: Use spectral centroid measurements to study aspects of the audio signal.
- Reference-based comparison: Compare a participant's recording against a reference recording to identify measurable delivery differences.
- Timestamped evidence: Locate relevant intervals so users can investigate specific parts of their speech.
The application is designed to support evidence-based practice rather than relying entirely on subjective feedback. Acoustic measurements indicate delivery differences; they do not independently establish a speaker's emotions, intentions, or personality.
How We Built It
We developed RhetorIQ as a Python-based speech analytics application with an interactive Streamlit interface.
The workflow begins with the user's speech recording and, when available, a reference recording and transcript. The application processes the audio, extracts relevant acoustic features, and presents the resulting measurements through a structured analysis dashboard.
Our implementation focuses on several stages:
- Input handling: Accept speech recordings and optional transcript inputs.
- Audio processing: Prepare the recordings for acoustic analysis using the project's existing processing pipeline.
- Feature extraction: Measure speech rate, pause behavior, relative pitch variation, energy, and spectral characteristics.
- Reference comparison: Evaluate supported delivery differences between the participant and reference recordings.
- Results visualization: Present scores, timelines, and findings in an interface designed to make the analysis easier to understand.
We also improved the user experience by organizing the workflow into clearer steps and separating standard speech analysis from reference-based comparison.
Challenges We Faced
One of the main challenges was turning complex acoustic measurements into feedback that users could understand and act upon.
Different speakers may pronounce the same words at different speeds, pitches, and energy levels. These differences do not necessarily indicate better or worse communication. Designing useful comparisons therefore requires care in selecting metrics and interpreting their results.
Another challenge was ensuring that reference-based scores remain meaningful. Two recordings should not be treated as directly comparable simply because both contain speech. Their spoken content and alignment need to be considered before interpreting differences.
We also worked on making the interface easier to navigate, handling uploaded audio and transcript inputs, and presenting multiple metrics without overwhelming users.
These challenges reinforced the importance of validating inputs, distinguishing independent acoustic measurements from reference-relative scores, and communicating the limitations of automated analysis.
What We Learned
Building RhetorIQ helped us understand how audio signal processing, feature extraction, and interactive visualization can work together in an AI application.
We learned that a useful AI product requires more than calculating metrics. Input validation, reliable comparisons, understandable visualizations, and honest explanations of limitations are equally important.
We also gained practical experience in Python application development, Streamlit UI design, audio-processing workflows, debugging, and preparing an application for deployment.
What's Next
We want to continue improving RhetorIQ by strengthening transcript and recording verification, refining reference alignment, and making the results more actionable.
Future improvements could include more detailed word-level feedback, personalized practice suggestions based on measured delivery patterns, and progress tracking across multiple practice sessions.
Our long-term vision is to make speech practice more accessible, measurable, and consistent—helping people understand not only what they say, but also how they deliver it.
RhetorIQ turns speech into measurable insights, giving speakers a clearer starting point for improvement.
Log in or sign up for Devpost to join the conversation.