01
Aligns speech over time
Multi-feature dynamic time warping maps transcript tokens and pause intervals across reference and participant audio.
Multimodal AI Hackathon 2026 · Track C
Speech feedback, grounded in the moment.
OratoMetric explores how acoustic signal processing can turn speech-performance critique into precise, time-aligned insights and practical coaching.
Contrastive analysis16 kHz audio
Participant recordingAligned to speech
00:0000:0800:1600:24
!Rushed cadence spikeFlaw region linked to the aligned speech segment
4performance formats
6speech flaw categories
5controlled quality tiers
16 kHzaudio processing pipeline
What it does
From acoustic detail to useful feedback.
Compare a performance with a reference, locate meaningful differences, and connect each observation to the part of the speech where it occurs.
02
Finds specific vocal patterns
Examines pitch, cadence, pauses, energy and articulation to identify six defined rubric failure modes.
03
Makes findings actionable
Flaw regions connect acoustic measurements to rhetorical impact, coaching guidance and A/B audio audition.
The approach
Compare. Locate. Explain.
Designed to focus on changes in delivery rather than raw differences between individual voices or recording levels.
1
Normalize the signalAnalyze 16 kHz mono audio with pitch- and energy-aware speaker normalization.
2
Align reference and performanceUse multi-feature DTW to associate words and pauses across both streams.
3
Ground the coachingMark the affected time region, explain the detected pattern and suggest a practice focus.
Monotone pitch compressionRushed cadenceDragging tempoMissing rhetorical pauseEnergy dropoutArticulation slur
Built for speaking
One analysis framework. Four performance formats.
The project includes curated reference baselines and a controlled degradation spectrum for contrastive evaluation.
A
Persuasive oratory
Evaluate the shape and timing of expressive delivery.
B
Dramatic declamation
Inspect vocal contrast, pacing and rhetorical pauses.
C
Debate & interpretive reading
Explore performance consistency across competitive speaking and text interpretation.
Take a closer look at the code.
Explore the repository for implementation details, evaluation notes, technical deliverables and project updates.
View on GitHub