SpeakTraceANALYSIS STUDIO
Connecting to studio

THE ANALYSIS STUDIO

Hear it. See it. Refine it.

A closer look at the pace, pauses, and energy behind a reading.

CONNECTING

COMPARISON

Choose a recording to begin

EXAMPLE

Explore a real archive recording and compare a controlled variation.

Compare with

Generated variations are labelled separately from analysis findings.

Actual audio alignment. Measured evidence.

LISTEN & COMPARE

Every difference has a location.

AWAITING ANALYSIS
ReferenceParticipantEach track has its own time axis
Run a comparison to reveal measured waveforms and word alignment.

ACOUSTIC DETAIL

Compare the contours

Analyze audio to compare measured acoustic features at corresponding transcript words.

WORD BY WORD

The shared transcript

Select a word after analysis

Your transcript will appear here.

PacingPausesEnergyHighlighted words link to measured evidence

MEASURED DIFFERENCES

Evidence timeline

0 INTERVALS

Start with the sound.

Analyze a comparison to see where delivery changes, and why it was flagged.

BEHIND THE ANALYSIS

Evidence, with context.

SpeakTrace compares two readings of the same words. It shows how they differ, without pretending there is one right way to speak.

01

Align the words

A pretrained English speech model locates words in each recording. Poor alignment can cause scoring to be withheld.

02

Measure the difference

A fixed rubric compares local pace, pauses, and relative energy. Findings are tied to audio intervals and measurements.

03

Listen with context

Replay each interval. A flagged difference is a practice cue, not a diagnosis, emotion label, or universal quality judgment.

How recordings are processed

Connecting to check this studio’s deployment and storage policy. Analysis is enabled after the connection is verified.

How to read the evidence

Energy is relative to each recording's voiced median, not calibrated loudness. Different recording durations have separate time axes. Generated demo variations are identified as controlled examples; their labels do not enter inference.

Loading the versioned rubric…