Larger pills mean the word was used more frequently.

The word map shows keywords from both speakers in real time. Each bubble is a word — its color shows who uses it and whether both speakers share its meaning.

Tap any word bubble on the map to see how each speaker uses it. Tap a speaker name at the top to read their full transcript, with key words highlighted.

Speaker A / Speaker B — read each speaker's full transcript as it was spoken. Words from the map are highlighted inline so you can see them in their original flow.

Insights — shows word counts, vocabulary overlap percentage, a breakdown of words by category, and a convergence-over-time graph tracking how the speakers' vocabularies have evolved during the performance.

Search — find any word or phrase across both speakers' transcripts. Results show excerpts with surrounding context, and you can filter by speaker.

Every word spoken passes through five filters before it can appear on the map:

1. Length — Words shorter than 4 characters are discarded (the, is, to, an).

2. Stop words — About 200 common English words are removed: conjunctions (that, this, which), filler (actually, really, basically), conversational noise (yeah, okay), and transcription artifacts (laughs, applause).

3. Stemming — Words are reduced to their root form. Vulnerability, vulnerable, and vulnerabilities all become one entry. This helps detect overlap between speakers who use different forms of the same word. The stemmer is algorithmic, not a dictionary — safe and safety are treated as different roots.

4. Frequency — A word must appear at least twice in a speaker's transcript to qualify. This filters out one-off words and transcription noise.

5. Pruning — When both speakers' combined vocabulary produces more candidates than the map can hold, the system raises the bar — only the most frequently used words survive. Longer conversations produce more candidates, so only the most prominent terms make the cut.

This system uses co-occurrence analysis to map how each speaker uses language. For every keyword a speaker says, the system records which other keywords appear within five words of it. This builds a unique linguistic fingerprint — a co-occurrence neighborhood — for each word in each speaker's vocabulary.

Each speaker's transcript produces its own network of words and connections. The visualization overlays these two networks: words unique to one speaker appear on their side (blue or amber), while words used by both speakers appear in the center as shared (purple).

For example, a psychotherapist might say "trust" near vulnerability, repair, attachment. A cryptographer might say "trust" near verification, protocol, certificate. The word is the same, but the surrounding context is entirely different — revealing how professional frameworks shape the meaning of shared vocabulary.

The system measures these patterns; you interpret what they mean.

The speakers' voices are transcribed using AI speech-to-text. After that, the analysis is entirely mathematical — word counting and co-occurrence comparison. There is no AI interpreting what the words mean or drawing conclusions.

This is deliberate. Rather than handing you an AI-generated summary, this system gives you the raw linguistic patterns and invites you to be the analyst. You decide what it means when two people use the same word differently. The interpretation is yours.

Pinch to zoom the map. Drag to pan. Double-tap to reset the view.