INNOVART 2027 / ART × TECHNOLOGY

A rehearsal partner.
The performer decides.

In spoken-word performance, a pause or a quiet line may be the point. SpeechTrace helps a performer hear exactly where a take differs from a reference, then decide which differences to keep and which to practice.

Try the arts workflow

  1. Load a moderate-volume or extra-pause example. Listen to both versions at the flagged passage.
  2. Write your rehearsal intention: for example, “Keep the deliberate pause; explore steadier volume.”
  3. Mark each moment Intentional choice, Practice this passage, or Review later. Your choice does not alter the numeric similarity score.
  4. Download rehearsal notes. The on-screen preview and plain-text file contain the intention, interval, words and your decision. No account or notes database is needed.
  5. Compare a new take. Changing recordings or reanalyzing clears old annotations so they cannot be mistaken for new feedback.

Actual outcome

A deployed, runnable speech comparison engine and a working rehearsal annotation/export interface. The existing data provides a controlled demonstration, not validation with poets or performers. Testing with consenting artists is proposed future work.

Seven-minute presentation

The editable deck includes speaker notes and a live demonstration plan. Leave approximately three minutes for questions.

Download the editable presentation

Required event steps

Devpost submission and the Discord-shared presentation form are due January 2, 2027 at noon EST. A live presentation on January 3 is mandatory. A Devpost entry alone does not complete InnovArt participation.

Official event · InnovArt Discord

Provenance & limits

The core app was built October 6, 2026 for Multimodal AI Track C and was entered there. October 7 adds performer-authored intentions, intentional/practice decisions, downloadable rehearsal notes, and event guides. Codex assisted with implementation and writing. Vosk is pretrained. Similarity does not measure artistry, correctness, disability, accent quality or fairness. Uploaded audio is processed for each request without application persistence; hosting request metadata exists. Rehearsal intentions and decisions remain in this tab until downloaded and clear when results are invalidated.