Live voice debate against an AI opponent.
Speak your speech under the clock. The AI listens, interrupts, takes points of information, and gives you the next speech in real time. Sub-200ms voice latency via OpenAI's Realtime API; the model thinks in voice, not in transcripts, so the back-and-forth feels like a real round.
Five modes for different prep needs
- Full round · open speeches, POIs mid-speech, judge RFD when the timer ends. The flagship.
- Cross-examination drill · pure cross-ex Q&A under timer; sharpen questions and concessions.
- Opening speech · drill your PMC/1AC/first prop with an AI judge listening but not arguing back.
- Rebuttal drill · the AI gives a speech; you respond. POIs not allowed; pure rebuttal craft.
- Viva (oral exam) · academic defense mode; the AI examiner pushes on the weak link in your reasoning.
Sixteen distinct opponents
Each persona has its own voice and its own argumentative tendency. The Professor argues from first principles. The Surgeon goes for the throat on warrants. The Diplomat sounds reasonable until you read the next sentence. The Heckler uses POIs like a weapon. Dr. Iyer is a measured Indian-English academic examiner for viva mode. Eight strengths and eight weaknesses per persona so you can scout before the round.
Format-aware, language-aware
The same engine knows how Asian Parli, WSDC, BP, APDA, PF, LD, Policy, Congress, MUN, and viva each differ on evidence rules, structure, and register. POIs in WSDC and BP. Framework first in LD. Tagged cards in Policy. No fabricated citations in impromptu formats. Voice rounds run in 14 languages including Hindi, Mandarin, Spanish, French, Arabic, and Bengali; the RFD comes back in the language you spoke.
Example RFD excerpt
Gov wins, 28 to 27 speaker points. PMC framed the round on stable foreign investment; LOC's three responses were under-warranted and Gov extended the framing through MG. The dispositive clash was the Taiwan-strait disruption probability, which Gov gave a real link chain on (CHIPS Act build timelines) and LOR never weighed back. PMR collapsed cleanly to one voter. To fix: signpost weighing earlier, tighten the rebuttal of the LOC's free-trade counter-narrative into one analytic chain.