Every speech coach knows the math. Thirty students record a four-minute speech and you’re holding two hours of raw footage before you’ve scored a dimension or typed a comment. Graded the way most of us learned — watch, pause, rewind, type a paragraph — it swallows five or six hours, which is why weekly feedback quietly dies mid-season in so many programs. Here is the system we use to hold it to one afternoon, and what we eventually automated.
Three rules before the workflow
- One watch per video. Rewinding is where afternoons go to die. If you missed something, the audience probably lost it too — score what landed.
- Score during the watch, not after. Impressions reconstructed later drift toward the middle bands and blur between students.
- Comments are evidence plus one next step — not essays. A timestamp and an instruction outperform a paragraph of adjectives.
The system, step by step
- Batch by assignment, not by student. Grade all thirty takes of the same prompt back to back. Your ear calibrates to the task and holds steady; interleaving different assignments resets it every time.
- Set up a two-screen cockpit. Video on one side; rubric, roster checklist, and a notes column on the other. Paper works fine.
- Watch at 1.25–1.5×. Middle-school pacing tolerates it easily for content and structure. Drop to normal speed only when the drill is specifically about delivery.
- Score every dimension on the first pass, using anchored 1–5 bands (our free rubric pack has full descriptors). Torn between two bands? Take the one you could defend to a parent with a timestamp.
- Write timestamps, not transcripts. “2:14 — signpost dropped, audience loses the second point.” Two per student: one strength, one fix.
- Say the comment instead of typing it. Sixty to ninety seconds of spoken feedback is roughly a 150-word note — two to three times faster than typing, and it comes out in your coaching voice. Transcribe it or attach the audio.
- End every student with exactly one next step. “Next take: number your responses — two answers, first… second…” One instruction gets executed. Five get skimmed.
- Publish on a fixed weekday and track trends. A Clarity line that goes 2 → 3 → 4 over a month is the story worth telling students and parents — not any single week’s number.
The honest arithmetic
Done well: about three minutes of watching (at 1.3×), one minute of scoring, two minutes of spoken comments — call it six minutes per student. Thirty students is three hours, once a week, every week of the season. The system works. It is also the ceiling on your roster: past thirty-something students per grading coach, weekly feedback collapses into “great job!” by November. That ceiling — not rubric quality, not coach effort — is usually why feedback dies.
What we automated (and what we didn’t)
We build Augmind Studio and run our own middle-school program on it, so this part is first-hand. The split that works is AI drafts, coach judges:
- The AI does the drafting pass — it processes each take, transcribes it, scores every rubric dimension, and writes draft feedback with timestamped evidence, minutes after the student uploads.
- The coach does the judging — opens the review workbench, clicks any evidence line to jump the video to that exact moment, edits or overrides scores, dictates a personal note, and signs it.
- Nothing reaches a student or family without a coach’s name on it.
Verification runs one to two minutes per student. In our own program, feedback that used to slip to the weekend now goes out the same day — and “grade everyone, every week” stopped being aspirational and became the default.
Frequently asked questions
How long should grading one speech video take?
With a batching system and an anchored 1–5 rubric, about six minutes per student: three minutes of watching at 1.25–1.5× speed, one minute of scoring, two minutes of spoken comments. With an AI drafting pass that a coach verifies and signs, one to two minutes per student.
Is watching at 1.25–1.5× speed fair to students?
For content, structure, and argument quality, yes — middle-school pacing tolerates it easily. Drop back to normal speed when the assignment is specifically about delivery: pacing drills, dramatic interpretation, pause work.
What should written feedback on a speech include?
Three things: one timestamped strength, one timestamped fix, and exactly one next step for the next take. Scores justified by rubric band language rather than general impressions. Students act on one instruction; they skim five.
Does AI grading replace the coach?
No — the division of labor is drafting versus judging. The AI watches, transcribes, scores against the coach's rubric, and drafts feedback with timestamped evidence. The coach reviews it, overrides what's wrong, adds a personal note, and signs it. Nothing reaches a student or family without a coach's name on it.
Keep the afternoon. Ship the feedback.
Students record; AI drafts rubric-scored feedback with timestamped evidence; your coach reviews and signs. Free 21-day trial, no credit card.