Farmers Insurance | The James River District
Finds every word that will expose a Filipino speaker, and what to write instead. Your cleaned script is loaded below.
Paste what a VA actually said. Ranked by how often they said it, and split into traps that came from the script versus ones they reached for on their own.
Measures the two tells that survive good vocabulary: rising pitch on statements, and syllable-timed rhythm. Nothing is uploaded. The audio is analyzed inside your browser and never leaves this device.
Solo read, one take, no second person in the room. Sixty to ninety seconds. Read it like you are talking to a customer, not like you are reading.
Every trap word and what to say instead. Search it when you are writing something new.
| Never say | Say instead | Why |
|---|
These get drilled, not replaced. Your own carrier name is one of them.
| Word | How to say it |
|---|
Rising endings. American statements fall in pitch at the end. A rise reads as a question nobody asked, and as uncertainty. This is the loudest tell after word choice, and the most coachable. The page splits the audio into utterances, measures the pitch change across the last 280 milliseconds of each, and reports what share rise.
Rhythm. English is stress-timed, so neighbouring syllables differ a lot in length. Tagalog is closer to syllable-timed, where syllables come out even. The score is higher when delivery is more English-like.
Two speakers. If a recording has two people on it, their voices are separated by pitch before scoring, so the other person's delivery does not pollute the result. If the two voices are too close to tell apart, the page says so instead of guessing.
It does not detect V said as B, or F said as P. That needs a speech model too heavy to run in a browser. The passage is built to load those sounds deliberately, so one person listening once can catch them. It also measures delivery and word choice, not whether someone is easy to understand. Use it to decide what to change, not to pass or fail people.
The pass and watch thresholds are starting points, not published norms. Record two or three of your own US agents reading the same passage, run them through, and use their numbers as your baseline. Same passage, same page, same room. That comparison is valid; comparing to a published figure is not.
Everything runs in this browser. No transcript, recording, or result is uploaded anywhere. Close the tab and it is gone, apart from the script text and baseline, which are saved on this device only.