AI reading assessment for K-5 automatic WCPM scoring in 60 seconds

AI-powered assessment

ReadingFluency uses AI to transcribe oral reading, classify errors, and compare WCPM to grade-level benchmarks in one workflow.

This page explains how ReadingFluency turns a 60-second oral reading into a scored assessment. Students read aloud, AI aligns the speech to the passage, classifies errors, and returns WCPM plus benchmark context automatically.

Every error is reviewable | Passage-aware transcription | FERPA and COPPA aligned

ReadingFluency does not hand a recording to generic dictation and read back the result. The transcription model is given the passage the student is reading, and the scoring pipeline aligns the transcript to that passage, classifies each error by ORF convention, and keeps every call reviewable.

Passage-aware transcription

The transcription model is given the text the student is reading, so it expects the passage's vocabulary and names. The scoring pipeline then handles what young readers actually do:

• Developing pronunciation
• Slower reading rates
• Hesitations and self-corrections
• Dialect variants like toward/towards, matched rather than marked

• Passage-aware transcription, not generic dictation
• Research-standard error counting for mispronunciations, substitutions, omissions, transpositions, and missed proper nouns
• Instant WCPM plus Hasbrouck & Tindal benchmark context

The result: every scored error is reviewable. Inspect the transcript, overturn any call, and the saved WCPM recalculates.

1. Audio capture. The student reads aloud while the device microphone records. The completed recording is uploaded for transcription and scoring.
2. Speech-to-text processing. A speech-to-text provider transcribes the completed recording. The passage text gives the model context for the vocabulary and names the student is reading.
3. Word alignment. Each spoken word is aligned to the passage text, identifying matches, mismatches, and omissions.
4. Error classification. Mismatches are categorized: mispronunciation, substitution, omission, transposition, or a name or place said wrong. Self-corrections and repetitions are flagged but not counted as errors.
5. WCPM calculation. Words read correctly in the 60-second assessment are counted using the same ORF error rules on every assessment. Teachers can review and overturn flagged errors.
6. Benchmark comparison. Score compared to Hasbrouck & Tindal 2017 norms for instant grade-level context.

"But is AI really accurate enough for assessment?"

Every call is reviewable.

Every error reviewable

The teacher inspects the transcript, overturns any flagged error, and the saved WCPM recalculates to match the call.

Consistent scoring rules

The same error-counting rules apply on every assessment: counted errors, self-corrections, and repetitions follow the shared ORF convention.

Research aligned

WCPM is words read correctly in 60 seconds, compared to Hasbrouck & Tindal 2017 norms. The teacher stays the authority on every flagged word.

We know student data privacy isn't optional—it's essential. Here's how we protect every assessment:

Audio processing

• Completed recordings are uploaded for transcription
• Recordings are deleted from our servers after processing
• ReadingFluency does not keep recordings for playback
• Transcription providers process audio under their own retention and security terms

Data protection

• Encrypted connections carry recordings and results
• Access to student records is checked against the signed-in account
• Service providers process information to deliver the assessment

School privacy

• FERPA and COPPA aligned
• Data processing agreements available on request
• Data ownership remains with schools and teachers

You control your data

• Request an export of student records
• Request deletion of student records
• We never sell student data
• No advertising in the product

The models behind the score are maintained and pinned deliberately, and the teacher always has the final word.

Model updates

The transcription model is vendor-maintained and pinned explicitly in the scoring pipeline, so a model change is a deliberate decision rather than a silent one.

Two-step scoring

A second model verifies each flagged error before the score is saved, dismissing false positives such as a number read aloud in a different form than it was printed.

Teacher override

When you overturn a call, the saved score recalculates. Your judgment is the final word on every assessment.

You always get the current models—no upgrades to purchase, no new versions to install.