Transcript accuracy means faithfully representing recorded speech under the agreed transcription instructions. A percentage is useful only when you know the reference, counting rules, and exclusions behind it. Legal transcription accuracy standards should address both measurable word errors and mistakes that change meaning or assign speech to the wrong person.
What ATS’s published accuracy claim means
American Transcription Services’ pricing page advertises a “99%+ accuracy guarantee,” human-led transcription, and multiple quality checks. Its legal transcription services page states that trained legal transcription professionals review every transcript. These are ATS’s published claims.
Those pages do not explain the percentage’s calculation or provide an independent accuracy audit. The advertised figure should not be treated as an independently verified result or a universal legal requirement. Ask what the guarantee measures, which audio conditions it covers, and what correction terms apply.
How word error rate measures transcript errors
Word error rate, or WER, compares a transcript with a reference transcript and counts three kinds of word errors:
Substitution: a spoken word is replaced with another word.
Deletion: a spoken word is missing.
Insertion: an unspoken word is added.
NIST’s speech-recognition evaluation materials define the calculation as:
WER (%) = (substitutions + deletions + insertions) ÷ reference word count × 100
The denominator is the reference word count, not the delivered transcript’s length. NIST’s SCLITE documentation explains that scoring follows word alignment and that reference accuracy affects the statistics.
Agree on how numbers, contractions, punctuation, and unclear passages are handled. For edited transcription, define permitted omissions before scoring; otherwise, agreed filler removal may count as deletions.
Hypothetical calculation of 99% word-based accuracy
Suppose a checked reference contains 1,000 words. The transcript has six substitutions, three deletions, and one insertion:
Total word errors: 6 + 3 + 1 = 10.
WER: 10 ÷ 1,000 × 100 = 1%.
Word-based accuracy, defined here as 100% minus WER: 99%.
Counting only correctly matched reference words would yield 99.1%, because that measure does not subtract the insertion. The measurement method changes the answer.
This is an illustrative calculation, not ATS performance data. It describes this sample only and does not predict every future recording.
Why a high percentage can hide consequential errors
Standard word-error scoring weights errors equally. A missing filler and a missing negation each count as one deletion, although their effects differ. A whole-file average can also conceal a difficult passage with concentrated errors.
Hypothetical example of one missing word
A witness says, “I did not authorize the payment.” The transcript reads, “I did authorize the payment.” Removing “not” reverses the statement. If this were the only error in a 1,000-word reference, the calculation above would yield 99.9% word-based accuracy. This is a hypothetical example, not an ATS case study.
Speaker attribution also needs its own assessment. A word-only comparison may preserve every word while assigning the statement to the wrong speaker. NIST’s Rich Transcription evaluations distinguish speech-to-text, speaker identification over time, and speaker-attributed transcription tasks.
| Error type | Hypothetical example | Potential effect on meaning | Review action |
| Name substitution | “Miller” becomes “Milner.” | Identifies a different person. | Replay; check supplied name spellings. |
| Negation deletion | “Did not agree” becomes “did agree.” | Reverses the assertion. | Replay the full clause; restore the audible word. |
| Word insertion | “I saw him” becomes “I probably saw him.” | Adds uncertainty the speaker did not express. | Remove the unsupported word after audio review. |
| Date substitution | “May thirteen” becomes “May thirty.” | Changes the chronology. | Replay the date; flag unresolved ambiguity. |
| Amount substitution | “Fifteen thousand” becomes “fifty thousand.” | Changes the stated amount. | Replay the amount; avoid inferring it from documents. |
| Speaker attribution | Witness’s words labeled “Attorney.” | Assigns a statement to the wrong person. | Verify the voice against the recording and speaker log. |
Factors that affect legal transcription accuracy
Recording quality: low volume, clipping, background noise, and dropouts can obscure speech. Preserve the original file.
Overlapping speakers: simultaneous voices can make words or speaker changes inseparable.
Unfamiliar terminology: legal terms, acronyms, and proper names need spelling references checked against the audio.
Accents and speech patterns: careful listening to actual pronunciation helps avoid replacing words with expected phrasing.
Missing context: participant names, roles, and topic information can resolve ambiguities, but context cannot establish an inaudible word.
Flag uncertainty instead of guessing
Reviewers should replay unclear passages and preserve what is recoverable. Use an agreed marker such as “[inaudible 00:14:22]” for unresolved audio or “[overlapping speech 00:18:05]” where voices cannot be separated. Mark uncertain speaker identity too.
Reference materials can support spelling, but plausible wording is not evidence of what was said. Report excluded or unresolved spans alongside an accuracy score so its scope is visible.
Accuracy, formatting, and official transcript status
Linguistic accuracy concerns the recorded words and their attribution. Formatting requirements concern layout, labels, timestamps, and page-and-line numbering. Official transcript status concerns the applicable court’s authorization, preparation, and certification requirements.
For example, federal judiciary policy specifies transcript formatting and requires the designated transcriber of electronically recorded court proceedings to certify the original and copies. In the Southern District of New York, transcripts of electronic court recordings must be requested through Records. An advertised accuracy percentage does not establish compliance with those requirements. Check the court, jurisdiction, and proceeding before ordering an official transcript.
Checklist for assessing a sample transcript
Choose a representative excerpt that includes difficult audio and speaker changes. Review it while listening, then:
Confirm verbatim or edited instructions and permitted omissions.
Check completeness against the recording, including the beginning and end.
Track substitutions, deletions, and insertions against a checked reference.
Check names, dates, amounts, and negations individually.
Verify speaker labels at each change.
Confirm visible uncertainty markers and usable timestamps.
Check formatting separately from word accuracy.
Document corrections with audio timestamps and ask how disputed passages are resolved.
Frequently asked questions
How is transcript accuracy measured?
WER counts substitutions, deletions, and insertions against reference words. Ask for the formula, scoring rules, exclusions, and sample scope. Assess speaker attribution and consequential errors separately.
What should happen when audio is unclear?
The reviewer should flag the passage with an agreed marker and timestamp. Guessing hides uncertainty. Any excluded unclear spans should be disclosed when reporting a percentage.
Does proofreading establish transcript accuracy?
Reading text alone can catch typos or inconsistencies. Checking linguistic fidelity requires listening to the recording; fluent sentences can still contain missing or incorrect words.
Does a word accuracy score include speaker identification?
Only if the scoring method explicitly assesses attribution. Confirm whether incorrect labels are counted, reported separately, or excluded; a word-only score does not establish who spoke.
Does 99% accuracy make a transcript certified or official?
No. A percentage is a measurement or claim. Certification and official status depend on the applicable requirements. Confirm the court’s authorized ordering route and required certification.
Request a free sample or quote from ATS with your recording details and transcript requirements.