Certifying a Transcript From Someone Else's Recording
A firm asks you to produce a certified transcript from a digital video recording of a two-hour-forty-seven-minute administrative hearing, recorded by the agency's own AV system. You did not attend. The file has two audio channels: channel 2 is dead throughout, and the camera switches between a wide shot and a close-up of the presiding officer. There is a consistent offset of roughly five seconds between the audio and the video, so that lip movement and speech do not align. Several speakers are never introduced by name, and two of them sound similar. At least twice, speakers overlap for several seconds. The firm wants a certified transcript with speaker identifications and a five-business-day turnaround, and asks whether you can 'just label the unknowns as Speaker 1 and Speaker 2 and be done.'
1. What do you do before transcribing a single word, and what do you document?
Read the worked answer
The reliability of the source is the whole question, so establish it first.
- Verify the file. Duration, format, checksum, creation metadata, and whether it is the original or a copy. Record all of it.
- Test both channels. Confirm channel 2 is dead across the entire file rather than in one segment, and note it. If the surviving channel is mono-summed and partly phase-cancelled, that affects intelligibility and must be documented.
- Measure the A/V offset on a few known events and record the figure. Do not attempt to 'fix' it in the file you were given; work from the original and document the offset in a note so any reader can reconcile video timestamps to transcript timestamps.
- Listen at increased playback speed for the whole file before beginning, to catch dropouts, clipping, and unintelligible passages while you can still ask questions.
- Ask the firm, in writing, for the hearing's agenda, appearance list, and any exhibits or caption page. Names are the highest-value thing they can give you.
- Set expectations in writing about what a certification from a recording can and cannot attest to.
2. How do you handle speaker identification and the unidentified voices?
Read the worked answer
Do not accept the firm's instruction as given. Arbitrary labels with no basis become permanent, and at the next stage someone will assert that Speaker 2 was the agency director.
- Identify everything the recording itself supports - direct address ('Mr. Chairman,' 'Counsel, your witness'), self-identification, sequence, and role. Where an identification rests on inference, that is acceptable; where it rests on nothing, it is not.
- For genuine unknowns, use a neutral label with a stated basis, e.g. 'Unidentified Speaker 1 (male voice, appears from 00:41:12),' and list the labels in a preface or speaker key. That is materially better than 'Speaker 1' because it tells the reader what was and was not determined.
- Do not merge two similar voices without support. Over-identification is a worse error than an honest 'unidentified,' because it silently attributes testimony to the wrong person.
- Ask. Offer the firm the list of unidentified spans with timestamps and invite them to supply names from the appearance list; accepting their answer is fine if it is confirmed in writing and the confirmation is noted.
3. How do you render the overlapping speech, and what should your certification actually say?
Read the worked answer
Overlap. Render both speakers as best the single channel allows, marked so the reader knows simultaneity occurred - a bracketed note such as '[speaking simultaneously]' or a stacked-format convention applied consistently. Where one voice is genuinely unrecoverable, say so at that point rather than inventing a coherent exchange. Do not resolve a crosstalk collision into a clean question and answer: the fact of the interruption may be the point.
Certification. A certification is a personal attestation that the transcript is a true record of the proceedings. Where you worked from a recording you did not make, do not sign language that implies you heard the room. Appropriate certification states what you did: that the transcript was prepared from a specified recording, identified by file name and duration, and is a true and accurate transcription of that recording to the best of your ability. If the jurisdiction's prescribed form cannot be varied, disclose the source in a separate reporter's note and say so to the firm before starting.
Finally: neither the A/V offset nor the dead channel needs to appear in the transcript body - both belong in a dated note accompanying the delivery, copied to whoever ordered the work.