BATCH CHANNEL AND SPEAKER EVALUATION WORKSHEET Resource checked October 17, 2026 Companion: https://burki.dev/blog/scribe-v2-vs-assemblyai-batch-channels Original, unperformed evaluation plan. No provider jobs were submitted. This free worksheet does not include paid processing. Keep recordings and credentials in an authorized private system; use internal references here. 1. STATE THE DECISION BEFORE SELECTING SETTINGS Business question: __________________________________________________________ Must distinguish individual people, roles, or only recording sides? ___________ Languages / accents / recording routes: ______________________________________ Funding and recording-use authorization: _____________________________________ Required reviewed-result quality and turnaround: _____________________________ Acceptance owner: ___________________________________________________________ 2. INSPECT EACH SOURCE FILE Private file reference: ______________________________________________________ Duration: _____________________ Channel count: _______________________________ Sample format / rate: ________________________________________________________ Verified channel-to-side mapping and its evidence: ___________________________ Isolated tracks, duplicated mix, crosstalk, or unknown? ________________________ Number of people actually heard on each track: _______________________________ Reference transcript / person map / role map reviewed by: _____________________ Language checked against selected model contract: ____________________________ Do not treat a channel number as a verified individual or an authenticated identity. Preserve the file's original mapping before adding business labels. 3. RECORD THE EXACT CANDIDATE Provider: ___________________________________________________________________ Requested batch model: _______________________________________________________ Selected/returned model evidence: ____________________________________________ Request date / SDK version / endpoint: ________________________________________ Channel mode and diarization settings: ________________________________________ Role inference requested separately? Contract checked? ________________________ Request/job ID: ______________________________________________________________ Submitted / processing / completed / failed / unknown: ________________________ Failure message and retained evidence: _______________________________________ Selected guide models: scribe_v2 and universal-3-5-pro. Separate the production contract from any preview or realtime model. Consult current contracts before a run. Scribe role mode cannot be combined with its multichannel mode. Sources: https://elevenlabs.io/docs/api-reference/speech-to-text/convert https://www.assemblyai.com/docs/pre-recorded-audio/select-the-speech-model 4. ORIGINAL TEST CASES, ALL RESULTS BLANK A: One person on each of two isolated tracks. B: Two staff members successively on one track, caller on the other. C: A mono mix containing all three people. D: Same layout as B with simultaneous speech and a long silence. E: English worked correction: "Tuesday. Sorry, I meant Thursday." Use equivalent authorized source material for both candidates. Keep layouts and languages separate when reporting results. These cases are plans, not evidence that either model passes them. Case/file | Provider/model | Job usable? | Critical correction kept? | Evidence _________ | ______________ | ___________ | _________________________ | ________ _________ | ______________ | ___________ | _________________________ | ________ _________ | ______________ | ___________ | _________________________ | ________ 5. SCORE DISTINCT QUESTIONS, DO NOT HIDE FAILURES Attempted jobs: ______ Completed usable: ______ Failed/incomplete: ____________ Word error count/reference words, normalization rule: _________________________ Person-attribution error and review method: ___________________________________ Business-role errors and reference labels: ____________________________________ Important detail/correction errors: ___________________________________________ Channel mapping preserved? Evidence: __________________________________________ Reviewer disagreements and unresolved examples: ______________________________ UTC job submission: __________________________________________________________ UTC usable completed result available: _______________________________________ Measured boundary and duration: ______________________________________________ Do not compare this batch duration with a realtime first-token measurement. 6. COST AND HANDOFF Account rate/plan/region/currency/date: _______________________________________ Billable duration and channels: ______________________________________________ Additional features and charges: _____________________________________________ Repeated/failed work and billing evidence: ____________________________________ Storage/transfer/review effort: _______________________________________________ Total supported cost / usable reviewed recordings: ___________________________ Fictional target review record: Requested day Thursday, corrected from Tuesday. Receptionist then coordinator on the staff side. Customer role follows a verified channel map. Appointment remains an enquiry, not an accepted booking. This is not generated output. Decision: candidate fits / unsuitable / insufficient evidence Why, with retained examples: __________________________________________________ Next review owner/date: ______________________________________________________ In Burki, authorized operators can start at Calls and locate Call history. Keep its retained transcript separate from any new provider-produced artifact. This guide does not establish a native batch-upload or retranscription connector.