Verbatim vs. Polished Transcription: Choosing the Right Mode
Not all transcripts are created equal. A customer support manager reviewing agent performance needs to hear every "um," hesitation, and self-correction. A business analyst just needs the clean facts. That's why VoxParse offers two transcription modes: verbatim and polished.
Both modes use the same AI pipeline and cost the same per hour. The only difference is how faithfully the transcript represents the original speech. Listen to the sample call below, then compare the output side-by-side.
Listen to the original call
This is a real customer service call (telephony quality, 8kHz mono). The customer is calling about a device exchange offer.
Side-by-side comparison
Both transcripts below were generated from the same audio file using VoxParse's POST /v1/transcribe endpoint. The only difference is the mode parameter.
Agent: Thanks for calling AT&T. My name is Erica. Can I have the wireless telephone number you are calling about today?
Customer: 406-539-1202.
Agent: Confirm that number is 406-539-1202. Is that correct?
Customer: Yes.
Agent: How may I help you today?
Customer: I got a letter saying I can get the BlackBerry Torch.
Agent: Okay, I'd be happy to help with that, but first I need to verify some information for security.
...
Agent: Once you receive the return label, attach it to the phone box and drop it in any USPS blue collection box.
Agent: I have completed the exchange setup. Do you have any questions, Mr. Milgen?
Customer: No.
Agent: We value and appreciate your business. Thank you for calling AT&T. Have a great day.
Agent: Thanks for calling AT&T, my name is Erica. Can I have the wireless telephone number that you're calling about today?
Customer: 406-539-1202.
Agent: So confirm that number was 406-539-1202, is that correct?
Customer: Yes.
Agent: So how may I help you today?
Customer: Oh, I got this letter saying I can get the BlackBerry Torch.
Agent: Mm-hmm. Okay, I don't worry, I'd be more than happy to see with us, but before we proceed, for security reasons of your account, can you please verify some information for me first?
...
Agent: Okay, and also once you receive the Black... I mean the return label, just simply have fixed it to the phone box and drop it in any USPS blue box, okay?
Customer: Drop it in what?
Agent: In any USPS blue box.
Agent: Okay, so I'm done feeling that all the information needed here for the exchange. So do you have any questions regarding the information that I provide to you, Mr. Milligan?
Agent: Okay, we do value and appreciate your business. Thank you for calling AT&T and have a great day.
Highlighted spans show elements preserved in verbatim that polished removes: filler words, false starts, self-corrections, and natural speech patterns.
Key differences
| Aspect | Polished | Verbatim |
|---|---|---|
| Filler words | Removed (um, uh, so, oh) | Preserved exactly as spoken |
| False starts | Cleaned up | Kept ("the Black... I mean the return label") |
| Self-corrections | Smoothed to final intent | Both attempts preserved |
| ASR errors | Fixed (e.g., "USBs" → "USPS") | Fixed (same — ASR errors are always corrected) |
| Readability | Professional, report-ready | Natural speech, may read rough |
| Word count | ~15-30% shorter | Full length |
| Cost | Same ($0.49/hr) | Same ($0.49/hr) |
When to use each mode
Use Polished for:
- Business reporting — Summaries for managers who need clean, scannable transcripts
- CRM integration — Storing call notes that agents and account managers will reference
- Analytics dashboards — Clean text produces better keyword and topic extraction
- Customer-facing summaries — Follow-up emails quoting what was discussed
Use Verbatim for:
- QA and agent coaching — Hearing exactly how agents communicate, including hesitation patterns
- Compliance audits — Legal review requires word-for-word accuracy, not paraphrased content
- Dispute resolution — When a customer claims "that's not what was said," verbatim is your evidence
- Training data — Building speech models or fine-tuning NLP requires unedited source material
How to use it
Add the mode parameter to your POST /v1/transcribe request. Default is polished.
# Polished mode (default)
curl -X POST https://api.voxparse.com/v1/transcribe \
-H "X-API-Key: YOUR_API_KEY" \
-F "[email protected]"
# Verbatim mode
curl -X POST https://api.voxparse.com/v1/transcribe \
-H "X-API-Key: YOUR_API_KEY" \
-F "[email protected]" \
-F "mode=verbatim"Both modes return the same structured JSON response with AI analysis, compliance flags, sentiment, and financial extraction. The only difference is the transcript_cleaned field in the response.
Pro tip: Run the same file through both modes and store both transcripts. Use polished for your dashboard and verbatim for your compliance archive.
Try both modes free
Start with $10 in prepaid credits. Same price for both modes — $0.49/hr, everything included.
Get your API key →Bottom line
Most transcription APIs give you one output and call it done. VoxParse gives you the choice: a polished transcript for business consumption, or a verbatim transcript for when every word matters. Same audio, same price, same API call — just add mode=verbatim.