Transcribe Audio with Background Noise
How to get good transcriptions from recordings with noise, strong accents, or poor quality: the right engine, correct language, key terms, and recording tips.
Street recordings, echoing hearings, WhatsApp audios captured from far away: noise is the biggest enemy of transcription. The good news is that there is an engine specialized in this — Standard — and a few simple tweaks that greatly improve the result.
The first choice: the Standard engine
The Standard engine was built for difficult recordings: background noise, varied accents, and low quality. And it has a bonus: it's the cheapest engine on the platform, consuming 1 cycle per minute (vs. 4 for Premium).
It is available on all plans, including the free one. On the Pro and Premium plans (and on hourly packages), Standard runs with the most robust model in the family (large-v3), even better for difficult audio.
With the 120 cycles on the free plan, you can transcribe up to 120 minutes on Standard — the same balance yields 30 minutes on Premium. For noisy audio, Standard is usually better AND cheaper.
Checklist for difficult audio
Choose the Standard engine when uploading
On the new transcription screen, switch the engine from Premium to Standard before starting.
Set the language manually
Noise interferes with automatic language detection. If you know the audio is in Portuguese, select Portuguese explicitly.
Provide key terms (if using Premium)
If you prefer Premium, take advantage of key terms to reinforce proper names and jargon — in noisy audio, those are the words that most often come out wrong. See Custom vocabulary.
Review in the editor
For very degraded audio, set aside a few minutes to review the text in the editor with the synchronized player by your side — listen to the passage, fix it, save. See Edit transcription.
Prevention: how to record better next time
- Move the microphone closer to the speaker — distance is worse than constant noise.
- Avoid talking at the same time as someone else; overlapping voices are the hardest scenario for any AI.
- In meetings, a phone at the center of the table works better than one in the corner of the room.
- If possible, do a 30-second test before the important recording.
No engine can recover what is not in the audio. If the voices are inaudible even to a human, the transcription will reflect that. In such cases, the result works as a draft for manual review, not as a final text.
Frequently asked questions
Which engine is best for audio with a lot of background noise?
Standard. It is the specialist in difficult recordings — and costs 4 times fewer cycles than Premium.
Does the free plan have access to the Standard engine?
Yes. Free, Lite, and Advanced use Standard on the small/medium models; Pro and higher use the large-v3 model.
Is there any noise filter before transcription?
There is no configurable filter — noise robustness comes from the engine itself. Choosing the right engine is what makes the difference.
Does WhatsApp audio work well?
Yes, WhatsApp voice format (OGG/Opus) is natively supported. If the recording itself is noisy, the same tips in this article apply.
Related articles
Didn't resolve? Open a ticket and our team will help you.