How to improve audio quality for better transcription
Microphone, room, distance, file format: 10 simple tips to record clear sound and get a much more accurate automatic transcription.
A transcription AI can only write what it hears. With a clear recording, the transcript is close to perfect; with distant, distorted or noisy audio, errors pile up. The good news: a few adjustments are enough to go from an average result to text that is almost ready to use.
Here are 10 practical tips, from the simplest to the most technical.
1. Bring the microphone closer to the speaker
This is the tip that changes the result the most. At 6 to 12 inches (15 to 30 cm) from the mouth, the voice clearly dominates room noise. A smartphone lying in the middle of a large table mostly picks up echo and air conditioning.
For a meeting, place the phone or microphone as close as possible to the people who speak the most, or use one microphone per speaker.
2. Pick a quiet, "soft" room
Bare walls, tiled floors and large windows bounce sound around: the voice becomes distant and reverberant. A room with curtains, a sofa, a rug or a bookcase absorbs those echoes.
Also think about background noise you no longer notice: computer fans, the fridge, traffic, notifications. Turn off whatever you can before you start recording.
3. Use an external microphone if you record often
A laptop's built-in microphone also picks up the keyboard and the fan. On a small budget:
- A lavalier (clip-on) microphone: ideal for interviews or lectures, it always stays at the same distance from the mouth.
- A USB microphone: perfect for podcasts and video calls.
- Earphones with a microphone: much better than the laptop microphone for calls.
4. Check the recording level
Audio that is too quiet has to be turned up, and the hiss gets louder with it. Audio that is too loud "clips": peaks are flattened and words become hard to recognise.
Record a 10-second test and listen back. The voice should be clear, with no crackling on the loud parts.
5. Avoid overlapping voices
When two people speak at once, neither a human nor an AI can catch everything. In meetings, ask everyone to finish their sentence. In interviews, leave a short pause before answering.
This also improves speaker identification (available on the AudioWhiz Premium plan): each speaker is separated more cleanly.
6. Speak clearly, at a natural pace
No need to sound like a robot. Just articulate, don't swallow the end of your sentences, and avoid turning away from the microphone while speaking. Names, acronyms and technical terms deserve to be said a little more slowly.
7. Record in a good format
If your app lets you choose:
| Setting |
Recommendation |
| Format |
WAV, FLAC or M4A; MP3 works very well |
| Bitrate |
128 kbps or more for MP3, never below 64 kbps |
| Sample rate |
16 kHz minimum, ideally 44.1 or 48 kHz |
| Channels |
Mono is enough for voice |
Avoid converting a file several times from one compressed format to another: each conversion degrades the sound a little.
8. Trim long silences and unnecessary parts
The minutes of silence before a meeting starts or the hold music count towards your transcription time. A quick trim before uploading saves minutes and gives you a cleaner transcript.
9. Select the right language
Automatic detection works well, but if the recording mixes several languages or starts with a passage in another language, choosing the main language yourself avoids surprises.
10. Proofread the key passages
Even with perfect audio, quickly checking numbers, names and quotes is a good habit. With a good recording, this review takes a few minutes instead of an hour.
In short
Microphone close by, quiet room, correct level, one person at a time: these four habits are enough to get a very reliable transcript in the vast majority of cases.
Ready to try? Upload your next recording to AudioWhiz: the free plan includes 60 minutes of transcription per month, so you can compare the result before and after applying these tips.