Audio quality is the single biggest factor in transcription accuracy. A clean recording transcribes at 99% accuracy. A noisy one drops to 80% or worse — and the gap between those two numbers is measured in hours of manual editing.

The good news is that almost none of the fix is expensive. Most bad transcripts come from three or four avoidable decisions made in the minutes before someone hits record. This guide covers all of it: what to set up before you record, what to do while the recording is running, and the mistakes that quietly ruin otherwise usable audio.

## Before You Record: Pick a Microphone

Your microphone matters more than every other piece of gear combined.

### USB microphones ($30-150)

The best value for most people. Plug into your computer and record directly.

- **Budget**: Any USB condenser mic ($30-50) beats your laptop mic by a wide margin
- **Mid-range**: Dynamic USB mics ($80-120) reject background noise better
- **Pro**: Studio-quality USB condensers ($120-150) deliver broadcast-grade audio

If you are choosing between spending more on a mic and spending more on a room, buy the mic first.

### Headset microphones

Great for meetings and calls. The mic sits close to your mouth, which reduces background noise. Most gaming headsets and business headsets work well.

### Lapel microphones ($15-50)

Clip-on mics are ideal for interviews and presentations. They stay close to the speaker's mouth regardless of movement, which makes them forgiving when your subject turns their head or talks with their hands.

### Smartphone microphone

Surprisingly good in quiet environments. Hold the phone 6-12 inches from your mouth. For better results, use the earbuds' built-in mic.

## Before You Record: Fix the Room

### Reduce echo

Hard surfaces (walls, desks, windows) cause echo that degrades transcription accuracy. Soft surfaces absorb sound:

- Close curtains and blinds
- Place a rug on hard floors
- Add soft furniture (couches, cushions)
- Hang blankets on walls for recording sessions (budget acoustic treatment)

### Minimize background noise

- Close windows and doors
- Turn off fans, heaters, and air conditioners during recording
- Silence phone notifications
- Choose the quietest room available
- Record during off-peak hours if in an office

A small carpeted room with the door shut will beat a beautiful glass-walled conference room every time.

## Before You Record: Dial In Your Settings

### Sample rate

Use 44.1 kHz or 48 kHz. Higher is unnecessary for speech and creates larger files without improving transcription.

### Bit depth

16-bit is sufficient for transcription. 24-bit provides more dynamic range but most speech does not need it.

### Format

Record in WAV or FLAC for maximum quality. If file size is a concern, high-bitrate MP3 (192 kbps+) or M4A (128 kbps+) works fine.

### Gain and levels

- Set gain so speech peaks at -12 to -6 dB
- Leave headroom to avoid clipping (distortion from levels that are too high)
- Do a test recording and adjust before the real session

That last point deserves emphasis. Record thirty seconds, play it back, and actually listen. A test recording is the cheapest insurance in this entire guide, and it is the step people skip most often — usually right before discovering that the mic was muted, the wrong input was selected, or the fan they stopped hearing an hour ago is sitting on top of every word.

## While You're Recording

Setup gets you a good starting point. These habits keep it good for the length of the session.

**Get close, and stay close.** Proximity does more for clarity than any setting. Pick a distance, and hold it — a speaker who drifts in and out gives you a recording with inconsistent levels, which is harder to work with than one that is consistently quiet.

**Speak across the mic, not straight into it.** Angling slightly off-axis keeps hard consonants from hitting the capsule head-on and popping.

**Keep hands, papers, and laptops off the mic's surface.** A desk mic picks up every tap, shuffle, and typed note through the table. If you need to take notes during an interview, do it on paper, away from the mic — or better, let the transcript do it.

**Have everyone say their name at the top.** Blazescribe applies speaker labels automatically, and a clean roll call at the start of the file makes those labels trivial to map to real people afterward. It takes ten seconds and saves you from guessing who Speaker 3 was.

**Say problems out loud when they happen.** When a siren goes past or someone knocks, note it: "sorry, siren." You are leaving yourself a searchable landmark. Combined with word-level timestamps, it turns a vague "somewhere in the middle" into an exact spot you can jump to.

**Don't change the environment mid-session.** The AC you turned off stays off. Moving the mic, opening a window, or switching rooms halfway through means the second half of your file sounds like a different recording.

**Monitor with headphones.** Listening live is how you catch a dead mic, a loose cable, or a channel recording silence — while you can still do something about it, rather than an hour later.

**Record a backup on a second device.** A phone running a voice memo in the corner costs you nothing and has saved a lot of interviews. Files corrupt, batteries die, apps crash.

## Common Mistakes That Wreck a Transcript

Most bad transcripts trace back to the same handful of errors.

**1. Trying to fix it in post instead of fixing the source.** Cleanup tools can only work with what was captured. If a voice was too far from the mic to be recorded clearly, no amount of processing will invent the missing detail. Spend the effort before you record, not after.

**2. Recording too hot.** This is the asymmetry that catches people out: audio that is too quiet can be normalized back up afterward, but audio that clipped is permanently distorted. Those peaks are gone. When you are unsure, err on the quiet side and leave the headroom.

**3. Using a Bluetooth headset as the recording mic.** The moment a Bluetooth headset's microphone activates, the connection typically drops into a low-quality voice mode, and everything you record inherits it. Wired is better. A cheap wired headset will out-record an expensive wireless one.

**4. Assuming names and jargon will come through.** Proper nouns, acronyms, product names, and technical terms are the hardest things in any recording, because they are the words with the fewest contextual clues around them. Say them clearly. If a recording is dense with unusual names, spell the important ones out at the start of the file.

**5. Letting people talk over each other.** Crosstalk is the fastest way to degrade both the words and the speaker labels — when two voices land on top of each other, there is no clean signal for either. In group recordings, one person at a time, with a small pause between turns, is worth far more than any hardware upgrade.

**6. Re-encoding an already-compressed file.** Every pass through a lossy codec throws away detail permanently. Exporting your MP3 to another MP3 to "make it smaller" degrades it again. Go from your original source file to the transcript in one step, and keep the original.

**7. Mixing languages without meaning to.** Blazescribe transcribes audio in 50 languages, and it handles each of them well — but a recording that hops between two of them mid-sentence is the hardest case you can hand any transcription system. Where you have a choice, keep one recording to one language.

**8. Proofreading the transcript from the top.** If you know the audio got rough in three places, you do not need to reread the whole thing. Use word-level timestamps to jump straight to the moments you flagged during recording, check those, and move on.

## After the Recording: Cleanup That Actually Helps

If your recording already has background noise, you can improve it before transcription.

### Noise reduction

Tools like Audacity (free) have noise reduction features:

1. Select a segment of pure background noise (no speech)
2. Use it as a noise profile
3. Apply noise reduction to the entire recording
4. Be conservative — aggressive noise reduction makes speech sound robotic

That last step is where most people overshoot. Heavy-handed noise reduction hollows out the speech along with the hum, and a robotic voice transcribes worse than a slightly noisy one. Take out less than you think you need.

### Normalization

If the audio is too quiet, normalize it to bring levels up to a standard range without clipping.

## The Impact on Accuracy

| Audio quality | Typical accuracy |
|---------------|-----------------|
| Studio/quiet room + good mic | 98-99% |
| Quiet room + laptop mic | 94-97% |
| Office + headset mic | 92-96% |
| Noisy environment + phone mic | 80-90% |
| Very noisy + poor mic | 70-80% |

Read that table as a budget. Every row you move up is editing time you never have to spend. Investing 5 minutes in audio setup can be the difference between a usable transcript and one that requires heavy editing — and unlike editing, the setup cost is paid once and applies to every recording you make afterward.

Get the most from your recordings. [Sign up for Blazescribe](/signup) and experience what 98%+ accuracy looks like on quality audio.