Skip to content
VidzTranscripts

12 Recording Tips for More Accurate Transcriptions

By the VidzTranscripts team · · Updated · 3 min read

Modern speech recognition is remarkably accurate, often above 95% on clean speech. Most mistakes in automatic transcripts come from the recording, not the software. The good news is that small changes in how you record make a big difference, and most of them cost nothing.

1. Get the microphone close

Distance is the single biggest factor. A cheap lapel microphone clipped to a shirt beats an expensive camera microphone three metres away, because the voice is much louder than the room around it. As a rule of thumb, every time you double the distance, the voice gets four times quieter relative to the noise and echo.

2. Choose a quiet, soft room

Hard walls and bare floors create echo, which smears words together. Rooms with curtains, carpets, sofas and bookshelves sound much drier. Turn off fans and air conditioning while recording if you can. A small room full of soft furnishings beats a large empty conference room.

3. Avoid music under speech

Background music is great for viewers but hard for transcription. If you edit videos, transcribe the voice track before you add music, or export a voice-only version for transcription.

4. One person at a time

Overlapping speech is difficult for both humans and machines. In interviews and meetings, let people finish before replying. A short pause before answering also helps the transcript break into paragraphs at the right places.

5. Record each speaker separately when possible

For podcasts and remote interviews, tools like Riverside, Zencastr or a multitrack recorder capture each person on their own track. Clean individual tracks give the best transcripts.

6. Set the right language

Auto-detect works well for long recordings in one language, but choosing the language yourself avoids mistakes on short clips, heavy accents or recordings that start with music or silence.

7. Check your levels

Audio that is too quiet gets lost in noise; audio that is too loud distorts, and distortion can't be undone. Aim for speech peaking around -12 dB to -6 dB on your recorder's meter, and do a ten-second test recording before anything important.

8. Speak clearly, not slowly

You do not need to talk like a robot. Normal, clearly pronounced speech at a natural pace is ideal. Mumbling and trailing off at the end of sentences cause most errors.

9. Recording on a phone

10. Online calls

In video calls, each person's microphone matters more than the software. Ask participants to use headsets, to mute when not speaking and to avoid speaker-phone mode. If your call software can record separate audio tracks for each participant, turn that option on.

11. Rescuing a recording that's already made

If the recording is noisy, you can sometimes improve it before transcribing. Free audio editors such as Audacity include noise reduction and normalisation. Use them gently: aggressive noise reduction makes voices sound watery and can lower accuracy rather than raise it. Two steps that are almost always safe are trimming long silences or music at the start, and normalising a very quiet recording so the speech is at a sensible level.

12. Proofread names and numbers

Even with perfect audio, uncommon names, brand names and figures are the most likely words to be wrong. A quick search through the transcript for these is usually all the editing you need.

How much difference does it make?

Accuracy is usually measured as word error rate: the share of words that are wrong, missing or added. On a clean, close-miked recording, modern models often get fewer than one word in twenty wrong. The same speech recorded from across a noisy room can easily have several times as many errors. That is the difference between a transcript you can use after a quick skim and one you have to correct line by line, and it's decided at the moment you press record.

Need a transcript?

Convert your video or audio to text and subtitles for free.

Transcribe a file

Popular converters