Audio quality checker
Find out whether a recording is clear enough to transcribe before you spend anything on it, or test your microphone before the interview starts. Everything is measured in your browser and nothing is uploaded.
- Checks a recording before you pay to transcribe it
- Tests your microphone before a call starts
- Noise, loudness and clipping, explained
- Measured in the browser, nothing uploaded
Pick the microphone you will record with. The check takes about eight seconds: a moment of silence to measure the room, then one sentence at your normal volume.
Pick the microphone you will use in the session and stand where you will speak. The check takes ten seconds: three seconds of silence to measure the room, then one sentence read aloud at your normal volume.
Welcome, everyone. I will speak at my normal pace, and my words will appear on your screen while I am still saying them.
Measured with the browser’s noise suppression and automatic gain switched off, so you see what the microphone itself picks up. Nothing is recorded or saved.
Measured the way Speecho listens in a live session: one channel, with the browser’s echo cancellation and noise suppression on. Nothing is recorded or saved.
What decides whether a transcript comes out right
Transcription models cope well with accents, fast speech and jargon. What they cannot recover is a word that never made it into the file cleanly: speech buried under a noisy room, a voice too far from the microphone, or a peak so loud the waveform was cut flat. Those are the three things this tool measures, because they are what separates a clean transcript from one full of gaps.
The numbers are the ones an audio engineer would look at. Signal-to-noise ratio is the distance between the voice and everything else; clipping counts the moments the level hit the ceiling. For how that translates into accuracy, see how accurate AI transcription really is. If the file passes, transcribe it. If long pauses are the problem, the silence remover shortens them first, and the voice recorder is there for a better take.
Audio quality checker - FAQ
Not covered here? Ask us, we usually reply within a business day.
Is my recording uploaded?
No. The file is decoded and measured inside this tab with your browser’s own audio engine. Nothing is sent anywhere.
What is a good signal-to-noise ratio for transcription?
25 dB or more is comfortable: the voice stands clearly apart from the room. Between 15 and 25 dB transcription usually works, but quiet words, names and numbers go first. Under 15 dB, expect gaps. These thresholds are our rule of thumb for speech, not a published standard.
Why does it call my file quiet when it sounds fine?
Your player probably turns it up. The measurement is in dBFS, relative to the loudest level a file can hold, and speech around -25 to -15 dBFS is typical of a well-set recording. Quiet audio is only a real problem when the noise is close behind it, which is why the verdict leans on signal-to-noise more than on loudness.
What does clipping mean?
The level went above what the file can store, so the tops of the waveform were cut flat. You hear it as crackle or distortion on loud syllables. It cannot be repaired afterwards; the fix is lower input gain on the next recording.
Does the microphone test record me?
No. It listens for eight to ten seconds to measure the levels and then releases the microphone. Nothing is saved or uploaded.
Which files can it read?
Anything your browser can decode: MP3, M4A, WAV, OGG and the audio track of most MP4 and MOV files. Up to three hours, since the whole file is held in memory.
Is my microphone good enough for live captions or live translation?
Open “Test my microphone”, choose “Live captions or interpretation” and run the ten-second check in the room where you will speak. It listens the way a Speecho live session does and tells you whether your voice stands far enough above the room: 25 dB or more is good, under 15 dB is risky. Those numbers rest on a test we ran on 4 October 2026 with synthetic voices and added noise, passed through the browser and streamed to the live recogniser: it got every word right from 16 dB up, began to get single words wrong below 15 dB, and lost whole phrases once the voice no longer stood out from the noise. Real voices and echoing rooms are harder than synthetic ones, which is why “good” starts well above that point. Treat it as a guide, not a guarantee.
Why does the live check give different numbers from the recording check?
Because the two listen differently, on purpose. The recording check switches the browser’s noise suppression and automatic gain off, to show what the microphone and the room put into a file. A live session has no second pass, so Speecho opens the microphone with echo cancellation and noise suppression on, and the live check measures that processed sound. Steady noise such as a fan is largely filtered out, so the same room usually scores a higher signal-to-noise ratio in the live check. Other people talking are not filtered out.