Audio file inspector
See what a clip actually contains before you do anything with it — how long it is, how much of it survived compression, and whether it is worth submitting at all.
This runs entirely on your device. The page makes no network requests. Your browser reads the file with the File API and decodes it with the Web Audio API; nothing is uploaded, nothing is stored, and closing the tab clears it. This is not a detector — it measures the container, not the speech. Detection lives at /check/.
Decoding…
What each number tells you
Duration. The single most useful figure. Ten to thirty seconds of one person speaking gives the best result. Below about five seconds, an analysis will usually return unclear rather than guess, because there is not enough continuous speech to measure anything against.
Sample rate. The ceiling on what frequencies the file can contain — a recording can only represent frequencies up to half its sample rate. Narrowband telephone audio is sampled at 8 kHz, which means everything above roughly 4 kHz is simply not present. Much of the evidence of synthesis sits in that discarded region, which is why phone audio is the hard case.
Channels. Mono is normal for voice and is not a problem. Stereo sometimes means the file was produced or edited rather than captured from a call, which is worth knowing but proves nothing on its own.
Size, and the average bitrate derived from it. A small file holding a long recording has thrown a great deal away. The lower the average bitrate, the more aggressively the audio has been compressed, and compression removes exactly the fine detail a detector reads. This is why a forwarded copy so often performs worse than the original.
A caveat about sample rate
Browsers resample audio while decoding it, so the decoded sample rate usually reflects your own device’s audio output rate rather than what is stored in the file. This tool reports the decoded value honestly labelled as such. For WAV files it additionally reads the rate, channel count and bit depth straight out of the file header, which is the true stored value. For compressed formats, treat the decoded figure as a property of playback, not of the file.
Whatever the numbers say, do not convert or re-encode the file to “improve” it. Every re-encode discards more. Getting a usable recording explains the handling that preserves evidence.
Questions people ask
Is my file uploaded anywhere?
No. There are no network calls on this page at all. The file is read and decoded by your browser and never leaves the device.
Can this tell me whether the voice is real?
No, and it never will. It reads the file’s properties. Deciding whether the speech was generated is a different question and is answered at /check/.
My file will not decode. Is it broken?
Not necessarily. Browsers vary in which codecs they can decode. A file this page cannot open may still be analysable, so try submitting it anyway.
The duration looks longer than the speech in it.
Silence, hold music and ring tone all count towards duration but contribute nothing usable. What matters is the longest continuous stretch of one person speaking.
Reviewed