WAV files are enormous because they are uncompressed. This transcribes them without uploading a single gigabyte: the recording is read in pieces on your own machine.
Start with the arithmetic that makes this page necessary. WAV stores sound uncompressed, so the file size is roughly the sample rate times the bit depth times the number of channels times the duration. In round numbers, an hour of CD-quality stereo works out to about a gigabyte, and a long field session or a multi-hour hearing can run to several. That is the exact number an "upload your WAV" service asks you to push across your internet connection before it will transcribe a single word, and then it still meters your minutes at the far end.
Here that whole upload never happens, so there is no gigabyte wait to sit through. The WAV is opened on your disk and read in small slices, which means text starts appearing within seconds of dropping the file, no matter how large the file is. Your connection speed stops being part of the equation entirely.
WAV is the working format of people who record for a living or for the record. Hardware field recorders (the handheld units documentary crews, sound designers, and wildlife recordists carry) write WAV by default, precisely because it keeps every bit of the capture. Studio and podcast session masters live as WAV too: the untouched multitrack bounce you archive before anything gets compressed for release. Court reporters and legal dictation systems produce WAV for evidentiary reasons, since an uncompressed file is a faithful record. Call centres and telephony systems export WAV as well, though usually the narrow-band 8 kHz kind, and those transcribe, just with lower accuracy, which is worth setting your expectations around. And oral-history collections, where archivists digitise interviews to preserve them, are almost always stored as WAV.
The common thread is that none of these people generated a WAV so it could be convenient to upload; they generated it to be an accurate, complete recording. Transcribing it should not force them to hand that recording to a third party.
The instinct with a two-gigabyte file is to expect the browser to seize up trying to hold it all in memory at once. That is not what happens, because the WAV is never loaded whole. It is read as a stream of small chunks straight from disk: a chunk comes in, gets converted to what the recogniser needs, is transcribed, and is released before the next chunk arrives. Memory use stays flat whether the file is thirty megabytes or three gigabytes: the size on disk simply never becomes the size in memory.
Progress is reported by seconds of speech processed rather than by how much of the file has been read, which keeps the estimate honest on a long recording with quiet passages. This is a size story, not a duration story: the point is that the raw heft of an uncompressed file (the thing that makes WAV painful to upload) is exactly the thing this streaming approach neutralises.
.wav in, however big it is. No pre-conversion, no splitting the file into parts first.Working with a different file type? Use the full audio transcriber, or jump to a format-specific guide: