WAV is what you get when the equipment does not compress. Handheld field recorders such as the Zoom H-series and Tascam units write WAV by default, studio software exports it, courtroom and dictation systems often produce it, and many researchers record interviews on a device that only knows this one format. The files are heavy, but the audio inside is exactly what the microphone heard, with no encoder in between.
That makes WAV pleasant to transcribe. There are no compression artifacts, the frequency range is intact and quiet passages are not smeared the way they can be in a low-bitrate MP3. If the recording was made with a reasonable microphone at a reasonable distance, the transcript will reflect that.
Transcrify reads the file as uploaded. There is no need to convert it to MP3 to make it smaller, and doing so would only throw information away. Mono and stereo files, 16-bit and 24-bit depth, sample rates from 8 kHz telephone quality up to 96 kHz are all handled.
An hour of stereo WAV at 16-bit 44.1 kHz is a little over 600 MB. Three hours is close to two gigabytes. Many browser-based tools stop at a fraction of that, which is why people with recorders end up searching for a converter before they can even start. Here the upload is divided into one-megabyte chunks sent one after another, so the size of the file is not what limits you.
What limits a guest upload is length: up to 3 hours per recording. That covers a full deposition, a long focus group or an entire rehearsal in a single file. If you record at 96 kHz or in 24-bit, the file grows but the limit stays the same, because the limit is measured in time, not in bytes.
A practical note for future recordings: speech does not need high sample rates. 16-bit at 44.1 kHz, or even 48 kHz mono, captures everything a transcription model can use. Recording in 24-bit at 96 kHz is fine for music, but for an interview it only means a slower upload.
Every segment in the transcript carries a timestamp and a speaker label. With a recorder placed between two people on a table, the two voices are usually separated reliably. In a focus group with six participants and overlapping talk, the model does its best, and the passages where people speak over each other are worth a listen afterwards. The timestamps make that quick: the text tells you where to put the playhead.
More than fifty languages are recognised automatically, so a field recording that switches between English and the local language needs no manual setting. A typical hour of audio is transcribed in three to five minutes once the upload is complete.
WAV files from recorders are often the raw material of something confidential: a legal proceeding, a clinical interview, an unreleased recording session. The speech models run on hardware we operate, and the audio is not passed on to a third-party speech API. Guest transcriptions are kept under the same rules and move into your account if you later create one.
As a guest you can read the transcript and copy it. A free account adds the editor, your transcription history and downloads in SRT, VTT, ASS and TXT, with 5 hours across 10 files. AI chat with the recording, summaries and export to PDF or Word are part of the paid plans.