29 August 2026
Upload the file to a transcription service built for it — most run on Whisper or a comparable model and return text in minutes to an hour, depending on length. What you get back depends far more on how the audio was captured than on which service you pick: a single microphone in a room with five people will always transcribe worse and label speakers worse than a call recorded with one channel per participant.
Speech recognition models are good now. The variable that ruins a transcript almost every time is the room, not the model. A phone propped on a conference table picks up the person nearest it clearly and everyone else as mush — overlapping voices, HVAC hum, someone typing. That same conversation recorded through a video call, where each participant has their own microphone and their own audio channel, comes out close to word-perfect because the model never has to separate two voices from one waveform.
A transcript with no names attached — just a wall of text — is only half useful. Speaker labels come from diarization, a separate step that clusters segments of audio by voice and assigns them Speaker 1, Speaker 2, and so on before a human (or the tool, from context) swaps in real names.
Diarization works from acoustic differences between voices, so it needs those differences to be audible. On a single-mic recording where two people have similar pitch and both come through the same channel at similar volume, the model will genuinely confuse them — this is the same problem covered in our piece on speaker diarization, and it does not go away just because the transcription itself was accurate. On a per-participant recording, labeling is close to solved: each channel already tells you who spoke.
AVAY is a video meeting platform that transcribes the call itself — no bot joins, because there is nothing to join. Start one at avay.ai, read how each part works in the documentation, or see what it costs.