Because we record in dual channel, the rep and the prospect are already on separate audio tracks before transcription starts.
Twilio records dual channel. Channel one is the rep's leg, channel two is the dialed leg. AssemblyAI's multichannel model transcribes each independently. There is no inference step to get wrong.
Every line in the transcript is timestamped. Click and the audio jumps there.
When a transcription fails — a provider hiccup, a quota limit — a sweeper retries it automatically, up to three attempts, waiting long enough that a live call is never transcribed twice.
Diarization guesses who spoke. Dual channel knows. On a call where both people talk over each other, that's the difference between a usable transcript and a confusing one.
English today.
It retries automatically.
Explore
