What is speaker diarisation?
Speaker diarisation is the process of working out who spoke when in an audio recording, so that each part of the transcript is labelled with the right speaker.
Why it matters
A transcript without speaker labels is hard to read and close to useless for analysis. Reviewers need to know whether the agent or the customer said a phrase. Measures such as talk time per speaker, interruptions and customer sentiment all depend on separating the voices correctly. Summaries are also more accurate when the system knows who made a promise and who made a complaint.
How it works
The system splits the audio into short segments and analyses the characteristics of the voice in each one. Segments that sound like the same person are grouped and given a label such as speaker one or speaker two. When each party is recorded on a separate channel, the task is simple. On a single mixed channel, overlapping speech and similar voices make it harder.
Example
A recorded call between an agent and a customer about a late delivery is saved as one mixed audio track. Diarisation labels each line of the transcript as agent or customer. A quality reviewer can then see that the customer raised the delay three times before the agent acknowledged it, and that the agent spoke for most of the call.
