Speaker identification

When transcript of some audio or video file is generated you can identify speakers in the transcript. Double-click the transcript in the list of transcripts to see additional options for editing and exporting.

Transcription view screen has an option to identify speakers. Click on the "Identify speakers" button to see the available options. With the MSDD model, you can leave the number of speakers on Auto or select the known number of speakers. Providing the known number can help when automatic detection merges similar or quieter voices. Sortformer detects the number automatically.

If the audio file is still present on the system, speaker identification will mark each speaker's sentences with an appropriate label. You can preview 10 seconds of a random sentence from an identified speaker and rename the automatically identified label to the speaker's real name. If the "Merge speaker sentences" checkbox is selected when you save the speaker labels, all consecutive sentences from the same speaker will be merged into one segment. Speaker identification is not available on Intel macOS.

Setting an exact count guarantees that MSDD creates that many speaker clusters, but it cannot guarantee that every sentence is assigned to the correct person. Recordings with overlapping speech, large differences in microphone distance, or very little speech from one participant can still require manual correction.