Speaker identification
Attach diarized speaker turns to known identities using enrolled voiceprints, for workflows where who spoke matters as much as what was said.
Speaker identification maps an unknown speaker label to a known identity. It builds on diarization and requires application-specific enrollment or reference context.
Typical flow
- Diarize the audio into anonymous speaker turns.
- Compare eligible turns with the reference identities available to the workload.
- Return an identity match and confidence information where the configured flow supports it.
- Let the application decide how to handle uncertain or unmatched speakers.
Product and privacy considerations
Voice identity can be sensitive. Collect and use reference audio only with the appropriate authority and disclosure for your jurisdiction and use case. Keep raw enrollment audio, derived representations, and match output within a deliberate retention and access policy.
Do not force a match
An unknown result is safer than assigning the wrong identity. Product logic should preserve anonymous speaker labels when a match is not sufficiently reliable.
Speaker diarization
Separate a recording into speaker turns and align speaker labels with the transcript, either on its own or as one stage of a pipeline.
Transcript formatting
Turn raw ASR output into readable transcripts: punctuation, casing, numbers, and paragraphing that stay stable for people and downstream systems.