Recognize people by the sound of their voice.
Enroll someone with a short recording. Later, add any clip and VoiceID tells you who is speaking and how sure it is.
What happens to your audio
Prepare
The recording is converted to one channel at 16,000 samples per second.
Find speech
Silence and pauses are cut out so only the voice is kept.
Check clarity
We measure how much louder the voice is than the background noise.
Make a voiceprint
An AI model turns the voice into 192 numbers that describe how it sounds.
Enroll a voice
Give the person a name, then add a recording of them speaking. Any length works, but 10 to 30 seconds of clear speech gives the most reliable voiceprint.
Who is speaking?
Enrolling the same name again adds another sample for that person.
Add their voice
Identify a speaker
Add a recording and VoiceID compares it with everyone who is enrolled. You get a similarity score from 0 to 1 for each person.
Add a recording
Choose how strict matching is
Higher is stricter. If a person you know is enrolled keeps landing in yellow, lower it a little. The result below updates as you move the slider.
Batch
Add many recordings at once. VoiceID processes them one after another, and you can download the results as a CSV file.
What should happen to each file?
Add recordings
People
Everyone VoiceID can recognize. Each row is one enrolled recording. Voiceprints are saved in data/voiceprints.json on the server.