I have an idea for an automated push to talk feature, leveraging the "biometric speech identification" models I've seen floating around.
Honestly I could probably implement this myself. Grab a high quality voice embedding pipeline, have a "enrollment" phrase (like Sphinx of Black Quartz...) and grab vectors for the speaker, then continuously check audio chunks for those vectors.
I have an idea for an automated push to talk feature, leveraging the "biometric speech identification" models I've seen floating around.
Honestly I could probably implement this myself. Grab a high quality voice embedding pipeline, have a "enrollment" phrase (like Sphinx of Black Quartz...) and grab vectors for the speaker, then continuously check audio chunks for those vectors.