DhawalM/mira-invocation-audio

Mira invocation audio An exact snapshot of the active audio dataset used by the Mira Whisper and HuBERT invocation classifiers. It includes 167 derivatives from one real microphone training session. The label is 1 when the glasses wearer directly addresses the Mira assistant, and 0 for incidental mentions, other conversation, or background audio. Only the wearer's direct address should activate Mira. Synthetic positives assume the synthetic speaker is the wearer; the classifiers… See the full description on the dataset page: https://huggingface.co/datasets/DhawalM/mira-invocation-audio.