Humod is a model-human conversational dataset
Inspiration behind this dataset: humans interact differently with models than they do with other humans, and human-human interactions don't capture that
We get a full-duplex model (gemini-2.5-flash) to speak to the human
Mostly American accent for now (although can expand)
Each sample includes metadata: age, current country, country grew up, race, ethnicity, and emotional state

Technical Analysis

	Value
Avg session duration	Just over 1 min
Sample rate	24 kHz
Bit depth	16-bit PCM
File format	WAV
Mean SNR	~27 dB
Median RMS	-24 dBFS
Average speech ratio	0.31
Spectral centroid	~1.98 kHz
Frequency content	11.3 kHz (averaged over 10-second clips)