liva-ai/couples-call-samples

Description of Data Source

An app that helps couples connect through interactive games and guided conversations. Partners can answer questions together and explore their relationship in a fun and engaging way. Audio is recorded separately on each device, producing time-aligned, channel-separated tracks.

Technical Analysis

Measure	Value
Sample rate	48 kHz
Bit depth	24-bit PCM
File format	WAV (stereo, one person per channel)
Processing	None: no noise suppression, echo cancellation or automatic gain
Full-band audio	82 % of hours carry speech above 20 kHz (the rest are headset mics)
Average call duration	17.3 min (median 11.8 min) for two-person calls

These metrics vary from call to call. Contact us for a breakdown at the quality level you need. Pricing and collection rate depend on that level.

Measure	Range
SNR	about 5-27 dB (median 17 dB)
Speech level (RMS)	about -40 to -23 dBFS
Speech ratio	median 0.36 per conversation
Audio bleed	present in about 80 % of hours of two-person calls
Spectral centroid	about 0.4-1.7 kHz

SNR (signal-to-noise ratio) is how much louder the speech is than the background. No noise suppression is applied, so the background is the natural sound of each person's room.

Audio Bleed

Audio bleed occurs when a participant’s microphone picks up the other person’s voice. Most of this dataset contains bleed because many users don’t wear headphones. Audio is recorded losslessly without noise suppression. Some researchers intentionally keep the bleed, others process the audio themselves, and others prefer to purchase only recordings without bleed. Pricing and collection rates depend on your requirements. Contact us for details.

Without audio bleed

Sample 1

Sample 2

With audio bleed (the majority of the dataset)

Sample 1

Sample 2

Languages

Distribution of conversation hours by language.

Language	Share of hours
English	84.2 %
Spanish	4.8 %
German	2.8 %
Finnish	1.8 %
Italian	1.4 %
Filipino	1.2 %
Afrikaans	1.1 %
Swedish	1.0 %
Portuguese	0.8 %
Unknown	0.9 %

Speakers per Conversation

Distribution of conversation hours by number of speakers.

Speakers	Share of hours
2	100 %

Conversation Categories

Distribution of two-person conversation hours by category. Categories were defined and conversations classified from transcripts using an LLM.

Questions

Question category	Share of hours
Deeper emotional questions	5.8 %
Shared memories	2.1 %
Spicy / intimate topics	1.5 %
Relationship boundaries	1.2 %
Future plans	0.7 %
Funny hypotheticals	0.4 %

Games

Game	Share of hours
Draw & Guess	15.8 %
Hot Take	9.3 %
Mini Golf	7.5 %
Same Wavelength	7.1 %
Word games	7.1 %
Connect 4	6.8 %
Who's More Likely	6.6 %
Cooperative puzzles / mysteries	2.3 %
Other games	2.0 %

Other

Activity	Share of hours
Free conversation	16.9 %
Setup / technical	6.9 %

Accents

Distribution of English conversation hours by estimated accent. At least one participant in each conversation has the specified accent. Accents are inferred from IP addresses.

Accent	Share of hours
General American	27.5 %
Southern American English	22.0 %
Western European influenced English	8.6 %
Unknown	5.6 %
Australian English	4.1 %
Southern European influenced English	4.1 %
Southern British English	4.1 %
Canadian English	3.9 %
Northern British English	3.6 %
East Asian influenced English	1.8 %
Middle Eastern influenced English	0.9 %
Maritime Southeast Asian English	0.6 %
Irish English	0.5 %
Southern African English	0.4 %
Latin American influenced English	0.4 %
Caribbean English	0.4 %
Mainland Southeast Asian influenced English	0.4 %
South Asian English	0.3 %
West African English	0.1 %