liva-ai/yapdo-convo

Yapdo Sample Data

This dataset card shows details about the Yapdo conversational speech corpus by Liva AI (YC S25). This dataset card details information for 150,000+ hours of recordings from 8,000+ speakers across 17 languages. The source audio is natively recorded with separate speaker channels; the samples here are presented as combined conversations.

The strength of this dataset is its naturalness. Recorded among friends "in-the-wild," it preserves the spontaneity of real dialogue and supports the development of better conversational AI. It also reflects friendly interactions across cultures and captures realistic turn-taking dynamics, which are essential for training models that sound natural. We note one limitation is in the acoustic quality; however, a potential solution during collaboration is to collect pairwise data with better/studio-quality equipment to train a model that can enhance the data.

Dataset Configs

Each config maps to a single directory of audio files with an accompanying metadata.jsonl.

Group	Configs	Description
Languages	languages_ar, languages_arz, languages_bn, languages_es, languages_fr, languages_gu, languages_ha, languages_hi, languages_ig, languages_it, languages_kn, languages_pa, languages_pcm, languages_si, languages_sw, languages_ta, languages_te, languages_tl, languages_ur, languages_yo	A subset of our multilingual speech organized by language code. Metadata: nisqa
Accents	accents_Arabic_influenced_English, accents_East_African_English, accents_Filipino_English, accents_General_American, accents_Indian_English, accents_Nigerian_English, accents_South_African_English, accents_West_African_English	Accent-stratified English speech across 8 accent groups. Metadata: nisqa
Full-duplex	full_duplex_backchanneling, full_duplex_interruption_detection, full_duplex_pause_handling, full_duplex_turn_taking	Recordings with full-duplex benchmark characteristics (backchanneling, interruption, pause handling, turn-taking)
Paralinguistics	paralinguistics	Paralinguistic elements (laughing, singing, elongated words, etc.). For the sake of displaying these elements, we identified emotion-rich segments using Empathic-Insight-Voice-Small since we observed that emotion-rich segments were likely to contain such elements. Metadata: emotion, peak_score, avg_mos
Random 50	random_50	50 completely random samples from our QA'd pool. Metadata: nisqa, language, accent, speech_ratio
Gaming	gaming_assistant, gaming_companion	Gaming conversations: helper-type conversations vs. companion. Metadata: game, summary

Dataset Overview

Total audio	150,000+ hours
Unique conversations	31,822
Unique speakers	9,779
Languages	~17
Speakers per conversation	2–13 (avg 2.7)
Conversation duration	19s – 24.4 hrs (avg ~60 min)
Code-switching	~25% of conversations
Speech type	Spontaneous, unscripted, multi-party conversations
Quality score (NISQA)	1.8 - 4.8 (avg 2.5)
Common topics	Video games, daily life (jobs, school, relationships, earning money)

---

Languages

Language labels for each conversation were reviewed by a native human speaker.

Monolingual Conversations

17 languages with over 10 hours of monolingual conversation data. The below shows the approximate distribution of conversations and hours across languages.

Language	Code	Conversations (%)	Hours (%)
English	en	66.5%	65.5%
Egyptian Arabic	arz	7.5%	8.2%
Spanish	es	4.8%	7.7%
Swahili	sw	6.2%	4.6%
Nigerian Pidgin	pcm	3.6%	3.9%
Hindi	hi	3.8%	2.6%
Arabic	ar	2.6%	2.4%
Tagalog	tl	0.7%	1.3%
Tamil	ta	0.6%	0.9%
Hausa	ha	1.0%	0.8%
Yoruba	yo	0.9%	0.8%
Italian	it	1.2%	0.8%
French	fr	0.1%	0.2%
Igbo	ig	0.2%	0.1%
Telugu	te	0.1%	0.1%
Cebuano	ceb	0.1%	0.1%
Kannada	kn	0.1%	0.1%

Code-Switching Conversations

28 language combinations with over 10 hours of code-switching data, spanning roughly 25% of all conversations. The below shows the approximate distribution of conversations and hours across language groups.

Language Group	Conversations (%)	Hours (%)
English + Nigerian Pidgin	51.9%	49.3%
English + Tagalog	17.9%	21.2%
Cebuano + English + Tagalog	8.6%	10.1%
English + Swahili	5.4%	4.8%
English + Yoruba	3.0%	2.4%
English + Hausa	2.2%	2.3%
English + Nigerian Pidgin + Yoruba	1.1%	1.2%
Arabic + Egyptian Arabic	1.1%	0.9%
English + Hindi	1.8%	0.9%
Hausa + Swahili	0.8%	0.8%
English + Hiligaynon + Tagalog	0.6%	0.8%
English + Hausa + Swahili	0.5%	0.7%
English + Tamil	0.7%	0.6%
Nigerian Pidgin + Yoruba	0.8%	0.6%
Hindi + Urdu	0.5%	0.5%
English + Igbo + Nigerian Pidgin	0.2%	0.4%
English + Hausa + Nigerian Pidgin	0.3%	0.3%
English + Spanish	0.3%	0.3%
English + Igbo	0.3%	0.3%
Arabic + English	0.3%	0.3%
Egyptian Arabic + English	0.3%	0.3%
English + Telugu	0.3%	0.2%
English + Nigerian Pidgin + Swahili	0.2%	0.2%
Igbo + Nigerian Pidgin	0.2%	0.2%
Cebuano + English + Hiligaynon + Tagalog	0.1%	0.2%
Nigerian Pidgin + Swahili	0.2%	0.2%
Hausa + Nigerian Pidgin	0.1%	0.1%
English + Hindi + Urdu	0.1%	0.1%

---

Accents

These labels are for English only and were obtained from the city that participants self-reported they were from. The below shows the approximate distribution of hours across accent groups.

Accent Group	Hours (%)
Nigerian English	80.5%
Indian English	5.2%
East African English	4.6%
General American	3.9%
Filipino English	2.8%
West African English	1.8%
Arabic-influenced English	0.7%
South African English	0.5%

---

Labels

Language labels were assigned at the speaker-track level by native speakers who reviewed each individual track within a conversation. A single conversation may carry multiple language labels when speakers use different languages. Accent labels are derived from each speaker's self-reported city of origin, providing a natural geographic proxy for dialect and accent variation.

---

Technical Analysis

Sample rate	48 kHz
Bit depth	16-bit PCM
File format	WAV
Mean SNR	~33 dB
Median RMS	-26 dBFS
Average speech ratio	0.35
Spectral centroid	~0.66 kHz
Frequency content	3.3 kHz (averaged over 10–30 second clips)

Combined vs. Separated Audio

Each sample in this dataset is a combined mix of all speakers. The parent Yapdo corpus stores each speaker on a separate, time-aligned track. Here's what that difference sounds like — a Telugu conversation with 2 speakers:

Combined (all speakers mixed)

Speaker 1 (isolated track)

Speaker 2 (isolated track)

---

Audio Artifacts

Source audio passes through Opus VoIP pipeline.

Artifact	Prevalence
Dropouts / packet loss	98.6%
Bandwidth ceiling (< 4 kHz)	97.2%
Clicks / pops	93.6%
Mains hum (50/60 Hz)	82.4%
Silence / dead air	34.6%
Frame repetition	18.2%
Echo	15.2%
Low signal level	5.8%
Onset transients	5.2%
Clipping	0.6%

---