Multilingual Conversations

Data description

Multilingual Conversations is a dataset of unscripted, two-speaker conversations, each held in the pair’s shared native or fluent language. There are no prompts or topics — people talk as they naturally would, so the audio keeps the prosody, disfluencies, and overlap of real conversation. It spans European and Scandinavian languages alongside a long tail of lower-resource languages.

Each speaker’s microphone is captured locally as uncompressed 16-bit PCM and delivered as a separate stem, so every recording arrives with its native dynamics intact and a BCP-47 language code attached. It works as a low-resource-language training set, a cross-lingual prosody corpus, and a multilingual evaluation benchmark for ASR and conversational TTS.