Speaker-Aware Simulation Improves Conversational Speech Recognition
Máté Gedeon, Péter Mihajlik · 2026
Read Speaker-Aware Simulation Improves Conversational Speech Recognition by Máté Gedeon, Péter Mihajlik (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-04776.txt.
Description
Hungarian conversational speech is the testbed for Máté Gedeon and Péter Mihajlik in a 2026 study of speech recognition. Conversation data with many speakers is scarce for lower-resource languages, so the authors simulate dialogues from single-speaker recordings.
Their speaker-aware simulation stitches recordings into realistic multi-speaker exchanges. A variant also models pauses based on utterance length, capturing how timing works in real talk. Fake Hungarian dialogues were made from a large speech corpus and mixed with real conversation for training. Across many settings, speaker-aware data beat naive concatenation. The pause variant gave modest, systematic gains, strongest when the source conversation statistics matched the target domain.
Catalog record
| CREATOR | Máté Gedeon, Péter Mihajlik |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 50 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-04776