Qwen vs. Gemma Integration with Whisper: A Comparative Study in Multilingual SpeechLLM Systems
Tuan Nguyen, Long-Vu Hoang, Huy-Dat Tran · 2025
Read Qwen vs. Gemma Integration with Whisper: A Comparative Study in Multilingual SpeechLLM Systems by Tuan Nguyen, Long-Vu Hoang, Huy-Dat Tran (2025) 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-2506-13596.txt.
Description
Tuan Nguyen, Long-Vu Hoang and Huy-Dat Tran compare two language models as decoders in multilingual speech systems, in their 2025 arXiv paper. Their entry to a 2025 challenge pairs a fine-tuned Whisper encoder with projector layers and a decoder-only language model. Training proceeds in three stages, tuning the encoder, the projector and the language model in turn.
Two decoders were tried in the same frame: a twelve-billion-parameter Gemma model and a smaller Qwen model. The Gemma system scored better on the private test set, with lower average error rates across languages. The comparison guides decoder choice for speech language models.
Catalog record
| CREATOR | Tuan Nguyen, Long-Vu Hoang, Huy-Dat Tran |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 16 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-13596