What type of study is this?

This is a Quantitative Study study.

October 20, 2025Open Access

Qwen vs. Gemma Integration with Whisper: A Comparative Study in Multilingual SpeechLLM Systems

Key Points

The study finds that the integration of Whisper with LLMs allows for effective multilingual speech recognition.
A competitive average WER of 16.63% was achieved using Gemma3-12B, while Qwen2.5-7B resulted in 18.6%.
Utilizing a three-stage training methodology, the study progressively optimizes key components for better performance.
This approach suggests that efficient projector architectures significantly enhance language modeling in LLM applications.

Abstract

This paper presents our system for the MLC-SLM Challenge 2025, focusing on multilingual speech recognition and language modeling with large language models (LLMs). Our approach combines a fine-tuned Whisper-large-v3 encoder with efficient projector architectures and various decoder configurations. We employ a three-stage training methodology that progressively optimizes the encoder, projector, and LLM components. Our system achieves competitive performance with a private test average WER/CER result of 16.63% using the Gemma3-12B and 18.6% using the Qwen2.5-7B as decoder-only language model.

Read Full Paperexternally

KI fragen

Bookmark

View Full Paper