Transsion Multilingual Speech Recognition System for MLC-SLM 2025 Challenge
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866916910555201536 |
|---|---|
| author | Li, Xiaoxiao Zhu, An Jiang, Youhai Zhu, Fengjie |
| author_facet | Li, Xiaoxiao Zhu, An Jiang, Youhai Zhu, Fengjie |
| contents | This paper presents the architecture and performance of a novel Multilingual Automatic Speech Recognition (ASR) system developed by the Transsion Speech Team for Track 1 of the MLC-SLM 2025 Challenge. The proposed system comprises three key components: 1) a frozen Whisper-large-v3 based speech encoder, leveraging large-scale pretraining to ensure robust acoustic feature extraction; 2) a trainable adaptor module using Linear-ReLU-Linear transformation mechanisms to effectively align speech and text representations; and 3) a frozen Qwen2.5-7B-Instruct large language model (LLM) integrated with trainable LoRA for optimized contextual linguistic decoding. By systematically combining pretrained models with task specific fine-tuning, the system achieved a word/character error rate (WER/CER) of 9.83% across 11 languages in the evaluation set and ranked third place among global participants. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2508_14916 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Transsion Multilingual Speech Recognition System for MLC-SLM 2025 Challenge Li, Xiaoxiao Zhu, An Jiang, Youhai Zhu, Fengjie Audio and Speech Processing Artificial Intelligence Computation and Language This paper presents the architecture and performance of a novel Multilingual Automatic Speech Recognition (ASR) system developed by the Transsion Speech Team for Track 1 of the MLC-SLM 2025 Challenge. The proposed system comprises three key components: 1) a frozen Whisper-large-v3 based speech encoder, leveraging large-scale pretraining to ensure robust acoustic feature extraction; 2) a trainable adaptor module using Linear-ReLU-Linear transformation mechanisms to effectively align speech and text representations; and 3) a frozen Qwen2.5-7B-Instruct large language model (LLM) integrated with trainable LoRA for optimized contextual linguistic decoding. By systematically combining pretrained models with task specific fine-tuning, the system achieved a word/character error rate (WER/CER) of 9.83% across 11 languages in the evaluation set and ranked third place among global participants. |
| title | Transsion Multilingual Speech Recognition System for MLC-SLM 2025 Challenge |
| topic | Audio and Speech Processing Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2508.14916 |