# Direct Preference Optimization for English-Mandarin Code-Switching Speech Recognition in Audio LLMs

arXiv: https://arxiv.org/abs/2605.23975 · HTML: https://mlsanity.com/paper/2605.23975/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Placed in Preference alignment for recommenders and RAG, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 23 of its 28 references that are in the map.
- First posted on arXiv 13 May 2026.

## Placed in

- LLM alignment / RLHF / preference optimisation / [Preference alignment for recommenders and RAG](https://mlsanity.com/alignment/preference-alignment-for-recommenders-and-rag/)

## Builds on (most cited first, 23 of 23)

- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Robust Speech Recognition via Large-Scale Weak Supervision](https://mlsanity.com/paper/2212.04356/) (2022)
- [Common Voice: A Massively-Multilingual Speech Corpus](https://mlsanity.com/paper/1912.06670/) (2019)
- [SimPO: Simple Preference Optimization with a Reference-Free Reward](https://mlsanity.com/paper/2405.14734/) (2024)
- [Qwen2.5-Omni Technical Report](https://mlsanity.com/paper/2503.20215/) (2025)
- [Qwen2-Audio Technical Report](https://mlsanity.com/paper/2407.10759/) (2024)
- [Phi-4-Mini Technical Report: Compact yet Powerful Multimodal Language Models via Mixture-of-LoRAs](https://mlsanity.com/paper/2503.01743/) (2025)
- [SALMONN: Towards Generic Hearing Abilities for Large Language Models](https://mlsanity.com/paper/2310.13289/) (2023)
- [Qwen3-Omni Technical Report](https://mlsanity.com/paper/2509.17765/) (2025)
- [FLEURS: Few-shot Learning Evaluation of Universal Representations of Speech](https://mlsanity.com/paper/2205.12446/) (2022)
- [Emilia: An Extensive, Multilingual, and Diverse Speech Dataset for Large-Scale Speech Generation](https://mlsanity.com/paper/2407.05361/) (2024)
- [mDPO: Conditional Preference Optimization for Multimodal Large Language Models](https://mlsanity.com/paper/2406.11839/) (2024)
- [SpeechAlign: Aligning Speech Generation to Human Preferences](https://mlsanity.com/paper/2404.05600/) (2024)
- [On the End-to-End Solution to Mandarin-English Code-switching Speech Recognition](https://mlsanity.com/paper/1811.00241/) (2018)
- [Reducing Language confusion for Code-switching Speech Recognition with Token-level Language Diarization](https://mlsanity.com/paper/2210.14567/) (2022)
- [Enhancing Code-switching Speech Recognition with Interactive Language Biases](https://mlsanity.com/paper/2309.16953/) (2023)
- [Speech collage: code-switched audio generation by collaging monolingual corpora](https://mlsanity.com/paper/2309.15674/) (2023)
- [CS-Dialogue: A 104-Hour Dataset of Spontaneous Mandarin-English Code-Switching Dialogues for Speech Recognition](https://mlsanity.com/paper/2502.18913/) (2025)
- [Boosting Code-Switching ASR with Mixture of Experts Enhanced Speech-Conditioned LLM](https://mlsanity.com/paper/2409.15905/) (2024)
- [Adapting Whisper for Code-Switching through Encoding Refining and Language-Aware Decoding](https://mlsanity.com/paper/2412.16507/) (2024)
- [SC-MoE: Switch Conformer Mixture of Experts for Unified Streaming and Non-streaming Code-Switching ASR](https://mlsanity.com/paper/2406.18021/) (2024)
- [Can we train ASR systems on Code-switch without real code-switch data? Case study for Singapore's languages](https://mlsanity.com/paper/2506.14177/) (2025)

## Built on by (most cited first, 2 of 2)

- [Efficiently Adapting Spoken Language Models for the Singaporean Context](https://mlsanity.com/paper/2607.10092/) (2026)
- [DiaWhisper-DPO: Role-Attributed Transcription of Clinical Interviews via Failure-Mined Preference Optimization](https://mlsanity.com/paper/2609.16661/) (2026)

## Abstract (arXiv)

Audio large language models (Audio LLMs) exhibit systematic failures in transcribing code-switching speech despite strong multilingual capabilities. Focusing on English-Mandarin, we identify three failure modes: language omission, translation-instead-of-transcription, and hallucination. We apply Direct Preference Optimization (DPO) to align models, constructing preference pairs in which chosen responses preserve mixed-language content while rejected responses mimic failure patterns. Training three Audio LLMs on 100K pairs (570 hours), we observe consistent behavioral shifts: models learn to preserve language composition rather than translating when prompted for transcription. This alignment yields MER reductions up to 89.6% (in-distribution) and 20.0% (out-of-distribution). Our findings suggest DPO can effectively elicit correct code-switching transcription behavior from multilingual Audio LLMs.