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

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.

NTU Speechlab LLM-Based Multilingual ASR System for Interspeech MLC-SLM Challenge 2025

This report details the NTU Speechlab system developed for the Interspeech 2025 Multilingual Conversational Speech and Language Model (MLC-SLM) Challenge (Task I), where we achieved 5th place. We present comprehensive analyses of our multilingual automatic speech recognition system, highlighting key advancements in model architecture, data selection, and training strategies. In particular, language-specific prompts and model averaging techniques were instrumental in boosting system performance across diverse languages. Compared to the initial baseline system, our final model reduced the average Mix Error Rate from 20.2% to 10.6%, representing an absolute improvement of 9.6% (a relative improvement of 48%) on the evaluation set. Our results demonstrate the effectiveness of our approach and offer practical insights for future Speech Large Language Models.

Seewo's Submission to MLC-SLM: Lessons learned from Speech Reasoning Language Models

This paper presents Seewo's systems for both tracks of the Multilingual Conversational Speech Language Model Challenge (MLC-SLM), addressing automatic speech recognition (ASR) and speaker diarization with ASR (SD-ASR). We introduce a multi-stage training pipeline that explicitly enhances reasoning and self-correction in speech language models for ASR. Our approach combines curriculum learning for progressive capability acquisition, Chain-of-Thought data augmentation to foster intermediate reflection, and Reinforcement Learning with Verifiable Rewards (RLVR) to further refine self-correction through reward-driven optimization. This approach achieves substantial improvements over the official challenge baselines. On the evaluation set, our best system attains a WER/CER of 11.57% for Track 1 and a tcpWER/tcpCER of 17.67% for Track 2. Comprehensive ablation studies demonstrate the effectiveness of each component under challenge constraints.

Bi-directional Context-Enhanced Speech Large Language Models for Multilingual Conversational ASR

This paper introduces the integration of language-specific bi-directional context into a speech large language model (SLLM) to improve multilingual continuous conversational automatic speech recognition (ASR). We propose a character-level contextual masking strategy during training, which randomly removes portions of the context to enhance robustness and better emulate the flawed transcriptions that may occur during inference. For decoding, a two-stage pipeline is utilized: initial isolated segment decoding followed by context-aware re-decoding using neighboring hypotheses. Evaluated on the 1500-hour Multilingual Conversational Speech and Language Model (MLC-SLM) corpus covering eleven languages, our method achieves an 18% relative improvement compared to a strong baseline, outperforming even the model trained on 6000 hours of data for the MLC-SLM competition. These results underscore the significant benefit of incorporating contextual information in multilingual continuous conversational ASR.

BUT System for the MLC-SLM Challenge

We present a two-speaker automatic speech recognition (ASR) system that combines DiCoW -- a diarization-conditioned variant of Whisper -- with DiariZen, a diarization pipeline built on top of Pyannote. We first evaluate both systems in out-of-domain (OOD) multilingual scenarios without any fine-tuning. In this scenario, DiariZen consistently outperforms the baseline Pyannote diarization model, demonstrating strong generalization. Despite being fine-tuned on English-only data for target-speaker ASR, DiCoW retains solid multilingual performance, indicating that encoder modifications preserve Whisper's multilingual capabilities. We then fine-tune both DiCoW and DiariZen on the MLC-SLM challenge data. The fine-tuned DiariZen continues to outperform the fine-tuned Pyannote baseline, while DiCoW sees further gains from domain adaptation. Our final system achieves a micro-average tcpWER/CER of 16.75% and ranks second in Task 2 of the MLC-SLM challenge. Lastly, we identify several labeling inconsistencies in the training data -- such as missing speech segments and incorrect silence annotations -- which can hinder diarization fine-tuning. We propose simple mitigation strategies to address these issues and improve system robustness.

SC-SOT: Conditioning the Decoder on Diarized Speaker Information for End-to-End Overlapped Speech Recognition

We propose Speaker-Conditioned Serialized Output Training (SC-SOT), an enhanced SOT-based training for E2E multi-talker ASR. We first probe how SOT handles overlapped speech, and we found the decoder performs implicit speaker separation. We hypothesize this implicit separation is often insufficient due to ambiguous acoustic cues in overlapping regions. To address this, SC-SOT explicitly conditions the decoder on speaker information, providing detailed information about "who spoke when". Specifically, we enhance the decoder by incorporating: (1) speaker embeddings, which allow the model to focus on the acoustic characteristics of the target speaker, and (2) speaker activity information, which guides the model to suppress non-target speakers. The speaker embeddings are derived from a jointly trained E2E speaker diarization model, mitigating the need for speaker enrollment. Experimental results demonstrate the effectiveness of our conditioning approach on overlapped speech.

From Flat to Feeling: A Feasibility and Impact Study on Dynamic Facial Emotions in AI-Generated Avatars

Dynamic facial emotion is essential for believable AI-generated avatars; however, most systems remain visually inert, limiting their utility in high-stakes simulations such as virtual training for investigative interviews with abused children. We introduce and evaluate a real-time architecture fusing Unreal Engine 5 MetaHuman rendering with NVIDIA Omniverse Audio2Face to translate vocal prosody into high-fidelity facial expressions on photorealistic child avatars. We implemented a distributed two-PC setup that decouples language processing and speech synthesis from GPU-intensive rendering, designed to support low-latency interaction in desktop and VR environments. A between-subjects study ($N=70$) using audio+visual and visual-only conditions assessed perceptual impacts as participants rated emotional clarity, facial realism, and empathy for two avatars expressing joy, sadness, and anger. Results demonstrate that avatars could express emotions recognizably, with sadness and joy achieving high identification rates. However, anger recognition significantly dropped without audio, highlighting the importance of congruent vocal cues for high-arousal emotions. Interestingly, removing audio boosted perceived facial realism, suggesting that audiovisual desynchrony remains a key design challenge. These findings confirm the technical feasibility of generating emotionally expressive avatars and provide guidance for improving non-verbal communication in sensitive training simulations.

Adapting Whisper for Streaming Speech Recognition via Two-Pass Decoding

OpenAI Whisper is a family of robust Automatic Speech Recognition (ASR) models trained on 680,000 hours of audio. However, its encoder-decoder architecture, trained with a sequence-to-sequence objective, lacks native support for streaming ASR. In this paper, we fine-tune Whisper for streaming ASR using the WeNet toolkit by adopting a Unified Two-pass (U2) structure. We introduce an additional Connectionist Temporal Classification (CTC) decoder trained with causal attention masks to generate streaming partial transcripts, while the original Whisper decoder reranks these partial outputs. Our experiments on LibriSpeech and an earnings call dataset demonstrate that, with adequate fine-tuning data, Whisper can be adapted into a capable streaming ASR model. We also introduce a hybrid tokenizer approach, which uses a smaller token space for the CTC decoder while retaining Whisper's original token space for the attention decoder, resulting in improved data efficiency and generalization.

Lightweight and Robust Multi-Channel End-to-End Speech Recognition with Spherical Harmonic Transform

This paper presents SHTNet, a lightweight spherical harmonic transform (SHT) based framework, which is designed to address cross-array generalization challenges in multi-channel automatic speech recognition (ASR) through three key innovations. First, SHT based spatial sound field decomposition converts microphone signals into geometry-invariant spherical harmonic coefficients, isolating signal processing from array geometry. Second, the Spatio-Spectral Attention Fusion Network (SSAFN) combines coordinate-aware spatial modeling, refined self-attention channel combinator, and spectral noise suppression without conventional beamforming. Third, Rand-SHT training enhances robustness through random channel selection and array geometry reconstruction. The system achieves 39.26\% average CER across heterogeneous arrays (e.g., circular, square, and binaural) on datasets including Aishell-4, Alimeeting, and XMOS, with 97.1\% fewer computations than conventional neural beamformers.

(SimPhon Speech Test): A Data-Driven Method for In Silico Design and Validation of a Phonetically Balanced Speech Test

Traditional audiometry often provides an incomplete characterization of the functional impact of hearing loss on speech understanding, particularly for supra-threshold deficits common in presbycusis. This motivates the development of more diagnostically specific speech perception tests. We introduce the Simulated Phoneme Speech Test (SimPhon Speech Test) methodology, a novel, multi-stage computational pipeline for the in silico design and validation of a phonetically balanced minimal-pair speech test. This methodology leverages a modern Automatic Speech Recognition (ASR) system as a proxy for a human listener to simulate the perceptual effects of sensorineural hearing loss. By processing speech stimuli under controlled acoustic degradation, we first identify the most common phoneme confusion patterns. These patterns then guide the data-driven curation of a large set of candidate word pairs derived from a comprehensive linguistic corpus. Subsequent phases involving simulated diagnostic testing, expert human curation, and a final, targeted sensitivity analysis systematically reduce the candidates to a final, optimized set of 25 pairs (the SimPhon Speech Test-25). A key finding is that the diagnostic performance of the SimPhon Speech Test-25 test items shows no significant correlation with predictions from the standard Speech Intelligibility Index (SII), suggesting the SimPhon Speech Test captures perceptual deficits beyond simple audibility. This computationally optimized test set offers a significant increase in efficiency for audiological test development, ready for initial human trials.