End-to-End DOA-Guided Speech Extraction in Noisy Multi-Talker Scenarios

Target Speaker Extraction (TSE) plays a critical role in enhancing speech signals in noisy and multi-speaker environments. This paper presents an end-to-end TSE model that incorporates Direction of Arrival (DOA) and beamwidth embeddings to extract speech from a specified spatial region centered around the DOA. Our approach efficiently captures spatial and temporal features, enabling robust performance in highly complex scenarios with multiple simultaneous speakers. Experimental results demonstrate that the proposed model not only significantly enhances the target speech within the defined beamwidth but also effectively suppresses interference from other directions, producing a clear and isolated target voice. Furthermore, the model achieves remarkable improvements in downstream Automatic Speech Recognition (ASR) tasks, making it particularly suitable for real-world applications.

Step-Audio 2 Technical Report

This paper presents Step-Audio 2, an end-to-end multi-modal large language model designed for industry-strength audio understanding and speech conversation. By integrating a latent audio encoder and reasoning-centric reinforcement learning (RL), Step-Audio 2 achieves promising performance in automatic speech recognition (ASR) and audio understanding. To facilitate genuine end-to-end speech conversation, Step-Audio 2 incorporates the generation of discrete audio tokens into language modeling, significantly enhancing its responsiveness to paralinguistic information such as speaking styles and emotions. To effectively leverage the rich textual and acoustic knowledge in real-world data, Step-Audio 2 integrates retrieval-augmented generation (RAG) and is able to call external tools such as web search to mitigate hallucination and audio search to switch timbres. Trained on millions of hours of speech and audio data, Step-Audio 2 delivers intelligence and expressiveness across diverse conversational scenarios. Evaluation results demonstrate that Step-Audio 2 achieves state-of-the-art performance on various audio understanding and conversational benchmarks compared to other open-source and commercial solutions. Please visit https://github.com/stepfun-ai/Step-Audio2 for more information.

Improving Contextual ASR via Multi-grained Fusion with Large Language Models

While end-to-end Automatic Speech Recognition (ASR) models have shown impressive performance in transcribing general speech, they often struggle to accurately recognize contextually relevant keywords, such as proper nouns or user-specific entities. Previous approaches have explored leveraging keyword dictionaries in the textual modality to improve keyword recognition, either through token-level fusion that guides token-by-token generation or phrase-level fusion that enables direct copying of keyword phrases. However, these methods operate at different granularities and have their own limitations. In this paper, we propose a novel multi-grained fusion approach that jointly leverages the strengths of both token-level and phrase-level fusion with Large Language Models (LLMs). Our approach incorporates a late-fusion strategy that elegantly combines ASR's acoustic information with LLM's rich contextual knowledge, balancing fine-grained token precision with holistic phrase-level understanding. Experiments on Chinese and English datasets demonstrate that our approach achieves state-of-the-art performance on keyword-related metrics while preserving high accuracy on non-keyword text. Ablation studies further confirm that the token-level and phrase-level components both contribute significantly to the performance gains, complementing each other in our joint multi-grained framework. The code and models will be publicly available at https://github.com/.

Code-Switching in End-to-End Automatic Speech Recognition: A Systematic Literature Review

Motivated by a growing research interest into automatic speech recognition (ASR), and the growing body of work for languages in which code-switching (CS) often occurs, we present a systematic literature review of code-switching in end-to-end ASR models. We collect and manually annotate papers published in peer reviewed venues. We document the languages considered, datasets, metrics, model choices, and performance, and present a discussion of challenges in end-to-end ASR for code-switching. Our analysis thus provides insights on current research efforts and available resources as well as opportunities and gaps to guide future research.

End-to-end Acoustic-linguistic Emotion and Intent Recognition Enhanced by Semi-supervised Learning

Emotion and intent recognition from speech is essential and has been widely investigated in human-computer interaction. The rapid development of social media platforms, chatbots, and other technologies has led to a large volume of speech data streaming from users. Nevertheless, annotating such data manually is expensive, making it challenging to train machine learning models for recognition purposes. To this end, we propose applying semi-supervised learning to incorporate a large scale of unlabelled data alongside a relatively smaller set of labelled data. We train end-to-end acoustic and linguistic models, each employing multi-task learning for emotion and intent recognition. Two semi-supervised learning approaches, including fix-match learning and full-match learning, are compared. The experimental results demonstrate that the semi-supervised learning approaches improve model performance in speech emotion and intent recognition from both acoustic and text data. The late fusion of the best models outperforms the acoustic and text baselines by joint recognition balance metrics of 12.3% and 10.4%, respectively.

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.

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.

Improving Practical Aspects of End-to-End Multi-Talker Speech Recognition for Online and Offline Scenarios

We extend the frameworks of Serialized Output Training (SOT) to address practical needs of both streaming and offline automatic speech recognition (ASR) applications. Our approach focuses on balancing latency and accuracy, catering to real-time captioning and summarization requirements. We propose several key improvements: (1) Leveraging Continuous Speech Separation (CSS) single-channel front-end with end-to-end (E2E) systems for highly overlapping scenarios, challenging the conventional wisdom of E2E versus cascaded setups. The CSS framework improves the accuracy of the ASR system by separating overlapped speech from multiple speakers. (2) Implementing dual models -- Conformer Transducer for streaming and Sequence-to-Sequence for offline -- or alternatively, a two-pass model based on cascaded encoders. (3) Exploring segment-based SOT (segSOT) which is better suited for offline scenarios while also enhancing readability of multi-talker transcriptions.

Regularizing Learnable Feature Extraction for Automatic Speech Recognition

Neural front-ends are an appealing alternative to traditional, fixed feature extraction pipelines for automatic speech recognition (ASR) systems since they can be directly trained to fit the acoustic model. However, their performance often falls short compared to classical methods, which we show is largely due to their increased susceptibility to overfitting. This work therefore investigates regularization methods for training ASR models with learnable feature extraction front-ends. First, we examine audio perturbation methods and show that larger relative improvements can be obtained for learnable features. Additionally, we identify two limitations in the standard use of SpecAugment for these front-ends and propose masking in the short time Fourier transform (STFT)-domain as a simple but effective modification to address these challenges. Finally, integrating both regularization approaches effectively closes the performance gap between traditional and learnable features.

Unifying Streaming and Non-streaming Zipformer-based ASR

There has been increasing interest in unifying streaming and non-streaming automatic speech recognition (ASR) models to reduce development, training, and deployment costs. We present a unified framework that trains a single end-to-end ASR model for both streaming and non-streaming applications, leveraging future context information. We propose to use dynamic right-context through the chunked attention masking in the training of zipformer-based ASR models. We demonstrate that using right-context is more effective in zipformer models compared to other conformer models due to its multi-scale nature. We analyze the effect of varying the number of right-context frames on accuracy and latency of the streaming ASR models. We use Librispeech and large in-house conversational datasets to train different versions of streaming and non-streaming models and evaluate them in a production grade server-client setup across diverse testsets of different domains. The proposed strategy reduces word error by relative 7.9\% with a small degradation in user-perceived latency. By adding more right-context frames, we are able to achieve streaming performance close to that of non-streaming models. Our approach also allows flexible control of the latency-accuracy tradeoff according to customers requirements.