Code-switching automatic speech recognition (ASR) aims to transcribe speech that contains two or more languages accurately. To better capture language-specific speech representations and address language confusion in code-switching ASR, the mixture-of-experts (MoE) architecture and an additional language diarization (LD) decoder are commonly employed. However, most researches remain stagnant in simple operations like weighted summation or concatenation to fuse language-specific speech representations, leaving significant opportunities to explore the enhancement of integrating language bias information. In this paper, we introduce CAMEL, a cross-attention-based MoE and language bias approach for code-switching ASR. Specifically, after each MoE layer, we fuse language-specific speech representations with cross-attention, leveraging its strong contextual modeling abilities. Additionally, we design a source attention-based mechanism to incorporate the language information from the LD decoder output into text embeddings. Experimental results demonstrate that our approach achieves state-of-the-art performance on the SEAME, ASRU200, and ASRU700+LibriSpeech460 Mandarin-English code-switching ASR datasets.

In hours-long meeting scenarios, real-time speech stream often struggles with achieving accurate speaker diarization, commonly leading to speaker identification and speaker count errors. To address this challenge, we propose SCDiar, a system that operates on speech segments, split at the token level by a speaker change detection (SCD) module. Building on these segments, we introduce several enhancements to efficiently select the best available segment for each speaker. These improvements lead to significant gains across various benchmarks. Notably, on real-world meeting data involving more than ten participants, SCDiar outperforms previous systems by up to 53.6% in accuracy, substantially narrowing the performance gap between online and offline systems.

Target speech extraction (TSE) systems are designed to extract target speech from a multi-talker mixture. The popular training objective for most prior TSE networks is to enhance reconstruction performance of extracted speech waveform. However, it has been reported that a TSE system delivers high reconstruction performance may still suffer low-quality experience problems in practice. One such experience problem is wrong speaker extraction (called speaker confusion, SC), which leads to strong negative experience and hampers effective conversations. To mitigate the imperative SC issue, we reformulate the training objective and propose two novel loss schemes that explore the metric of reconstruction improvement performance defined at small chunk-level and leverage the metric associated distribution information. Both loss schemes aim to encourage a TSE network to pay attention to those SC chunks based on the said distribution information. On this basis, we present X-SepFormer, an end-to-end TSE model with proposed loss schemes and a backbone of SepFormer. Experimental results on the benchmark WSJ0-2mix dataset validate the effectiveness of our proposals, showing consistent improvements on SC errors (by 14.8% relative). Moreover, with SI-SDRi of 19.4 dB and PESQ of 3.81, our best system significantly outperforms the current SOTA systems and offers the top TSE results reported till date on the WSJ0-2mix.

In this paper, we propose multi-band MelGAN, a much faster waveform generation model targeting to high-quality text-to-speech. Specifically, we improve the original MelGAN by the following aspects. First, we increase the receptive field of the generator, which is proven to be beneficial to speech generation. Second, we substitute the feature matching loss with the multi-resolution STFT loss to better measure the difference between fake and real speech. Together with pre-training, this improvement leads to both better quality and better training stability. More importantly, we extend MelGAN with multi-band processing: the generator takes mel-spectrograms as input and produces sub-band signals which are subsequently summed back to full-band signals as discriminator input. The proposed multi-band MelGAN has achieved high MOS of 4.34 and 4.22 in waveform generation and TTS, respectively. With only 1.91M parameters, our model effectively reduces the total computational complexity of the original MelGAN from 5.85 to 0.95 GFLOPS. Our Pytorch implementation can achieve a real-time factor of 0.03 on CPU without hardware specific optimization.

A text-independent speaker verification system suffers severe performance degradation under short utterance condition. To address the problem, in this paper, we propose an adversarially learned embedding mapping model that directly maps a short embedding to an enhanced embedding with increased discriminability. In particular, a Wasserstein GAN with a bunch of loss criteria are investigated. These loss functions have distinct optimization objectives and some of them are less favoured for the speaker verification research area. Different from most prior studies, our main objective in this study is to investigate the effectiveness of those loss criteria by conducting numerous ablation studies. Experiments on Voxceleb dataset showed that some criteria are beneficial to the verification performance while some have trivial effects. Lastly, a Wasserstein GAN with chosen loss criteria, without finetuning, achieves meaningful advancements over the baseline, with 4% relative improvements on EER and 7% on minDCF in the challenging scenario of short 2second utterances.