Recent advances in text-to-speech synthesis have achieved notable success in generating high-quality short utterances for individual speakers. However, these systems still face challenges when extending their capabilities to long, multi-speaker, and spontaneous dialogues, typical of real-world scenarios such as podcasts. These limitations arise from two primary challenges: 1) long speech: podcasts typically span several minutes, exceeding the upper limit of most existing work; 2) spontaneity: podcasts are marked by their spontaneous, oral nature, which sharply contrasts with formal, written contexts; existing works often fall short in capturing this spontaneity. In this paper, we propose MoonCast, a solution for high-quality zero-shot podcast generation, aiming to synthesize natural podcast-style speech from text-only sources (e.g., stories, technical reports, news in TXT, PDF, or Web URL formats) using the voices of unseen speakers. To generate long audio, we adopt a long-context language model-based audio modeling approach utilizing large-scale long-context speech data. To enhance spontaneity, we utilize a podcast generation module to generate scripts with spontaneous details, which have been empirically shown to be as crucial as the text-to-speech modeling itself. Experiments demonstrate that MoonCast outperforms baselines, with particularly notable improvements in spontaneity and coherence.

We tackle the task of long-form music generation--particularly the challenging \textbf{lyrics-to-song} problem--by introducing YuE, a family of open foundation models based on the LLaMA2 architecture. Specifically, YuE scales to trillions of tokens and generates up to five minutes of music while maintaining lyrical alignment, coherent musical structure, and engaging vocal melodies with appropriate accompaniment. It achieves this through (1) track-decoupled next-token prediction to overcome dense mixture signals, (2) structural progressive conditioning for long-context lyrical alignment, and (3) a multitask, multiphase pre-training recipe to converge and generalize. In addition, we redesign the in-context learning technique for music generation, enabling versatile style transfer (e.g., converting Japanese city pop into an English rap while preserving the original accompaniment) and bidirectional generation. Through extensive evaluation, we demonstrate that YuE matches or even surpasses some of the proprietary systems in musicality and vocal agility. In addition, fine-tuning YuE enables additional controls and enhanced support for tail languages. Furthermore, beyond generation, we show that YuE's learned representations can perform well on music understanding tasks, where the results of YuE match or exceed state-of-the-art methods on the MARBLE benchmark. Keywords: lyrics2song, song generation, long-form, foundation model, music generation

Recent years have witnessed the success of foundation models pre-trained with self-supervised learning (SSL) in various music informatics understanding tasks, including music tagging, instrument classification, key detection, and more. In this paper, we propose a self-supervised music representation learning model for music understanding. Distinguished from previous studies adopting random projection or existing neural codec, the proposed model, named MuQ, is trained to predict tokens generated by Mel Residual Vector Quantization (Mel-RVQ). Our Mel-RVQ utilizes residual linear projection structure for Mel spectrum quantization to enhance the stability and efficiency of target extraction and lead to better performance. Experiments in a large variety of downstream tasks demonstrate that MuQ outperforms previous self-supervised music representation models with only 0.9K hours of open-source pre-training data. Scaling up the data to over 160K hours and adopting iterative training consistently improve the model performance. To further validate the strength of our model, we present MuQ-MuLan, a joint music-text embedding model based on contrastive learning, which achieves state-of-the-art performance in the zero-shot music tagging task on the MagnaTagATune dataset. Code and checkpoints are open source in https://github.com/tencent-ailab/MuQ.

We introduce Gull, a generative multifunctional audio codec. Gull is a general purpose neural audio compression and decompression model which can be applied to a wide range of tasks and applications such as real-time communication, audio super-resolution, and codec language models. The key components of Gull include (1) universal-sample-rate modeling via subband modeling schemes motivated by recent progress in audio source separation, (2) gain-shape representations motivated by traditional audio codecs, (3) improved residual vector quantization modules, (4) elastic decoder network that enables user-defined model size and complexity during inference time, (5) built-in ability for audio super-resolution without the increase of bitrate. We compare Gull with existing traditional and neural audio codecs and show that Gull is able to achieve on par or better performance across various sample rates, bitrates and model complexities in both subjective and objective evaluation metrics.

Humans possess an extraordinary ability to selectively focus on the sound source of interest amidst complex acoustic environments, commonly referred to as cocktail party scenarios. In an attempt to replicate this remarkable auditory attention capability in machines, target speaker extraction (TSE) models have been developed. These models leverage the pre-registered cues of the target speaker to extract the sound source of interest. However, the effectiveness of these models is hindered in real-world scenarios due to the unreliable or even absence of pre-registered cues. To address this limitation, this study investigates the integration of natural language description to enhance the feasibility, controllability, and performance of existing TSE models. Specifically, we propose a model named LLM-TSE, wherein a large language model (LLM) extracts useful semantic cues from the user's typed text input. These cues can serve as independent extraction cues, task selectors to control the TSE process or complement the pre-registered cues. Our experimental results demonstrate competitive performance when only text-based cues are presented, the effectiveness of using input text as a task selector, and a new state-of-the-art when combining text-based cues with pre-registered cues. To our knowledge, this is the first study to successfully incorporate LLMs to guide target speaker extraction, which can be a cornerstone for cocktail party problem research. Demos are provided at https://github.com/haoxiangsnr/llm-tse Colin, 1953) - a term coined to describe a scenario where multiple sound sources are engaged in simultaneous conversation, yet a listener can selectively concentrate on a single sound source. This scenario represents a complex challenge in auditory perception (Haykin & Chen, 2005; Mesgarani & Chang, 2012; Bizley & Cohen, 2013) and serves as a remarkable demonstration of the intricate sound processing that occurs within the human auditory system.

We choose the dual-path transformer-based full-subband network (DPT-FSNet) [12] to explore compression methods for Echo cancellation and noise reduction are essential for fullduplex three reasons. First, the model has exhibited high wide-band communication, yet most existing neural networks have perceptual evaluation of speech quality (WB-PESQ) scores on high computational costs and are inflexible in tuning model the NS task with a small number of parameters but suffers from complexity. In this paper, we introduce time-frequency dualpath large computational cost. Second, DPT-FSNet is conducted compression to achieve a wide range of compression ratios on complete time-frequency (T-F) feature maps, indicating its on computational cost. Specifically, for frequency compression, complexity being closely related to the number of frames and trainable filters are used to replace manually designed frequency bands. Third, the model involves a 2D convolution filters for dimension reduction. For time compression, only using encoder, a dual-path transformer and a 2D convolution decoder, frame skipped prediction causes large performance degradation, implying that compression methods should be applicable to different which can be alleviated by a post-processing network modules.

End-to-end models, such as the neural Transducer, have been successful in integrating acoustic and linguistic information jointly to achieve excellent recognition performance. However, adapting these models with text-only data is challenging. Factorized neural Transducer (FNT) aims to address this issue by introducing a separate vocabulary decoder to predict the vocabulary, which can effectively perform traditional text data adaptation. Nonetheless, this approach has limitations in fusing acoustic and language information seamlessly. Moreover, a degradation in word error rate (WER) on the general test sets was also observed, leading to doubts about its overall performance. In response to this challenge, we present an improved factorized neural Transducer (IFNT) model structure designed to comprehensively integrate acoustic and language information while enabling effective text adaptation. We evaluate the performance of our proposed methods through in-domain experiments on GigaSpeech and out-of-domain experiments adapting to EuroParl, TED-LIUM, and Medical datasets. After text-only adaptation, IFNT yields 7.9% to 28.5% relative WER improvements over the standard neural Transducer with shallow fusion, and relative WER reductions ranging from 1.6% to 8.2% on the three test sets compared to the FNT model.

Automatic speech recognition (ASR) based on transducers is widely used. In training, a transducer maximizes the summed posteriors of all paths. The path with the highest posterior is commonly defined as the predicted alignment between the speech and the transcription. While the vanilla transducer does not have a prior preference for any of the valid paths, this work intends to enforce the preferred paths and achieve controllable alignment prediction. Specifically, this work proposes Bayes Risk Transducer (BRT), which uses a Bayes risk function to set lower risk values to the preferred paths so that the predicted alignment is more likely to satisfy specific desired properties. We further demonstrate that these predicted alignments with intentionally designed properties can provide practical advantages over the vanilla transducer. Experimentally, the proposed BRT saves inference cost by up to 46% for non-streaming ASR and reduces overall system latency by 41% for streaming ASR.

Accurate recognition of dysarthric and elderly speech remain challenging tasks to date. Speaker-level heterogeneity attributed to accent or gender, when aggregated with age and speech impairment, create large diversity among these speakers. Scarcity of speaker-level data limits the practical use of data-intensive model based speaker adaptation methods. To this end, this paper proposes two novel forms of data-efficient, feature-based on-the-fly speaker adaptation methods: variance-regularized spectral basis embedding (SVR) and spectral feature driven f-LHUC transforms. Experiments conducted on UASpeech dysarthric and DementiaBank Pitt elderly speech corpora suggest the proposed on-the-fly speaker adaptation approaches consistently outperform baseline iVector adapted hybrid DNN/TDNN and E2E Conformer systems by statistically significant WER reduction of 2.48%-2.85% absolute (7.92%-8.06% relative), and offline model based LHUC adaptation by 1.82% absolute (5.63% relative) respectively.

A key challenge in dysarthric speech recognition is the speaker-level diversity attributed to both speaker-identity associated factors such as gender, and speech impairment severity. Most prior researches on addressing this issue focused on using speaker-identity only. To this end, this paper proposes a novel set of techniques to use both severity and speaker-identity in dysarthric speech recognition: a) multitask training incorporating severity prediction error; b) speaker-severity aware auxiliary feature adaptation; and c) structured LHUC transforms separately conditioned on speaker-identity and severity. Experiments conducted on UASpeech suggest incorporating additional speech impairment severity into state-of-the-art hybrid DNN, E2E Conformer and pre-trained Wav2vec 2.0 ASR systems produced statistically significant WER reductions up to 4.78% (14.03% relative). Using the best system the lowest published WER of 17.82% (51.25% on very low intelligibility) was obtained on UASpeech.