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.

Recent advancements in audio tokenization have significantly enhanced the integration of audio capabilities into large language models (LLMs). However, audio understanding and generation are often treated as distinct tasks, hindering the development of truly unified audio-language models. While instruction tuning has demonstrated remarkable success in improving generalization and zero-shot learning across text and vision, its application to audio remains largely unexplored. A major obstacle is the lack of comprehensive datasets that unify audio understanding and generation. To address this, we introduce Audio-FLAN, a large-scale instruction-tuning dataset covering 80 diverse tasks across speech, music, and sound domains, with over 100 million instances. Audio-FLAN lays the foundation for unified audio-language models that can seamlessly handle both understanding (e.g., transcription, comprehension) and generation (e.g., speech, music, sound) tasks across a wide range of audio domains in a zero-shot manner. The Audio-FLAN dataset is available on HuggingFace and GitHub and will be continuously updated.

Multilingual audio-text retrieval (ML-ATR) is a challenging task that aims to retrieve audio clips or multilingual texts from databases. However, existing ML-ATR schemes suffer from inconsistencies for instance similarity matching across languages. We theoretically analyze the inconsistency in terms of both multilingual modal alignment direction error and weight error, and propose the theoretical weight error upper bound for quantifying the inconsistency. Based on the analysis of the weight error upper bound, we find that the inconsistency problem stems from the data distribution error caused by random sampling of languages. We propose a consistent ML-ATR scheme using 1-to-k contrastive learning and audio-English co-anchor contrastive learning, aiming to mitigate the negative impact of data distribution error on recall and consistency in ML-ATR. Experimental results on the translated AudioCaps and Clotho datasets show that our scheme achieves state-of-the-art performance on recall and consistency metrics for eight mainstream languages, including English. Our code will be available at https://github.com/ATRI-ACL/ATRI-ACL.

With the rise of Speech Large Language Models (Speech LLMs), there has been growing interest in discrete speech tokens for their ability to integrate with text-based tokens seamlessly. Compared to most studies that focus on continuous speech features, although discrete-token based LLMs have shown promising results on certain tasks, the performance gap between these two paradigms is rarely explored. In this paper, we present a fair and thorough comparison between discrete and continuous features across a variety of semantic-related tasks using a light-weight LLM (Qwen1.5-0.5B). Our findings reveal that continuous features generally outperform discrete tokens, particularly in tasks requiring fine-grained semantic understanding. Moreover, this study goes beyond surface-level comparison by identifying key factors behind the under-performance of discrete tokens, such as limited token granularity and inefficient information retention. To enhance the performance of discrete tokens, we explore potential aspects based on our analysis. We hope our results can offer new insights into the opportunities for advancing discrete speech tokens in Speech LLMs.

We present RALL-E, a robust language modeling method for text-to-speech (TTS) synthesis. While previous work based on large language models (LLMs) shows impressive performance on zero-shot TTS, such methods often suffer from poor robustness, such as unstable prosody (weird pitch and rhythm/duration) and a high word error rate (WER), due to the autoregressive prediction style of language models. The core idea behind RALL-E is chain-of-thought (CoT) prompting, which decomposes the task into simpler steps to enhance the robustness of LLM-based TTS. To accomplish this idea, RALL-E first predicts prosody features (pitch and duration) of the input text and uses them as intermediate conditions to predict speech tokens in a CoT style. Second, RALL-E utilizes the predicted duration prompt to guide the computing of self-attention weights in Transformer to enforce the model to focus on the corresponding phonemes and prosody features when predicting speech tokens. Results of comprehensive objective and subjective evaluations demonstrate that, compared to a powerful baseline method VALL-E, RALL-E significantly improves the WER of zero-shot TTS from $5.6\%$ (without reranking) and $1.7\%$ (with reranking) to $2.5\%$ and $1.0\%$, respectively. Furthermore, we demonstrate that RALL-E correctly synthesizes sentences that are hard for VALL-E and reduces the error rate from $68\%$ to $4\%$.

While recent large-scale text-to-speech (TTS) models have achieved significant progress, they still fall short in speech quality, similarity, and prosody. Considering speech intricately encompasses various attributes (e.g., content, prosody, timbre, and acoustic details) that pose significant challenges for generation, a natural idea is to factorize speech into individual subspaces representing different attributes and generate them individually. Motivated by it, we propose NaturalSpeech 3, a TTS system with novel factorized diffusion models to generate natural speech in a zero-shot way. Specifically, 1) we design a neural codec with factorized vector quantization (FVQ) to disentangle speech waveform into subspaces of content, prosody, timbre, and acoustic details; 2) we propose a factorized diffusion model to generate attributes in each subspace following its corresponding prompt. With this factorization design, NaturalSpeech 3 can effectively and efficiently model intricate speech with disentangled subspaces in a divide-and-conquer way. Experiments show that NaturalSpeech 3 outperforms the state-of-the-art TTS systems on quality, similarity, prosody, and intelligibility, and achieves on-par quality with human recordings. Furthermore, we achieve better performance by scaling to 1B parameters and 200K hours of training data.

Speech conveys more information than text, as the same word can be uttered in various voices to convey diverse information. Compared to traditional text-to-speech (TTS) methods relying on speech prompts (reference speech) for voice variability, using text prompts (descriptions) is more user-friendly since speech prompts can be hard to find or may not exist at all. TTS approaches based on the text prompt face two main challenges: 1) the one-to-many problem, where not all details about voice variability can be described in the text prompt, and 2) the limited availability of text prompt datasets, where vendors and large cost of data labeling are required to write text prompts for speech. In this work, we introduce PromptTTS 2 to address these challenges with a variation network to provide variability information of voice not captured by text prompts, and a prompt generation pipeline to utilize the large language models (LLM) to compose high quality text prompts. Specifically, the variation network predicts the representation extracted from the reference speech (which contains full information about voice variability) based on the text prompt representation. For the prompt generation pipeline, it generates text prompts for speech with a speech language understanding model to recognize voice attributes (e.g., gender, speed) from speech and a large language model to formulate text prompts based on the recognition results. Experiments on a large-scale (44K hours) speech dataset demonstrate that compared to the previous works, PromptTTS 2 generates voices more consistent with text prompts and supports the sampling of diverse voice variability, thereby offering users more choices on voice generation. Additionally, the prompt generation pipeline produces high-quality text prompts, eliminating the large labeling cost. The demo page of PromptTTS 2 is available online.

Various applications of voice synthesis have been developed independently despite the fact that they generate "voice" as output in common. In addition, the majority of voice synthesis models currently rely on annotated audio data, but it is crucial to scale them to self-supervised datasets in order to effectively capture the wide range of acoustic variations present in human voice, including speaker identity, emotion, and prosody. In this work, we propose Make-A-Voice, a unified framework for synthesizing and manipulating voice signals from discrete representations. Make-A-Voice leverages a "coarse-to-fine" approach to model the human voice, which involves three stages: 1) semantic stage: model high-level transformation between linguistic content and self-supervised semantic tokens, 2) acoustic stage: introduce varying control signals as acoustic conditions for semantic-to-acoustic modeling, and 3) generation stage: synthesize high-fidelity waveforms from acoustic tokens. Make-A-Voice offers notable benefits as a unified voice synthesis framework: 1) Data scalability: the major backbone (i.e., acoustic and generation stage) does not require any annotations, and thus the training data could be scaled up. 2) Controllability and conditioning flexibility: we investigate different conditioning mechanisms and effectively handle three voice synthesis applications, including text-to-speech (TTS), voice conversion (VC), and singing voice synthesis (SVS) by re-synthesizing the discrete voice representations with prompt guidance. Experimental results demonstrate that Make-A-Voice exhibits superior audio quality and style similarity compared with competitive baseline models. Audio samples are available at https://Make-A-Voice.github.io

Large diffusion models have been successful in text-to-audio (T2A) synthesis tasks, but they often suffer from common issues such as semantic misalignment and poor temporal consistency due to limited natural language understanding and data scarcity. Additionally, 2D spatial structures widely used in T2A works lead to unsatisfactory audio quality when generating variable-length audio samples since they do not adequately prioritize temporal information. To address these challenges, we propose Make-an-Audio 2, a latent diffusion-based T2A method that builds on the success of Make-an-Audio. Our approach includes several techniques to improve semantic alignment and temporal consistency: Firstly, we use pre-trained large language models (LLMs) to parse the text into structured pairs for better temporal information capture. We also introduce another structured-text encoder to aid in learning semantic alignment during the diffusion denoising process. To improve the performance of variable length generation and enhance the temporal information extraction, we design a feed-forward Transformer-based diffusion denoiser. Finally, we use LLMs to augment and transform a large amount of audio-label data into audio-text datasets to alleviate the problem of scarcity of temporal data. Extensive experiments show that our method outperforms baseline models in both objective and subjective metrics, and achieves significant gains in temporal information understanding, semantic consistency, and sound quality.

Generating sound effects that humans want is an important topic. However, there are few studies in this area for sound generation. In this study, we investigate generating sound conditioned on a text prompt and propose a novel text-to-sound generation framework that consists of a text encoder, a Vector Quantized Variational Autoencoder (VQ-VAE), a decoder, and a vocoder. The framework first uses the decoder to transfer the text features extracted from the text encoder to a mel-spectrogram with the help of VQ-VAE, and then the vocoder is used to transform the generated mel-spectrogram into a waveform. We found that the decoder significantly influences the generation performance. Thus, we focus on designing a good decoder in this study. We begin with the traditional autoregressive decoder, which has been proved as a state-of-the-art method in previous sound generation works. However, the AR decoder always predicts the mel-spectrogram tokens one by one in order, which introduces the unidirectional bias and accumulation of errors problems. Moreover, with the AR decoder, the sound generation time increases linearly with the sound duration. To overcome the shortcomings introduced by AR decoders, we propose a non-autoregressive decoder based on the discrete diffusion model, named Diffsound. Specifically, the Diffsound predicts all of the mel-spectrogram tokens in one step and then refines the predicted tokens in the next step, so the best-predicted results can be obtained after several steps. Our experiments show that our proposed Diffsound not only produces better text-to-sound generation results when compared with the AR decoder but also has a faster generation speed, e.g., MOS: 3.56 \textit{v.s} 2.786, and the generation speed is five times faster than the AR decoder.