Sequence-to-Sequence (seq2seq) tasks transcribe the input sequence to a target sequence. The Connectionist Temporal Classification (CTC) criterion is widely used in multiple seq2seq tasks. Besides predicting the target sequence, a side product of CTC is to predict the alignment, which is the most probable input-long sequence that specifies a hard aligning relationship between the input and target units. As there are multiple potential aligning sequences (called paths) that are equally considered in CTC formulation, the choice of which path will be most probable and become the predicted alignment is always uncertain. In addition, it is usually observed that the alignment predicted by vanilla CTC will drift compared with its reference and rarely provides practical functionalities. Thus, the motivation of this work is to make the CTC alignment prediction controllable and thus equip CTC with extra functionalities. The Bayes risk CTC (BRCTC) criterion is then proposed in this work, in which a customizable Bayes risk function is adopted to enforce the desired characteristics of the predicted alignment. With the risk function, the BRCTC is a general framework to adopt some customizable preference over the paths in order to concentrate the posterior into a particular subset of the paths. In applications, we explore one particular preference which yields models with the down-sampling ability and reduced inference costs. By using BRCTC with another preference for early emissions, we obtain an improved performance-latency trade-off for online models. Experimentally, the proposed BRCTC reduces the inference cost of offline models by up to 47% without performance degradation and cuts down the overall latency of online systems to an unseen level.
Speaker identification, determining which character said each utterance in literary text, benefits many downstream tasks. Most existing approaches use expert-defined rules or rule-based features to directly approach this task, but these approaches come with significant drawbacks, such as lack of contextual reasoning and poor cross-lingual generalization. In this work, we propose a speaker identification framework that addresses these issues. We first extract large-scale distant supervision signals in English via general-purpose tools and heuristics, and then apply these weakly-labeled instances with a focus on encouraging contextual reasoning to train a cross-lingual language model. We show that the resulting model outperforms previous state-of-the-art methods on two English speaker identification benchmarks by up to 9% in accuracy and 5% with only distant supervision, as well as two Chinese speaker identification datasets by up to 4.7%.
Although large language models have achieved impressive zero-shot ability, the huge model size generally incurs high cost. Recently, semi-parametric language models, which augment a smaller language model with an external retriever, have demonstrated promising language modeling capabilities. However, it remains unclear whether such semi-parametric language models can perform competitively well as their fully-parametric counterparts on zero-shot generalization to downstream tasks. In this work, we introduce $\text{Zemi}$, a zero-shot semi-parametric language model. To our best knowledge, this is the first semi-parametric language model that can demonstrate strong zero-shot performance on a wide range of held-out unseen tasks. We train $\text{Zemi}$ with a novel semi-parametric multitask prompted training paradigm, which shows significant improvement compared with the parametric multitask training as proposed by T0. Specifically, we augment the multitask training and zero-shot evaluation with retrieval from a large-scale task-agnostic unlabeled corpus. In order to incorporate multiple potentially noisy retrieved augmentations, we further propose a novel $\text{augmentation fusion}$ module leveraging perceiver resampler and gated cross-attention. Notably, our proposed $\text{Zemi}_\text{LARGE}$ outperforms T0-3B by 16% on all seven evaluation tasks while being 3.9x smaller in model size.
Self-supervised learning (SSL) has drawn an increased attention in the field of speech processing. Recent studies have demonstrated that contrastive learning is able to learn discriminative speaker embeddings in a self-supervised manner. However, base contrastive self-supervised learning (CSSL) assumes that the pairs generated from a view of anchor instance and any view of other instances are all negative, which introduces many false negative pairs in constructing the loss function. The problem is referred as $class$-$collision$, which remains as one major issue that impedes the CSSL based speaker verification (SV) systems from achieving better performances. In the meanwhile, studies reveal that negative sample free SSL frameworks perform well in learning speaker or image representations. In this study, we investigate SSL techniques that lead to an improved SV performance. We first analyse the impact of false negative pairs in the CSSL systems. Then, a multi-stage Class-Collision Correction (C3) method is proposed, which leads to the state-of-the-art CSSL based speaker embedding system. On the basis of the pretrained CSSL model, we further propose to employ a negative sample free SSL objective (i.e., DINO) to fine-tune the speaker embedding network. The resulting speaker embedding system (C3-DINO) achieves 2.5% EER with a simple Cosine Distance Scoring method on Voxceleb1 test set, which outperforms the previous SOTA SSL system (4.86%) by a significant +45% relative improvement. With speaker clustering and pseudo labeling on Voxceleb2 training set, a LDA/CDS back-end applying on the C3-DINO speaker embeddings is able to further push the EER to 2.2%. Comprehensive experimental investigations of the Voxceleb benchmarks and our internal dataset demonstrate the effectiveness of our proposed methods, and the performance gap between the SSL SV and the supervised counterpart narrows further.
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.
Utterance rewriting aims to recover coreferences and omitted information from the latest turn of a multi-turn dialogue. Recently, methods that tag rather than linearly generate sequences have proven stronger in both in- and out-of-domain rewriting settings. This is due to a tagger's smaller search space as it can only copy tokens from the dialogue context. However, these methods may suffer from low coverage when phrases that must be added to a source utterance cannot be covered by a single context span. This can occur in languages like English that introduce tokens such as prepositions into the rewrite for grammaticality. We propose a hierarchical context tagger (HCT) that mitigates this issue by predicting slotted rules (e.g., "besides _") whose slots are later filled with context spans. HCT (i) tags the source string with token-level edit actions and slotted rules and (ii) fills in the resulting rule slots with spans from the dialogue context. This rule tagging allows HCT to add out-of-context tokens and multiple spans at once; we further cluster the rules to truncate the long tail of the rule distribution. Experiments on several benchmarks show that HCT can outperform state-of-the-art rewriting systems by ~2 BLEU points.
Prosodic boundary plays an important role in text-to-speech synthesis (TTS) in terms of naturalness and readability. However, the acquisition of prosodic boundary labels relies on manual annotation, which is costly and time-consuming. In this paper, we propose to automatically extract prosodic boundary labels from text-audio data via a neural text-speech model with pre-trained audio encoders. This model is pre-trained on text and speech data separately and jointly fine-tuned on TTS data in a triplet format: {speech, text, prosody}. The experimental results on both automatic evaluation and human evaluation demonstrate that: 1) the proposed text-speech prosody annotation framework significantly outperforms text-only baselines; 2) the quality of automatic prosodic boundary annotations is comparable to human annotations; 3) TTS systems trained with model-annotated boundaries are slightly better than systems that use manual ones.
In this paper, we propose a novel unsupervised text-to-speech (UTTS) framework which does not require text-audio pairs for the TTS acoustic modeling (AM). UTTS is a multi-speaker speech synthesizer developed from the perspective of disentangled speech representation learning. The framework offers a flexible choice of a speaker's duration model, timbre feature (identity) and content for TTS inference. We leverage recent advancements in self-supervised speech representation learning as well as speech synthesis front-end techniques for the system development. Specifically, we utilize a lexicon to map input text to the phoneme sequence, which is expanded to the frame-level forced alignment (FA) with a speaker-dependent duration model. Then, we develop an alignment mapping module that converts the FA to the unsupervised alignment (UA). Finally, a Conditional Disentangled Sequential Variational Auto-encoder (C-DSVAE), serving as the self-supervised TTS AM, takes the predicted UA and a target speaker embedding to generate the mel spectrogram, which is ultimately converted to waveform with a neural vocoder. We show how our method enables speech synthesis without using a paired TTS corpus. Experiments demonstrate that UTTS can synthesize speech of high naturalness and intelligibility measured by human and objective evaluations.
Despite the rapid progress in automatic speech recognition (ASR) research, recognizing multilingual speech using a unified ASR system remains highly challenging. Previous works on multilingual speech recognition mainly focus on two directions: recognizing multiple monolingual speech or recognizing code-switched speech that uses different languages interchangeably within a single utterance. However, a pragmatic multilingual recognizer is expected to be compatible with both directions. In this work, a novel language-aware encoder (LAE) architecture is proposed to handle both situations by disentangling language-specific information and generating frame-level language-aware representations during encoding. In the LAE, the primary encoding is implemented by the shared block while the language-specific blocks are used to extract specific representations for each language. To learn language-specific information discriminatively, a language-aware training method is proposed to optimize the language-specific blocks in LAE. Experiments conducted on Mandarin-English code-switched speech suggest that the proposed LAE is capable of discriminating different languages in frame-level and shows superior performance on both monolingual and multilingual ASR tasks. With either a real-recorded or simulated code-switched dataset, the proposed LAE achieves statistically significant improvements on both CTC and neural transducer systems. Code is released
Acoustic echo cancellation (AEC) plays an important role in the full-duplex speech communication as well as the front-end speech enhancement for recognition in the conditions when the loudspeaker plays back. In this paper, we present an all-deep-learning framework that implicitly estimates the second order statistics of echo/noise and target speech, and jointly solves echo and noise suppression through an attention based recurrent neural network. The proposed model outperforms the state-of-the-art joint echo cancellation and speech enhancement method F-T-LSTM in terms of objective speech quality metrics, speech recognition accuracy and model complexity. We show that this model can work with speaker embedding for better target speech enhancement and furthermore develop a branch for automatic gain control (AGC) task to form an all-in-one front-end speech enhancement system.