Large-scale speech self-supervised learning (SSL) has emerged to the main field of speech processing, however, the problem of computational cost arising from its vast size makes a high entry barrier to academia. In addition, existing distillation techniques of speech SSL models compress the model by reducing layers, which induces performance degradation in linguistic pattern recognition tasks such as phoneme recognition (PR). In this paper, we propose FitHuBERT, which makes thinner in dimension throughout almost all model components and deeper in layer compared to prior speech SSL distillation works. Moreover, we employ a time-reduction layer to speed up inference time and propose a method of hint-based distillation for less performance degradation. Our method reduces the model to 23.8% in size and 35.9% in inference time compared to HuBERT. Also, we achieve 12.1% word error rate and 13.3% phoneme error rate on the SUPERB benchmark which is superior than prior work.
The transcription quality of automatic speech recognition (ASR) systems degrades significantly when transcribing audios coming from unseen domains. We propose an unsupervised error correction method for unsupervised ASR domain adaption, aiming to recover transcription errors caused by domain mismatch. Unlike existing correction methods that rely on transcribed audios for training, our approach requires only unlabeled data of the target domains in which a pseudo-labeling technique is applied to generate correction training samples. To reduce over-fitting to the pseudo data, we also propose an encoder-decoder correction model that can take into account additional information such as dialogue context and acoustic features. Experiment results show that our method obtains a significant word error rate (WER) reduction over non-adapted ASR systems. The correction model can also be applied on top of other adaptation approaches to bring an additional improvement of 10% relatively.
Digital cryptocurrencies such as Bitcoin have exploded in recent years in both popularity and value. By their novelty, cryptocurrencies tend to be both volatile and highly speculative. The capricious nature of these coins is helped facilitated by social media networks such as Twitter. However, not everyone's opinion matters equally, with most posts garnering little to no attention. Additionally, the majority of tweets are retweeted from popular posts. We must determine whose opinion matters and the difference between influential and non-influential users. This study separates these two groups and analyzes the differences between them. It uses Hypertext-induced Topic Selection (HITS) algorithm, which segregates the dataset based on influence. Topic modeling is then employed to uncover differences in each group's speech types and what group may best represent the entire community. We found differences in language and interest between these two groups regarding Bitcoin and that the opinion leaders of Twitter are not aligned with the majority of users. There were 2559 opinion leaders (0.72% of users) who accounted for 80% of the authority and the majority (99.28%) users for the remaining 20% out of a total of 355,139 users.
Recent advances in neural text-to-speech research have been dominated by two-stage pipelines utilizing low-level intermediate speech representation such as mel-spectrograms. However, such predetermined features are fundamentally limited, because they do not allow to exploit the full potential of a data-driven approach through learning hidden representations. For this reason, several end-to-end methods have been proposed. However, such models are harder to train and require a large number of high-quality recordings with transcriptions. Here, we propose WavThruVec - a two-stage architecture that resolves the bottleneck by using high-dimensional Wav2Vec 2.0 embeddings as intermediate speech representation. Since these hidden activations provide high-level linguistic features, they are more robust to noise. That allows us to utilize annotated speech datasets of a lower quality to train the first-stage module. At the same time, the second-stage component can be trained on large-scale untranscribed audio corpora, as Wav2Vec 2.0 embeddings are time-aligned and speaker-independent. This results in an increased generalization capability to out-of-vocabulary words, as well as to a better generalization to unseen speakers. We show that the proposed model not only matches the quality of state-of-the-art neural models, but also presents useful properties enabling tasks like voice conversion or zero-shot synthesis.
In recent years, Generative Adversarial Networks (GANs) have produced significantly improved results in speech enhancement (SE) tasks. They are difficult to train, however. In this work, we introduce several improvements to the GAN training schemes, which can be applied to most GAN-based SE models. We propose using consistency loss functions, which target the inconsistency in time and time-frequency domains caused by Fourier and Inverse Fourier Transforms. We also present self-correcting optimization for training a GAN discriminator on SE tasks, which helps avoid "harmful" training directions for parts of the discriminator loss function. We have tested our proposed methods on several state-of-the-art GAN-based SE models and obtained consistent improvements, including new state-of-the-art results for the Voice Bank+DEMAND dataset.
For monaural speech enhancement, contextual information is important for accurate speech estimation. However, commonly used convolution neural networks (CNNs) are weak in capturing temporal contexts since they only build blocks that process one local neighborhood at a time. To address this problem, we learn from human auditory perception to introduce a two-stage trainable reasoning mechanism, referred as global-local dependency (GLD) block. GLD blocks capture long-term dependency of time-frequency bins both in global level and local level from the noisy spectrogram to help detecting correlations among speech part, noise part, and whole noisy input. What is more, we conduct a monaural speech enhancement network called GLD-Net, which adopts encoder-decoder architecture and consists of speech object branch, interference branch, and global noisy branch. The extracted speech feature at global-level and local-level are efficiently reasoned and aggregated in each of the branches. We compare the proposed GLD-Net with existing state-of-art methods on WSJ0 and DEMAND dataset. The results show that GLD-Net outperforms the state-of-the-art methods in terms of PESQ and STOI.
End-to-end speech-to-speech translation (S2ST) without relying on intermediate text representations is a rapidly emerging frontier of research. Recent works have demonstrated that the performance of such direct S2ST systems is approaching that of conventional cascade S2ST when trained on comparable datasets. However, in practice, the performance of direct S2ST is bounded by the availability of paired S2ST training data. In this work, we explore multiple approaches for leveraging much more widely available unsupervised and weakly-supervised speech and text data to improve the performance of direct S2ST based on Translatotron 2. With our most effective approaches, the average translation quality of direct S2ST on 21 language pairs on the CVSS-C corpus is improved by +13.6 BLEU (or +113% relatively), as compared to the previous state-of-the-art trained without additional data. The improvements on low-resource language are even more significant (+398% relatively on average). Our comparative studies suggest future research directions for S2ST and speech representation learning.
We explore the efficacy of multimodal behavioral cues for explainable prediction of personality and interview-specific traits. We utilize elementary head-motion units named kinemes, atomic facial movements termed action units and speech features to estimate these human-centered traits. Empirical results confirm that kinemes and action units enable discovery of multiple trait-specific behaviors while also enabling explainability in support of the predictions. For fusing cues, we explore decision and feature-level fusion, and an additive attention-based fusion strategy which quantifies the relative importance of the three modalities for trait prediction. Examining various long-short term memory (LSTM) architectures for classification and regression on the MIT Interview and First Impressions Candidate Screening (FICS) datasets, we note that: (1) Multimodal approaches outperform unimodal counterparts; (2) Efficient trait predictions and plausible explanations are achieved with both unimodal and multimodal approaches, and (3) Following the thin-slice approach, effective trait prediction is achieved even from two-second behavioral snippets.
Latent variable models are widely used to perform unsupervised segmentation of time series in different context such as robotics, speech recognition, and economics. One of the most widely used latent variable model is the Auto-Regressive Hidden Markov Model (ARHMM), which combines a latent mode governed by a Markov chain dynamics with a linear Auto-Regressive dynamics of the observed state. In this work, we propose two generalizations of the ARHMM. First, we propose a more general AR dynamics in Cartesian space, described as a linear combination of non-linear basis functions. Second, we propose a linear dynamics in unit quaternion space, in order to properly describe orientations. These extensions allow to describe more complex dynamics of the observed state. Although this extension is proposed for the ARHMM, it can be easily extended to other latent variable models with AR dynamics in the observed space, such as Auto-Regressive Hidden semi-Markov Models.
Speech utterances recorded under differing conditions exhibit varying degrees of confidence in their embedding estimates, i.e., uncertainty, even if they are extracted using the same neural network. This paper aims to incorporate the uncertainty estimate produced in the xi-vector network front-end with a probabilistic linear discriminant analysis (PLDA) back-end scoring for speaker verification. To achieve this we derive a posterior covariance matrix, which measures the uncertainty, from the frame-wise precisions to the embedding space. We propose a log-likelihood ratio function for the PLDA scoring with the uncertainty propagation. We also propose to replace the length normalization pre-processing technique with a length scaling technique for the application of uncertainty propagation in the back-end. Experimental results on the VoxCeleb-1, SITW test sets as well as a domain-mismatched CNCeleb1-E set show the effectiveness of the proposed techniques with 14.5%-41.3% EER reductions and 4.6%-25.3% minDCF reductions.