In this paper, we ask whether vocal source features (pitch, shimmer, jitter, etc) can improve the performance of automatic sung speech recognition, arguing that conclusions previously drawn from spoken speech studies may not be valid in the sung speech domain. We first use a parallel singing/speaking corpus (NUS-48E) to illustrate differences in sung vs spoken voicing characteristics including pitch range, syllables duration, vibrato, jitter and shimmer. We then use this analysis to inform speech recognition experiments on the sung speech DSing corpus, using a state of the art acoustic model and augmenting conventional features with various voice source parameters. Experiments are run with three standard (increasingly large) training sets, DSing1 (15.1 hours), DSing3 (44.7 hours) and DSing30 (149.1 hours). Pitch combined with degree of voicing produces a significant decrease in WER from 38.1% to 36.7% when training with DSing1 however smaller decreases in WER observed when training with the larger more varied DSing3 and DSing30 sets were not seen to be statistically significant. Voicing quality characteristics did not improve recognition performance although analysis suggests that they do contribute to an improved discrimination between voiced/unvoiced phoneme pairs.
Nowadays, most methods in end-to-end contextual speech recognition bias the recognition process towards contextual knowledge. Since all-neural contextual biasing methods rely on phrase-level contextual modeling and attention-based relevance modeling, they may encounter confusion between similar context-specific phrases, which hurts predictions at the token level. In this work, we focus on mitigating confusion problems with fine-grained contextual knowledge selection (FineCoS). In FineCoS, we introduce fine-grained knowledge to reduce the uncertainty of token predictions. Specifically, we first apply phrase selection to narrow the range of phrase candidates, and then conduct token attention on the tokens in the selected phrase candidates. Moreover, we re-normalize the attention weights of most relevant phrases in inference to obtain more focused phrase-level contextual representations, and inject position information to better discriminate phrases or tokens. On LibriSpeech and an in-house 160,000-hour dataset, we explore the proposed methods based on a controllable all-neural biasing method, collaborative decoding (ColDec). The proposed methods provide at most 6.1% relative word error rate reduction on LibriSpeech and 16.4% relative character error rate reduction on the in-house dataset over ColDec.
Convolutional models have been widely used in multiple domains. However, most existing models only use local convolution, making the model unable to handle long-range dependency efficiently. Attention overcomes this problem by aggregating global information but also makes the computational complexity quadratic to the sequence length. Recently, Gu et al. [2021] proposed a model called S4 inspired by the state space model. S4 can be efficiently implemented as a global convolutional model whose kernel size equals the input sequence length. S4 can model much longer sequences than Transformers and achieve significant gains over SoTA on several long-range tasks. Despite its empirical success, S4 is involved. It requires sophisticated parameterization and initialization schemes. As a result, S4 is less intuitive and hard to use. Here we aim to demystify S4 and extract basic principles that contribute to the success of S4 as a global convolutional model. We focus on the structure of the convolution kernel and identify two critical but intuitive principles enjoyed by S4 that are sufficient to make up an effective global convolutional model: 1) The parameterization of the convolutional kernel needs to be efficient in the sense that the number of parameters should scale sub-linearly with sequence length. 2) The kernel needs to satisfy a decaying structure that the weights for convolving with closer neighbors are larger than the more distant ones. Based on the two principles, we propose a simple yet effective convolutional model called Structured Global Convolution (SGConv). SGConv exhibits strong empirical performance over several tasks: 1) With faster speed, SGConv surpasses S4 on Long Range Arena and Speech Command datasets. 2) When plugging SGConv into standard language and vision models, it shows the potential to improve both efficiency and performance.
Self-supervised learning (SSL) has shown tremendous success in various speech-related downstream tasks, including Automatic Speech Recognition (ASR). The output embeddings of the SSL model are treated as powerful short-time representations of the speech signal. However, in the ASR task, the main objective is to get the correct sequence of acoustic units, characters, or byte-pair encodings (BPEs). Usually, encoder-decoder architecture works exceptionally well for a sequence-to-sequence task like ASR. Therefore, in this paper, we propose a new paradigm that exploits the power of a decoder during self-supervised learning. We use Hidden Unit BERT (HuBERT) SSL framework to compute the conventional masked prediction loss for the encoder. In addition, we have introduced a decoder in the SSL framework and proposed a target preparation strategy for the decoder. Finally, we use a multitask SSL setup wherein we jointly optimize both the encoder and decoder losses. We hypothesize that the presence of a decoder in the SSL model helps it learn an acoustic unit-based language model, which might improve the performance of an ASR downstream task. We compare our proposed SSL model with HuBERT and show up to 25% relative improvement in performance on ASR by finetuning on various LibriSpeech subsets.
In this paper, we explore a method for training speech-to-speech translation tasks without any transcription or linguistic supervision. Our proposed method consists of two steps: First, we train and generate discrete representation with unsupervised term discovery with a discrete quantized autoencoder. Second, we train a sequence-to-sequence model that directly maps the source language speech to the target language's discrete representation. Our proposed method can directly generate target speech without any auxiliary or pre-training steps with a source or target transcription. To the best of our knowledge, this is the first work that performed pure speech-to-speech translation between untranscribed unknown languages.
This paper describes the systems submitted to IWSLT 2021 by the Volctrans team. We participate in the offline speech translation and text-to-text simultaneous translation tracks. For offline speech translation, our best end-to-end model achieves 8.1 BLEU improvements over the benchmark on the MuST-C test set and is even approaching the results of a strong cascade solution. For text-to-text simultaneous translation, we explore the best practice to optimize the wait-k model. As a result, our final submitted systems exceed the benchmark at around 7 BLEU on the same latency regime. We will publish our code and model to facilitate both future research works and industrial applications.
Voice-driven services (VDS) are being used in a variety of applications ranging from smart home control to payments using digital assistants. The input to such services is often captured via an open voice channel, e.g., using a microphone, in an unsupervised setting. One of the key operational security requirements in such setting is the freshness of the input speech. We present AEOLUS, a security overlay that proactively embeds a dynamic acoustic nonce at the time of user interaction, and detects the presence of the embedded nonce in the recorded speech to ensure freshness. We demonstrate that acoustic nonce can (i) be reliably embedded and retrieved, and (ii) be non-disruptive (and even imperceptible) to a VDS user. Optimal parameters (acoustic nonce's operating frequency, amplitude, and bitrate) are determined for (i) and (ii) from a practical perspective. Experimental results show that AEOLUS yields 0.5% FRR at 0% FAR for speech re-use prevention upto a distance of 4 meters in three real-world environments with different background noise levels. We also conduct a user study with 120 participants, which shows that the acoustic nonce does not degrade overall user experience for 94.16% of speech samples, on average, in these environments. AEOLUS can therefore be used in practice to prevent speech re-use and ensure the freshness of speech input.
Citizen-generated counter speech is a promising way to fight hate speech and promote peaceful, non-polarized discourse. However, there is a lack of large-scale longitudinal studies of its effectiveness for reducing hate speech. We investigate the effectiveness of counter speech using several different macro- and micro-level measures of over 180,000 political conversations that took place on German Twitter over four years. We report on the dynamic interactions of hate and counter speech over time and provide insights into whether, as in `classic' bullying situations, organized efforts are more effective than independent individuals in steering online discourse. Taken together, our results build a multifaceted picture of the dynamics of hate and counter speech online. They suggest that organized hate speech produced changes in the public discourse. Counter speech, especially when organized, could help in curbing hate speech in online discussions.
Given a speaker's speech, it is interesting to see if it is possible to generate this speaker's face. One main challenge in this task is to alleviate the natural mismatch between face and speech. To this end, in this paper, we propose a novel Attention-based Residual Speech Portrait Model (AR-SPM) by introducing the ideal of the residual into a hybrid encoder-decoder architecture, where face prior features are merged with the output of speech encoder to form the final face feature. In particular, we innovatively establish a tri-item loss function, which is a weighted linear combination of the L2-norm, L1-norm and negative cosine loss, to train our model by comparing the final face feature and true face feature. Evaluation on AVSpeech dataset shows that our proposed model accelerates the convergence of training, outperforms the state-of-the-art in terms of quality of the generated face, and achieves superior recognition accuracy of gender and age compared with the ground truth.
Models of acoustic word embeddings (AWEs) learn to map variable-length spoken word segments onto fixed-dimensionality vector representations such that different acoustic exemplars of the same word are projected nearby in the embedding space. In addition to their speech technology applications, AWE models have been shown to predict human performance on a variety of auditory lexical processing tasks. Current AWE models are based on neural networks and trained in a bottom-up approach that integrates acoustic cues to build up a word representation given an acoustic or symbolic supervision signal. Therefore, these models do not leverage or capture high-level lexical knowledge during the learning process. In this paper, we propose a multi-task learning model that incorporates top-down lexical knowledge into the training procedure of AWEs. Our model learns a mapping between the acoustic input and a lexical representation that encodes high-level information such as word semantics in addition to bottom-up form-based supervision. We experiment with three languages and demonstrate that incorporating lexical knowledge improves the embedding space discriminability and encourages the model to better separate lexical categories.