Simultaneous Speech-to-Text translation serves a critical role in real-time crosslingual communication. Despite the advancements in recent years, challenges remain in achieving stability in the translation process, a concern primarily manifested in the flickering of partial results. In this paper, we propose a novel revision-controllable method designed to address this issue. Our method introduces an allowed revision window within the beam search pruning process to screen out candidate translations likely to cause extensive revisions, leading to a substantial reduction in flickering and, crucially, providing the capability to completely eliminate flickering. The experiments demonstrate the proposed method can significantly improve the decoding stability without compromising substantially on the translation quality.
The application of self-supervision to speech representation learning has garnered significant interest in recent years, due to its scalability to large amounts of unlabeled data. However, much progress, both in terms of pre-training and downstream evaluation, has remained concentrated in monolingual models that only consider English. Few models consider other languages, and even fewer consider indigenous ones. In our submission to the New Language Track of the ASRU 2023 ML-SUPERB Challenge, we present an ASR corpus for Quechua, an indigenous South American Language. We benchmark the efficacy of large SSL models on Quechua, along with 6 other indigenous languages such as Guarani and Bribri, on low-resource ASR. Our results show surprisingly strong performance by state-of-the-art SSL models, showing the potential generalizability of large-scale models to real-world data.
The integration of large language models (LLMs) with social robots has emerged as a promising avenue for enhancing human-robot interactions at a time when news reports generated by artificial intelligence (AI) are gaining in credibility. This integration is expected to intensify and become a more productive resource for journalism, media, communication, and education. In this paper a novel system is proposed that integrates AI's generative pretrained transformer (GPT) model with the Pepper robot, with the aim of improving the robot's natural language understanding and response generation capabilities for enhanced social interactions. By leveraging GPT's powerful language processing capabilities, this system offers a comprehensive pipeline that incorporates voice input recording, speech-to-text transcription, context analysis, and text-to-speech synthesis action generation. The Pepper robot is enabled to comprehend user queries, generate informative responses with general knowledge, maintain contextually relevant conversations, and act as a more domain-oriented news reporter. It is also linked with a news resource and powered with a Google search capability. To evaluate the performance of the framework, experiments were conducted involving a set of diverse questions. The robot's responses were assessed on the basis of eight criteria, including relevance, context, and fluency. Despite some identified limitations, this system contributes to the field of journalism and human-robot interaction by showcasing the potential of integrating LLMs with social robots. The proposed framework opens up opportunities for improving the conversational capabilities of robots, enabling interactions that are smoother, more engaging, and more context aware.
In this paper, we address the unsupervised speech enhancement problem based on recurrent variational autoencoder (RVAE). This approach offers promising generalization performance over the supervised counterpart. Nevertheless, the involved iterative variational expectation-maximization (VEM) process at test time, which relies on a variational inference method, results in high computational complexity. To tackle this issue, we present efficient sampling techniques based on Langevin dynamics and Metropolis-Hasting algorithms, adapted to the EM-based speech enhancement with RVAE. By directly sampling from the intractable posterior distribution within the EM process, we circumvent the intricacies of variational inference. We conduct a series of experiments, comparing the proposed methods with VEM and a state-of-the-art supervised speech enhancement approach based on diffusion models. The results reveal that our sampling-based algorithms significantly outperform VEM, not only in terms of computational efficiency but also in overall performance. Furthermore, when compared to the supervised baseline, our methods showcase robust generalization performance in mismatched test conditions.
This paper describes a system developed for the GENEA (Generation and Evaluation of Non-verbal Behaviour for Embodied Agents) Challenge 2023. Our solution builds on an existing diffusion-based motion synthesis model. We propose a contrastive speech and motion pretraining (CSMP) module, which learns a joint embedding for speech and gesture with the aim to learn a semantic coupling between these modalities. The output of the CSMP module is used as a conditioning signal in the diffusion-based gesture synthesis model in order to achieve semantically-aware co-speech gesture generation. Our entry achieved highest human-likeness and highest speech appropriateness rating among the submitted entries. This indicates that our system is a promising approach to achieve human-like co-speech gestures in agents that carry semantic meaning.
Large Language Models (LLMs) demonstrate impressive capabilities, yet interaction with these models is mostly facilitated through text. Using Text-To-Speech to synthesize LLM outputs typically results in notable latency, which is impractical for fluent voice conversations. We propose LLM2Speech, an architecture to synthesize speech while text is being generated by an LLM which yields significant latency reduction. LLM2Speech mimics the predictions of a non-streaming teacher model while limiting the exposure to future context in order to enable streaming. It exploits the hidden embeddings of the LLM, a by-product of the text generation that contains informative semantic context. Experimental results show that LLM2Speech maintains the teacher's quality while reducing the latency to enable natural conversations.
In indoor scenes, reverberation is a crucial factor in degrading the perceived quality and intelligibility of speech. In this work, we propose a generative dereverberation method. Our approach is based on a probabilistic model utilizing a recurrent variational auto-encoder (RVAE) network and the convolutive transfer function (CTF) approximation. Different from most previous approaches, the output of our RVAE serves as the prior of the clean speech. And our target is the maximum a posteriori (MAP) estimation of clean speech, which is achieved iteratively through the expectation maximization (EM) algorithm. The proposed method integrates the capabilities of network-based speech prior modelling and CTF-based observation modelling. Experiments on single-channel speech dereverberation show that the proposed generative method noticeably outperforms the advanced discriminative networks.
Multi-channel speech enhancement utilizes spatial information from multiple microphones to extract the target speech. However, most existing methods do not explicitly model spatial cues, instead relying on implicit learning from multi-channel spectra. To better leverage spatial information, we propose explicitly incorporating spatial modeling by applying spherical harmonic transforms (SHT) to the multi-channel input. In detail, a hierarchical framework is introduced whereby lower order harmonics capturing broader spatial patterns are estimated first, then combined with higher orders to recursively predict finer spatial details. Experiments on TIMIT demonstrate the proposed method can effectively recover target spatial patterns and achieve improved performance over baseline models, using fewer parameters and computations. Explicitly modeling spatial information hierarchically enables more effective multi-channel speech enhancement.
Diffusion models have shown promising results in speech enhancement, using a task-adapted diffusion process for the conditional generation of clean speech given a noisy mixture. However, at test time, the neural network used for score estimation is called multiple times to solve the iterative reverse process. This results in a slow inference process and causes discretization errors that accumulate over the sampling trajectory. In this paper, we address these limitations through a two-stage training approach. In the first stage, we train the diffusion model the usual way using the generative denoising score matching loss. In the second stage, we compute the enhanced signal by solving the reverse process and compare the resulting estimate to the clean speech target using a predictive loss. We show that using this second training stage enables achieving the same performance as the baseline model using only 5 function evaluations instead of 60 function evaluations. While the performance of usual generative diffusion algorithms drops dramatically when lowering the number of function evaluations (NFEs) to obtain single-step diffusion, we show that our proposed method keeps a steady performance and therefore largely outperforms the diffusion baseline in this setting and also generalizes better than its predictive counterpart.
This work explores the degree to which grammar acquisition is driven by language `simplicity' and the source modality (speech vs. text) of data. Using BabyBERTa as a probe, we find that grammar acquisition is largely driven by exposure to speech data, and in particular through exposure to two of the BabyLM training corpora: AO-Childes and Open Subtitles. We arrive at this finding by examining various ways of presenting input data to our model. First, we assess the impact of various sequence-level complexity based curricula. We then examine the impact of learning over `blocks' -- covering spans of text that are balanced for the number of tokens in each of the source corpora (rather than number of lines). Finally, we explore curricula that vary the degree to which the model is exposed to different corpora. In all cases, we find that over-exposure to AO-Childes and Open Subtitles significantly drives performance. We verify these findings through a comparable control dataset in which exposure to these corpora, and speech more generally, is limited by design. Our findings indicate that it is not the proportion of tokens occupied by high-utility data that aids acquisition, but rather the proportion of training steps assigned to such data. We hope this encourages future research into the use of more developmentally plausible linguistic data (which tends to be more scarce) to augment general purpose pre-training regimes.