This paper presents an end-to-end solution for the creation of fully automated conference meeting transcripts and their machine translations into various languages. This tool has been developed at the World Intellectual Property Organization (WIPO) using in-house developed speech-to-text (S2T) and machine translation (MT) components. Beyond describing data collection and fine-tuning, resulting in a highly customized and robust system, this paper describes the architecture and evolution of the technical components as well as highlights the business impact and benefits from the user side. We also point out particular challenges in the evolution and adoption of the system and how the new approach created a new product and replaced existing established workflows in conference management documentation.
While state-of-the-art Text-to-Speech systems can generate natural speech of very high quality at sentence level, they still meet great challenges in speech generation for paragraph / long-form reading. Such deficiencies are due to i) ignorance of cross-sentence contextual information, and ii) high computation and memory cost for long-form synthesis. To address these issues, this work develops a lightweight yet effective TTS system, ContextSpeech. Specifically, we first design a memory-cached recurrence mechanism to incorporate global text and speech context into sentence encoding. Then we construct hierarchically-structured textual semantics to broaden the scope for global context enhancement. Additionally, we integrate linearized self-attention to improve model efficiency. Experiments show that ContextSpeech significantly improves the voice quality and prosody expressiveness in paragraph reading with competitive model efficiency. Audio samples are available at: https://contextspeech.github.io/demo/
Amyotrophic lateral sclerosis is a fatal disease that not only affects movement, speech, and breath but also cognition. Recent studies have focused on the use of language analysis techniques to detect ALS and infer scales for monitoring functional progression. In this paper, we focused on another important aspect, cognitive impairment, which affects 35-50% of the ALS population. In an effort to reach the ALS population, which frequently exhibits mobility limitations, we implemented the digital version of the Edinburgh Cognitive and Behavioral ALS Screen (ECAS) test for the first time. This test which is designed to measure cognitive impairment was remotely performed by 56 participants from the EverythingALS Speech Study. As part of the study, participants (ALS and non-ALS) were asked to describe weekly one picture from a pool of many pictures with complex scenes displayed on their computer at home. We analyze the descriptions performed within +/- 60 days from the day the ECAS test was administered and extract different types of linguistic and acoustic features. We input those features into linear regression models to infer 5 ECAS sub-scores and the total score. Speech samples from the picture description are reliable enough to predict the ECAS subs-scores, achieving statistically significant Spearman correlation values between 0.32 and 0.51 for the model's performance using 10-fold cross-validation.
Cross-corpus speech emotion recognition (SER) seeks to generalize the ability of inferring speech emotion from a well-labeled corpus to an unlabeled one, which is a rather challenging task due to the significant discrepancy between two corpora. Existing methods, typically based on unsupervised domain adaptation (UDA), struggle to learn corpus-invariant features by global distribution alignment, but unfortunately, the resulting features are mixed with corpus-specific features or not class-discriminative. To tackle these challenges, we propose a novel Emotion Decoupling aNd Alignment learning framework (EMO-DNA) for cross-corpus SER, a novel UDA method to learn emotion-relevant corpus-invariant features. The novelties of EMO-DNA are two-fold: contrastive emotion decoupling and dual-level emotion alignment. On one hand, our contrastive emotion decoupling achieves decoupling learning via a contrastive decoupling loss to strengthen the separability of emotion-relevant features from corpus-specific ones. On the other hand, our dual-level emotion alignment introduces an adaptive threshold pseudo-labeling to select confident target samples for class-level alignment, and performs corpus-level alignment to jointly guide model for learning class-discriminative corpus-invariant features across corpora. Extensive experimental results demonstrate the superior performance of EMO-DNA over the state-of-the-art methods in several cross-corpus scenarios. Source code is available at https://github.com/Jiaxin-Ye/Emo-DNA.
In speech translation, leveraging multimodal data to improve model performance and address limitations of individual modalities has shown significant effectiveness. In this paper, we harness the complementary strengths of speech and text, which are disparate modalities. We observe three levels of modality gap between them, denoted by Modal input representation, Modal semantic, and Modal hidden states. To tackle these gaps, we propose \textbf{F}use-\textbf{S}peech-\textbf{T}ext (\textbf{FST}), a cross-modal model which supports three distinct input modalities for translation: speech, text, and fused speech-text. We leverage multiple techniques for cross-modal alignment and conduct a comprehensive analysis to assess its impact on speech translation, machine translation, and fused speech-text translation. We evaluate FST on MuST-C, GigaST, and newstest benchmark. Experiments show that the proposed FST achieves an average 34.0 BLEU on MuST-C En$\rightarrow$De/Es/Fr (vs SOTA +1.1 BLEU). Further experiments demonstrate that FST does not degrade on MT task, as observed in prior works. Instead, it yields an average improvement of 3.2 BLEU over the pre-trained MT model.
Visually grounded speech systems learn from paired images and their spoken captions. Recently, there have been attempts to utilize the visually grounded models trained from images and their corresponding text captions, such as CLIP, to improve speech-based visually grounded models' performance. However, the majority of these models only utilize the pretrained image encoder. Cascaded SpeechCLIP attempted to generate localized word-level information and utilize both the pretrained image and text encoders. Despite using both, they noticed a substantial drop in retrieval performance. We proposed Segmental SpeechCLIP which used a hierarchical segmental speech encoder to generate sequences of word-like units. We used the pretrained CLIP text encoder on top of these word-like unit representations and showed significant improvements over the cascaded variant of SpeechCLIP. Segmental SpeechCLIP directly learns the word embeddings as input to the CLIP text encoder bypassing the vocabulary embeddings. Here, we explore mapping audio to CLIP vocabulary embeddings via regularization and quantization. As our objective is to distill semantic information into the speech encoders, we explore the usage of large unimodal pretrained language models as the text encoders. Our method enables us to bridge image and text encoders e.g. DINO and RoBERTa trained with uni-modal data. Finally, we extend our framework in audio-only settings where only pairs of semantically related audio are available. Experiments show that audio-only systems perform close to the audio-visual system.
In this paper, we focus on Whisper, a recent automatic speech recognition model trained with a massive 680k hour labeled speech corpus recorded in diverse conditions. We first show an interesting finding that while Whisper is very robust against real-world background sounds (e.g., music), its audio representation is actually not noise-invariant, but is instead highly correlated to non-speech sounds, indicating that Whisper recognizes speech conditioned on the noise type. With this finding, we build a unified audio tagging and speech recognition model Whisper-AT by freezing the backbone of Whisper, and training a lightweight audio tagging model on top of it. With <1% extra computational cost, Whisper-AT can recognize audio events, in addition to spoken text, in a single forward pass.
In this paper, we present a groundbreaking paradigm for human-computer interaction that revolutionizes the traditional notion of an operating system. Within this innovative framework, user requests issued to the machine are handled by an interconnected ecosystem of generative AI models that seamlessly integrate with or even replace traditional software applications. At the core of this paradigm shift are large generative models, such as language and diffusion models, which serve as the central interface between users and computers. This pioneering approach leverages the abilities of advanced language models, empowering users to engage in natural language conversations with their computing devices. Users can articulate their intentions, tasks, and inquiries directly to the system, eliminating the need for explicit commands or complex navigation. The language model comprehends and interprets the user's prompts, generating and displaying contextual and meaningful responses that facilitate seamless and intuitive interactions. This paradigm shift not only streamlines user interactions but also opens up new possibilities for personalized experiences. Generative models can adapt to individual preferences, learning from user input and continuously improving their understanding and response generation. Furthermore, it enables enhanced accessibility, as users can interact with the system using speech or text, accommodating diverse communication preferences. However, this visionary concept raises significant challenges, including privacy, security, trustability, and the ethical use of generative models. Robust safeguards must be in place to protect user data and prevent potential misuse or manipulation of the language model. While the full realization of this paradigm is still far from being achieved, this paper serves as a starting point for envisioning this transformative potential.
Speech-to-speech translation is a typical sequence-to-sequence learning task that naturally has two directions. How to effectively leverage bidirectional supervision signals to produce high-fidelity audio for both directions? Existing approaches either train two separate models or a multitask-learned model with low efficiency and inferior performance. In this paper, we propose a duplex diffusion model that applies diffusion probabilistic models to both sides of a reversible duplex Conformer, so that either end can simultaneously input and output a distinct language's speech. Our model enables reversible speech translation by simply flipping the input and output ends. Experiments show that our model achieves the first success of reversible speech translation with significant improvements of ASR-BLEU scores compared with a list of state-of-the-art baselines.
In this paper, we introduce Libriheavy, a large-scale ASR corpus consisting of 50,000 hours of read English speech derived from LibriVox. To the best of our knowledge, Libriheavy is the largest freely-available corpus of speech with supervisions. Different from other open-sourced datasets that only provide normalized transcriptions, Libriheavy contains richer information such as punctuation, casing and text context, which brings more flexibility for system building. Specifically, we propose a general and efficient pipeline to locate, align and segment the audios in previously published Librilight to its corresponding texts. The same as Librilight, Libriheavy also has three training subsets small, medium, large of the sizes 500h, 5000h, 50000h respectively. We also extract the dev and test evaluation sets from the aligned audios and guarantee there is no overlapping speakers and books in training sets. Baseline systems are built on the popular CTC-Attention and transducer models. Additionally, we open-source our dataset creatation pipeline which can also be used to other audio alignment tasks.