This paper studies a novel pre-training technique with unpaired speech data, Speech2C, for encoder-decoder based automatic speech recognition (ASR). Within a multi-task learning framework, we introduce two pre-training tasks for the encoder-decoder network using acoustic units, i.e., pseudo codes, derived from an offline clustering model. One is to predict the pseudo codes via masked language modeling in encoder output, like HuBERT model, while the other lets the decoder learn to reconstruct pseudo codes autoregressively instead of generating textual scripts. In this way, the decoder learns to reconstruct original speech information with codes before learning to generate correct text. Comprehensive experiments on the LibriSpeech corpus show that the proposed Speech2C can relatively reduce the word error rate (WER) by 19.2% over the method without decoder pre-training, and also outperforms significantly the state-of-the-art wav2vec 2.0 and HuBERT on fine-tuning subsets of 10h and 100h.
The task of emotion recognition in conversations (ERC) benefits from the availability of multiple modalities, as offered, for example, in the video-based MELD dataset. However, only a few research approaches use both acoustic and visual information from the MELD videos. There are two reasons for this: First, label-to-video alignments in MELD are noisy, making those videos an unreliable source of emotional speech data. Second, conversations can involve several people in the same scene, which requires the detection of the person speaking the utterance. In this paper we demonstrate that by using recent automatic speech recognition and active speaker detection models, we are able to realign the videos of MELD, and capture the facial expressions from uttering speakers in 96.92% of the utterances provided in MELD. Experiments with a self-supervised voice recognition model indicate that the realigned MELD videos more closely match the corresponding utterances offered in the dataset. Finally, we devise a model for emotion recognition in conversations trained on the face and audio information of the MELD realigned videos, which outperforms state-of-the-art models for ERC based on vision alone. This indicates that active speaker detection is indeed effective for extracting facial expressions from the uttering speakers, and that faces provide more informative visual cues than the visual features state-of-the-art models have been using so far.
Denoising diffusion probabilistic models (DDPMs) are expressive generative models that have been used to solve a variety of speech synthesis problems. However, because of their high sampling costs, DDPMs are difficult to use in real-time speech processing applications. In this paper, we introduce DiffGAN-TTS, a novel DDPM-based text-to-speech (TTS) model achieving high-fidelity and efficient speech synthesis. DiffGAN-TTS is based on denoising diffusion generative adversarial networks (GANs), which adopt an adversarially-trained expressive model to approximate the denoising distribution. We show with multi-speaker TTS experiments that DiffGAN-TTS can generate high-fidelity speech samples within only 4 denoising steps. We present an active shallow diffusion mechanism to further speed up inference. A two-stage training scheme is proposed, with a basic TTS acoustic model trained at stage one providing valuable prior information for a DDPM trained at stage two. Our experiments show that DiffGAN-TTS can achieve high synthesis performance with only 1 denoising step.
In this paper we present two datasets for Tamasheq, a developing language mainly spoken in Mali and Niger. These two datasets were made available for the IWSLT 2022 low-resource speech translation track, and they consist of collections of radio recordings from the Studio Kalangou (Niger) and Studio Tamani (Mali) daily broadcast news. We share (i) a massive amount of unlabeled audio data (671 hours) in five languages: French from Niger, Fulfulde, Hausa, Tamasheq and Zarma, and (ii) a smaller parallel corpus of audio recordings (17 hours) in Tamasheq, with utterance-level translations in the French language. All this data is shared under the Creative Commons BY-NC-ND 3.0 license. We hope these resources will inspire the speech community to develop and benchmark models using the Tamasheq language.
Many archival recordings of speech from endangered languages remain unannotated and inaccessible to community members and language learning programs. One bottleneck is the time-intensive nature of annotation. An even narrower bottleneck occurs for recordings with access constraints, such as language that must be vetted or filtered by authorised community members before annotation can begin. We propose a privacy-preserving workflow to widen both bottlenecks for recordings where speech in the endangered language is intermixed with a more widely-used language such as English for meta-linguistic commentary and questions (e.g. What is the word for 'tree'?). We integrate voice activity detection (VAD), spoken language identification (SLI), and automatic speech recognition (ASR) to transcribe the metalinguistic content, which an authorised person can quickly scan to triage recordings that can be annotated by people with lower levels of access. We report work-in-progress processing 136 hours archival audio containing a mix of English and Muruwari. Our collaborative work with the Muruwari custodian of the archival materials show that this workflow reduces metalanguage transcription time by 20% even given only minimal amounts of annotated training data: 10 utterances per language for SLI and 39 minutes of the English for ASR.
Mixup is a popular data augmentation technique for training deep neural networks where additional samples are generated by linearly interpolating pairs of inputs and their labels. This technique is known to improve the generalization performance in many learning paradigms and applications. In this work, we first analyze Mixup and show that it implicitly regularizes infinitely many directional derivatives of all orders. We then propose a new method to improve Mixup based on the novel insight. To demonstrate the effectiveness of the proposed method, we conduct experiments across various domains such as images, tabular data, speech, and graphs. Our results show that the proposed method improves Mixup across various datasets using a variety of architectures, for instance, exhibiting an improvement over Mixup by 0.8% in ImageNet top-1 accuracy.
Speaker embedding extractors (EEs), which map input audio to a speaker discriminant latent space, are of paramount importance in speaker diarisation. However, there are several challenges when adopting EEs for diarisation, from which we tackle two key problems. First, the evaluation is not straightforward because the features required for better performance differ between speaker verification and diarisation. We show that better performance on widely adopted speaker verification evaluation protocols does not lead to better diarisation performance. Second, embedding extractors have not seen utterances in which multiple speakers exist. These inputs are inevitably present in speaker diarisation because of overlapped speech and speaker changes; they degrade the performance. To mitigate the first problem, we generate speaker verification evaluation protocols that mimic the diarisation scenario better. We propose two data augmentation techniques to alleviate the second problem, making embedding extractors aware of overlapped speech or speaker change input. One technique generates overlapped speech segments, and the other generates segments where two speakers utter sequentially. Extensive experimental results using three state-of-the-art speaker embedding extractors demonstrate that both proposed approaches are effective.
This work focuses on modelling a speaker's accent that does not have a dedicated text-to-speech (TTS) frontend, including a grapheme-to-phoneme (G2P) module. Prior work on modelling accents assumes a phonetic transcription is available for the target accent, which might not be the case for low-resource, regional accents. In our work, we propose an approach whereby we first augment the target accent data to sound like the donor voice via voice conversion, then train a multi-speaker multi-accent TTS model on the combination of recordings and synthetic data, to generate the donor's voice speaking in the target accent. Throughout the procedure, we use a TTS frontend developed for the same language but a different accent. We show qualitative and quantitative analysis where the proposed strategy achieves state-of-the-art results compared to other generative models. Our work demonstrates that low resource accents can be modelled with relatively little data and without developing an accent-specific TTS frontend. Audio samples of our model converting to multiple accents are available on our web page.
Distributed representations of words encode lexical semantic information, but how is that information encoded in word embeddings? Focusing on the skip-gram with negative-sampling method, we show theoretically and experimentally that the squared norm of word embedding encodes the information gain defined by the Kullback-Leibler divergence of the co-occurrence distribution of a word to the unigram distribution of the corpus. Furthermore, through experiments on tasks of keyword extraction, hypernym prediction, and part-of-speech discrimination, we confirmed that the KL divergence and the squared norm of embedding work as a measure of the informativeness of a word provided that the bias caused by word frequency is adequately corrected.
Self-Supervised Learning (SSL) from speech data has produced models that have achieved remarkable performance in many tasks, and that are known to implicitly represent many aspects of information latently present in speech signals. However, relatively little is known about the suitability of such models for prosody-related tasks or the extent to which they encode prosodic information. We present a new evaluation framework, SUPERB-prosody, consisting of three prosody-related downstream tasks and two pseudo tasks. We find that 13 of the 15 SSL models outperformed the baseline on all the prosody-related tasks. We also show good performance on two pseudo tasks: prosody reconstruction and future prosody prediction. We further analyze the layerwise contributions of the SSL models. Overall we conclude that SSL speech models are highly effective for prosody-related tasks.