In this letter, we propose a novel speaker conditioning technique that leverages a variable-length reference embedding sequence for flow-based text-to-speech (TTS) architecture in the context of zero-shot multi-speaker text-to-speech (ZSM-TTS). Unlike conventional ZSM-TTS methods, which usually rely on a single fixed-dimensional vector to represent the entire reference speech, our approach aims to extract variable-length embedding sequence for a more flexible and efficient conditioning. We enhance the current affine coupling function in flow-based TTS architecture by introducing an attentive speaker conditioning. This allows a local variation of the speaker conditioning. Our experiments demonstrate the effectiveness of the proposed method, highlighting improvements in terms of speaker similarity, speech naturalness, and speech intelligibility compared to the baseline methods.

Though neural text-to-speech (TTS) models show remarkable performance, they still require a large amount ofpaired dataset, which is expensive to collect. The heavy demand for collecting paired datasets makes the TTS models support only a small number of speakers and languages. To address this problem, we introduce a transfer learning framework for multi-lingual, zero-shot multi-speaker, and low-resource TTS. Firstly, we pretrain our model in an unsupervised manner with a multi-lingual multi-speaker speech-only dataset by leveraging the self-supervised speech representations as intermediate linguistic representations. Given this pretrained linguistic information, we then apply a supervised learning technique to the TTS model with a small amount of paired dataset. The pretrained linguistic representations extracted from the large-scale speech-only dataset facilitate phoneme-to-linguistic feature matching, which provides good guidance for supervised learning with a limited amount of labeled data. We evaluate the performance of our proposed model in low-resource, multi-lingual, and zero-shot multi-speaker TTS tasks. The experimental results demonstrate that our proposed method outperforms the baseline in terms of naturalness, intelligibility, and speaker similarity.

We introduce a text-to-speech (TTS) framework based on a neural transducer. We use discretized semantic tokens acquired from wav2vec 2.0 embeddings, which makes it easy to adopt a neural transducer for the TTS framework enjoying its monotonic alignment constraints. The proposed model first generates aligned semantic tokens using the neural transducer, then synthesizes a speech sample from the semantic tokens using a non-autoregressive (NAR) speech generator. This decoupled framework alleviates the training complexity of TTS and allows each stage to focus on 1) linguistic and alignment modeling and 2) fine-grained acoustic modeling, respectively. Experimental results on the zero-shot adaptive TTS show that the proposed model exceeds the baselines in speech quality and speaker similarity via objective and subjective measures. We also investigate the inference speed and prosody controllability of our proposed model, showing the potential of the neural transducer for TTS frameworks.

Training a text-to-speech (TTS) model requires a large scale text labeled speech corpus, which is troublesome to collect. In this paper, we propose a transfer learning framework for TTS that utilizes a large amount of unlabeled speech dataset for pre-training. By leveraging wav2vec2.0 representation, unlabeled speech can highly improve performance, especially in the lack of labeled speech. We also extend the proposed method to zero-shot multi-speaker TTS (ZS-TTS). The experimental results verify the effectiveness of the proposed method in terms of naturalness, intelligibility, and speaker generalization. We highlight that the single speaker TTS model fine-tuned on the only 10 minutes of labeled dataset outperforms the other baselines, and the ZS-TTS model fine-tuned on the only 30 minutes of single speaker dataset can generate the voice of the arbitrary speaker, by pre-training on unlabeled multi-speaker speech corpus.

In this letter, we propose a multivariate information minimization method that disentangles three or more latent representations. We show that control factors can be disentangled by minimizing interactive dependency, which can be expressed as a sum of mutual information upper bound terms. Since the upper bound estimate converges from the early training stage, there is little performance degradation due to auxiliary loss. The proposed technique is applied to train a text-to-speech synthesizer with multi-lingual, multi-speaker, and multi-style corpora. Subjective listening tests validate that the proposed method can improve the synthesizer in terms of quality as well as controllability.

Zero-shot multi-speaker text-to-speech (ZSM-TTS) models aim to generate a speech sample with the voice characteristic of an unseen speaker. The main challenge of ZSM-TTS is to increase the overall speaker similarity for unseen speakers. One of the most successful speaker conditioning methods for flow-based multi-speaker text-to-speech (TTS) models is to utilize the functions which predict the scale and bias parameters of the affine coupling layers according to the given speaker embedding vector. In this letter, we improve on the previous speaker conditioning method by introducing a speaker-normalized affine coupling (SNAC) layer which allows for unseen speaker speech synthesis in a zero-shot manner leveraging a normalization-based conditioning technique. The newly designed coupling layer explicitly normalizes the input by the parameters predicted from a speaker embedding vector while training, enabling an inverse process of denormalizing for a new speaker embedding at inference. The proposed conditioning scheme yields the state-of-the-art performance in terms of the speech quality and speaker similarity in a ZSM-TTS setting.

Although neural text-to-speech (TTS) models have attracted a lot of attention and succeeded in generating human-like speech, there is still room for improvements to its naturalness and architectural efficiency.In this work, we propose a novel nonautoregressive TTS model, namely Diff-TTS, which achieves highly natural and efficient speech synthesis.Given the text, Diff-TTS exploits a denoising diffusion framework to transform the noise signal into a mel-spectrogram via diffusion time steps.In order to learn the mel-spectrogram distribution conditioned on the text, we present a likelihood-based optimization method for TTS.Furthermore, to boost up the inference speed, we leverage the accelerated sampling method that allows Diff-TTS to generate raw waveforms much faster without significantly degrading perceptual quality.Through experiments, we verified that Diff-TTS generates 28 times faster than the real-time with a single NVIDIA 2080Ti GPU.

As recent text-to-speech (TTS) systems have been rapidly improved in speech quality and generation speed, many researchers now focus on a more challenging issue: expressive TTS. To control speaking styles, existing expressive TTS models use categorical style index or reference speech as style input. In this work, we propose StyleTagging-TTS (ST-TTS), a novel expressive TTS model that utilizes a style tag written in natural language. Using a style-tagged TTS dataset and a pre-trained language model, we modeled the relationship between linguistic embedding and speaking style domain, which enables our model to work even with style tags unseen during training. As style tag is written in natural language, it can control speaking style in a more intuitive, interpretable, and scalable way compared with style index or reference speech. In addition, in terms of model architecture, we propose an efficient non-autoregressive (NAR) TTS architecture with single-stage training. The experimental result shows that ST-TTS outperforms the existing expressive TTS model, Tacotron2-GST in speech quality and expressiveness.

Recent end-to-end (e2e) speech synthesis systems usually employ attention techniques to align an input text sequence against a mel-spectrogram sequence. Attention-based e2e approach has shown state-of-the-art performance in speech synthesis. However, generating stable and robust attention alignment to avoid some serious failures such as repeating, missing, and mumbling phones is still an ongoing challenge. In order to mitigate these alignment failures, we propose a novel attention method called memory attention for e2e speech synthesis, which is inspired by the gating mechanism of the long-short term memory (LSTM). Leveraging the sequence modeling power of the gating techniques, memory attention can produce a stable alignment by controlling the amount of content-based and location-based information. For performance evaluation, we compared our proposed memory attention algorithm with various conventional attention techniques in single speaker and emotional speech synthesis scenarios. From the experimental results, we conclude that memory attention can robustly generate various stylish speech.

In this paper, we propose a variational recurrent neural network (VRNN) based method for modeling and generating speech parameter sequences. In recent years. the performance of speech synthesis systems has been improved over conventional techniques thanks to deep learning-based acoustic models. Among the popular deep learning techniques, recurrent neural networks (RNNs) has been successful in modeling time-dependent sequential data efficiently. However, due to the deterministic nature of RNNs prediction, such models do not reflect the full complexity of highly structured data, like natural speech. In this regard, we propose adversarially trained variational recurrent neural network (AdVRNN) which use VRNN to better represent the variability of natural speech for acoustic modeling in speech synthesis. Also, we apply adversarial learning scheme in training AdVRNN to overcome oversmoothing problem. We conducted comparative experiments for the proposed VRNN with the conventional gated recurrent unit which is one of RNNs, for speech synthesis system. It is shown that the proposed AdVRNN based method performed better than the conventional GRU technique.