Audio-visual speech synthesis (AVSS) aims to produce an audio-visual stream that conveys a target speaker's speech. In this study, the AVSS system takes the input speech of a source speaker and generates the audio-visual stream of the target speaker while preserving the linguistic content of the source speech. The process involves two main components: voice conversion (VC), which adapts the vocal features from the source to the target speaker, and audio-visual synthesis (AVS), which generates the synchronized audio-visual stream from the transformed speech. This paper presents a novel generative framework based on multi-discriminative learning to enhance the realism and quality of AVSS outputs. The proposed approach integrates multiple discriminators, including capsule networks, co-occurrence neural networks, and vision transformers (ViTs), within the VC model to leverage their unique strengths in capturing diverse speech features. Additionally, the AVS model incorporates a co-occurrence neural network to improve video quality and achieve better temporal alignment between audio and visual data. Experimental evaluations on standard benchmarks demonstrate that the proposed method achieves significant improvements in both audio and video quality, offering a substantial advancement in AVSS technology.

In recent years, several text-to-speech systems have been proposed to synthesize natural speech in zero-shot, few-shot, and low-resource scenarios. However, these methods typically require training with data from many different speakers. The speech quality across the speaker set typically is diverse and imposes an upper limit on the quality achievable for the low-resource speaker. In the current work, we achieve high-quality speech synthesis using as little as five minutes of speech from the desired speaker by augmenting the low-resource speaker data with noise and employing multiple sampling techniques during training. Our method requires only four high-quality, high-resource speakers, which are easy to obtain and use in practice. Our low-complexity method achieves improved speaker similarity compared to the state-of-the-art zero-shot method HierSpeech++ and the recent low-resource method AdapterMix while maintaining comparable naturalness. Our proposed approach can also reduce the data requirements for speech synthesis for new speakers and languages.

2470

4958

In recent years, the quality of text-to-speech (TTS) synthesis vastly improved due to deep-learning techniques, with parallel architectures, in particular, providing excellent synthesis quality at fast inference. Training these models usually requires speech recordings, corresponding phoneme-level transcripts, and the temporal alignment of each phoneme to the utterances. Since manually creating such fine-grained alignments requires expert knowledge and is time-consuming, it is common practice to estimate them using automatic speech–phoneme alignment methods. In the literature, either the estimation methods’ accuracy or their impact on the TTS system’s synthesis quality is evaluated. In this study, we perform experiments with five state-of-the-art speech–phoneme aligners and evaluate their output with objective and subjective measures. As our main result, we show that small alignment errors (below 75 ms error) do not decrease the synthesis quality, which implies that the alignment error may not be the crucial factor when choosing an aligner for TTS training.

Attention-based Transformer models have been increasingly employed for automatic music generation. To condition the generation process of such a model with a user-specified sequence, a popular approach is to take that conditioning sequence as a priming sequence and ask a Transformer decoder to generate a continuation. However, thisprompt-based conditioningcannot guarantee that the conditioning sequence would develop or even simply repeat itself in the generated continuation. In this paper, we propose an alternative conditioning approach, calledtheme-based conditioning, that explicitly trains the Transformer to treat the conditioning sequence as a thematic material that has to manifest itself multiple times in its generation result. This is achieved with two main technical contributions. First, we propose a deep learning-based approach that uses contrastive representation learning and clustering to automatically retrieve thematic materials from music pieces in the training data. Second, we propose a novel gated parallel attention module to be used in a sequence-to-sequence (seq2seq) encoder/decoder architecture to more effectively account for a given conditioning thematic material in the generation process of the Transformer decoder. We report on objective and subjective evaluations of variants of the proposed Theme Transformer and the conventional prompt-based baseline, showing that our best model can generate, to some extent, polyphonic pop piano music with repetition and plausible variations of a given condition.

This paper deals with a scoreaudio music retrieval task where the aim is to find relevant audio recordings of Western classical music, given a short monophonic musical theme in symbolic notation as a query. Strategies for comparing score and audio data are often based on a common mid-level representation, such as chroma features, which capture melodic and harmonic properties. Recent studies demonstrated the effectiveness of deep neural networks that learn task-specific mid-level representations. Usually, such supervised learning approaches require scoreaudio pairs where individual note events of the score are aligned to the corresponding time positions of the audio excerpt. However, in practice, it is tedious to generate such strongly aligned training pairs. As one contribution of this paper, we show how to apply the Connectionist Temporal Classification (CTC) loss in the training procedure, which only uses weakly aligned training pairs. In such a pair, only the time positions of the beginning and end of a theme occurrence are annotated in an audio recording, rather than requiring local alignment annotations. We evaluate the resulting features in our theme retrieval scenario and show that they improve the state of the art for this task. As a main result, we demonstrate that with the CTC-based training procedure using weakly annotated data, we can achieve results almost as good as with strongly annotated data. Furthermore, we assess our chroma features in depth by inspecting their temporal smoothness or granularity as an important property and by analyzing the impact of different degrees of musical complexity on the features.

In this paper, we consider a cross-modal retrieval scenario of Western classical music. Given a short monophonic musical theme in symbolic notation as query, the objective is to find relevant audio recordings in a database. A major challenge of this retrieval task is the possible difference in the degree of polyphony between the monophonic query and the music recordings. Previous studies for popular music addressed this issue by performing the cross-modal comparison based on predominant melodies extracted from the recordings. For Western classical music, however, this approach is problematic since the underlying assumption of a single pre-dominant melody is often violated. Instead of extracting the melody explicitly, another strategy is to perform the cross-modal comparison directly on the basis of melody-enhanced salience representations. As the main contribution of this paper, we evaluate several conceptually different salience representations for our cross-modal retrieval scenario. Our extensive experimental results, which have been made available on a website, comprise more than 2000 musical themes and 100 hours of audio recordings.