While FastSpeech2 aims to integrate aspects of speech such as pitch, energy, and duration as conditional inputs, it still leaves scope for richer representations. As a part of this work, we leverage representations from various Self-Supervised Learning (SSL) models to enhance the quality of the synthesized speech. In particular, we pass the FastSpeech2 encoder's length-regulated outputs through a series of encoder layers with the objective of reconstructing the SSL representations. In the SALTTS-parallel implementation, the representations from this second encoder are used for an auxiliary reconstruction loss with the SSL features. The SALTTS-cascade implementation, however, passes these representations through the decoder in addition to having the reconstruction loss. The richness of speech characteristics from the SSL features reflects in the output speech quality, with the objective and subjective evaluation measures of the proposed approach outperforming the baseline FastSpeech2.
Diffusion model-based speech enhancement has received increased attention since it can generate very natural enhanced signals and generalizes well to unseen conditions. Diffusion models have been explored for several sub-tasks of speech enhancement, such as speech denoising, dereverberation, and source separation. In this paper, we investigate their use for target speech extraction (TSE), which consists of estimating the clean speech signal of a target speaker in a mixture of multi-talkers. TSE is realized by conditioning the extraction process on a clue identifying the target speaker. We show we can realize TSE using a conditional diffusion model conditioned on the clue. Besides, we introduce ensemble inference to reduce potential extraction errors caused by the diffusion process. In experiments on Libri2mix corpus, we show that the proposed diffusion model-based TSE combined with ensemble inference outperforms a comparable TSE system trained discriminatively.
The prevailing noise-resistant and reverberation-resistant localization algorithms primarily emphasize separating and providing directional output for each speaker in multi-speaker scenarios, without association with the identity of speakers. In this paper, we present a target speaker localization algorithm with a selective hearing mechanism. Given a reference speech of the target speaker, we first produce a speaker-dependent spectrogram mask to eliminate interfering speakers' speech. Subsequently, a Long short-term memory (LSTM) network is employed to extract the target speaker's location from the filtered spectrogram. Experiments validate the superiority of our proposed method over the existing algorithms for different scale invariant signal-to-noise ratios (SNR) conditions. Specifically, at SNR = -10 dB, our proposed network LocSelect achieves a mean absolute error (MAE) of 3.55 and an accuracy (ACC) of 87.40%.
For personalized speech generation, a neural text-to-speech (TTS) model must be successfully implemented with limited data from a target speaker. To this end, the baseline TTS model needs to be amply generalized to out-of-domain data (i.e., target speaker's speech). However, approaches to address this out-of-domain generalization problem in TTS have yet to be thoroughly studied. In this work, we propose an effective pruning method for a transformer known as sparse attention, to improve the TTS model's generalization abilities. In particular, we prune off redundant connections from self-attention layers whose attention weights are below the threshold. To flexibly determine the pruning strength for searching optimal degree of generalization, we also propose a new differentiable pruning method that allows the model to automatically learn the thresholds. Evaluations on zero-shot multi-speaker TTS verify the effectiveness of our method in terms of voice quality and speaker similarity.
This paper presents an end-to-end model designed to improve automatic speech recognition (ASR) for a particular speaker in a crowded, noisy environment. The model utilizes a single-channel speech enhancement module that isolates the speaker's voice from background noise, along with an ASR module. Through this approach, the model is able to decrease the word error rate (WER) of ASR from 80% to 26.4%. Typically, these two components are adjusted independently due to variations in data requirements. However, speech enhancement can create anomalies that decrease ASR efficiency. By implementing a joint fine-tuning strategy, the model can reduce the WER from 26.4% in separate tuning to 14.5% in joint tuning.
While federated learning (FL) has recently emerged as a promising approach to train machine learning models, it is limited to only preliminary explorations in the domain of automatic speech recognition (ASR). Moreover, FL does not inherently guarantee user privacy and requires the use of differential privacy (DP) for robust privacy guarantees. However, we are not aware of prior work on applying DP to FL for ASR. In this paper, we aim to bridge this research gap by formulating an ASR benchmark for FL with DP and establishing the first baselines. First, we extend the existing research on FL for ASR by exploring different aspects of recent $\textit{large end-to-end transformer models}$: architecture design, seed models, data heterogeneity, domain shift, and impact of cohort size. With a $\textit{practical}$ number of central aggregations we are able to train $\textbf{FL models}$ that are \textbf{nearly optimal} even with heterogeneous data, a seed model from another domain, or no pre-trained seed model. Second, we apply DP to FL for ASR, which is non-trivial since DP noise severely affects model training, especially for large transformer models, due to highly imbalanced gradients in the attention block. We counteract the adverse effect of DP noise by reviving per-layer clipping and explaining why its effect is more apparent in our case than in the prior work. Remarkably, we achieve user-level ($7.2$, $10^{-9}$)-$\textbf{DP}$ (resp. ($4.5$, $10^{-9}$)-$\textbf{DP}$) with a 1.3% (resp. 4.6%) absolute drop in the word error rate for extrapolation to high (resp. low) population scale for $\textbf{FL with DP in ASR}$.
As the size of pre-trained speech recognition models increases, running these large models in low-latency or resource-constrained environments becomes challenging. In this work, we leverage pseudo-labelling to assemble a large-scale open-source dataset which we use to distill the Whisper model into a smaller variant, called Distil-Whisper. Using a simple word error rate (WER) heuristic, we select only the highest quality pseudo-labels for training. The distilled model is 5.8 times faster with 51% fewer parameters, while performing to within 1% WER on out-of-distribution test data in a zero-shot transfer setting. Distil-Whisper maintains the robustness of the Whisper model to difficult acoustic conditions, while being less prone to hallucination errors on long-form audio. Distil-Whisper is designed to be paired with Whisper for speculative decoding, yielding a 2 times speed-up while mathematically ensuring the same outputs as the original model. To facilitate further research in this domain, we make our training code, inference code and models publicly accessible.
We present MM-VID, an integrated system that harnesses the capabilities of GPT-4V, combined with specialized tools in vision, audio, and speech, to facilitate advanced video understanding. MM-VID is designed to address the challenges posed by long-form videos and intricate tasks such as reasoning within hour-long content and grasping storylines spanning multiple episodes. MM-VID uses a video-to-script generation with GPT-4V to transcribe multimodal elements into a long textual script. The generated script details character movements, actions, expressions, and dialogues, paving the way for large language models (LLMs) to achieve video understanding. This enables advanced capabilities, including audio description, character identification, and multimodal high-level comprehension. Experimental results demonstrate the effectiveness of MM-VID in handling distinct video genres with various video lengths. Additionally, we showcase its potential when applied to interactive environments, such as video games and graphic user interfaces.
In contemporary society, voice-controlled devices, such as smartphones and home assistants, have become pervasive due to their advanced capabilities and functionality. The always-on nature of their microphones offers users the convenience of readily accessing these devices. However, recent research and events have revealed that such voice-controlled devices are prone to various forms of malicious attacks, hence making it a growing concern for both users and researchers to safeguard against such attacks. Despite the numerous studies that have investigated adversarial attacks and privacy preservation for images, a conclusive study of this nature has not been conducted for the audio domain. Therefore, this paper aims to examine existing approaches for privacy-preserving and privacy-attacking strategies for audio and speech. To achieve this goal, we classify the attack and defense scenarios into several categories and provide detailed analysis of each approach. We also interpret the dissimilarities between the various approaches, highlight their contributions, and examine their limitations. Our investigation reveals that voice-controlled devices based on neural networks are inherently susceptible to specific types of attacks. Although it is possible to enhance the robustness of such models to certain forms of attack, more sophisticated approaches are required to comprehensively safeguard user privacy.
Expressive text-to-speech (TTS) aims to synthesize speeches with human-like tones, moods, or even artistic attributes. Recent advancements in expressive TTS empower users with the ability to directly control synthesis style through natural language prompts. However, these methods often require excessive training with a significant amount of style-annotated data, which can be challenging to acquire. Moreover, they may have limited adaptability due to fixed style annotations. In this work, we present FreeStyleTTS (FS-TTS), a controllable expressive TTS model with minimal human annotations. Our approach utilizes a large language model (LLM) to transform expressive TTS into a style retrieval task. The LLM selects the best-matching style references from annotated utterances based on external style prompts, which can be raw input text or natural language style descriptions. The selected reference guides the TTS pipeline to synthesize speeches with the intended style. This innovative approach provides flexible, versatile, and precise style control with minimal human workload. Experiments on a Mandarin storytelling corpus demonstrate FS-TTS's proficiency in leveraging LLM's semantic inference ability to retrieve desired styles from either input text or user-defined descriptions. This results in synthetic speeches that are closely aligned with the specified styles.