Search

Low resource of parallel data is the key challenge of accent conversion(AC) problem in which both the pronunciation units and prosody pattern need to be converted. We propose a two-stage generative framework "convert-and-speak" in which the conversion is only operated on the semantic token level and the speech is synthesized conditioned on the converted semantic token with a speech generative model in target accent domain. The decoupling design enables the "speaking" module to use massive amount of target accent speech and relieves the parallel data required for the "conversion" module. Conversion with the bridge of semantic token also relieves the requirement for the data with text transcriptions and unlocks the usage of language pre-training technology to further efficiently reduce the need of parallel accent speech data. To reduce the complexity and latency of "speaking", a single-stage AR generative model is designed to achieve good quality as well as lower computation cost. Experiments on Indian-English to general American-English conversion show that the proposed framework achieves state-of-the-art performance in accent similarity, speech quality, and speaker maintenance with only 15 minutes of weakly parallel data which is not constrained to the same speaker. Extensive experimentation with diverse accent types suggests that this framework possesses a high degree of adaptability, making it readily scalable to accommodate other accents with low-resource data. Audio samples are available at https://www.microsoft.com/en-us/research/project/convert-and-speak-zero-shot-accent-conversion-with-minimumsupervision/., Comment: 9 pages, ACM MM2024(accepted)

Most existing masked audio modeling (MAM) methods learn audio representations by masking and reconstructing local spectrogram patches. However, the reconstruction loss mainly accounts for the signal-level quality of the reconstructed spectrogram and is still limited in extracting high-level audio semantics. In this paper, we propose to enhance the semantic modeling of MAM by distilling cross-modality knowledge from contrastive language-audio pretraining (CLAP) representations for both masked and unmasked regions (MAM-CLAP) and leveraging a multi-objective learning strategy with a supervised classification branch (SupMAM), thereby providing more semantic knowledge for MAM and enabling it to effectively learn global features from labels. Experiments show that our methods significantly improve the performance on multiple downstream tasks. Furthermore, by combining our MAM-CLAP with SupMAM, we can achieve new state-of-the-art results on various audio and speech classification tasks, exceeding previous self-supervised learning and supervised pretraining methods., Comment: Accepted by Interspeech2023

Existing deep learning based speech enhancement mainly employ a data-driven approach, which leverage large amounts of data with a variety of noise types to achieve noise removal from noisy signal. However, the high dependence on the data limits its generalization on the unseen complex noises in real-life environment. In this paper, we focus on the low-latency scenario and regard speech enhancement as a speech generation problem conditioned on the noisy signal, where we generate clean speech instead of identifying and removing noises. Specifically, we propose a conditional generative framework for speech enhancement, which models clean speech by acoustic codes of a neural speech codec and generates the speech codes conditioned on past noisy frames in an auto-regressive way. Moreover, we propose an explicit-alignment approach to align noisy frames with the generated speech tokens to improve the robustness and scalability to different input lengths. Different from other methods that leverage multiple stages to generate speech codes, we leverage a single-stage speech generation approach based on the TF-Codec neural codec to achieve high speech quality with low latency. Extensive results on both synthetic and real-recorded test set show its superiority over data-driven approaches in terms of noise robustness and temporal speech coherence., Comment: 5 pages, ICASSP2024(accepted)

Audiovisual segmentation (AVS) is a challenging task that aims to segment visual objects in videos according to their associated acoustic cues. With multiple sound sources and background disturbances involved, establishing robust correspondences between audio and visual contents poses unique challenges due to (1) complex entanglement across sound sources and (2) frequent changes in the occurrence of distinct sound events. Assuming sound events occur independently, the multi-source semantic space can be represented as the Cartesian product of single-source sub-spaces. We are motivated to decompose the multi-source audio semantics into single-source semantics for more effective interactions with visual content. We propose a semantic decomposition method based on product quantization, where the multi-source semantics can be decomposed and represented by several disentangled and noise-suppressed single-source semantics. Furthermore, we introduce a global-to-local quantization mechanism, which distills knowledge from stable global (clip-level) features into local (frame-level) ones, to handle frequent changes in audio semantics. Extensive experiments demonstrate that our semantically decomposed audio representation significantly improves AVS performance, e.g., +21.2% mIoU on the challenging AVS-Semantic benchmark with ResNet50 backbone. https://github.com/lxa9867/QSD.

Noise suppression (NS) models have been widely applied to enhance speech quality. Recently, Deep Learning-Based NS, which we denote as Deep Noise Suppression (DNS), became the mainstream NS method due to its excelling performance over traditional ones. However, DNS models face 2 major challenges for supporting the real-world applications. First, high-performing DNS models are usually large in size, causing deployment difficulties. Second, DNS models require extensive training data, including noisy audios as inputs and clean audios as labels. It is often difficult to obtain clean labels for training DNS models. We propose the use of knowledge distillation (KD) to resolve both challenges. Our study serves 2 main purposes. To begin with, we are among the first to comprehensively investigate mainstream KD techniques on DNS models to resolve the two challenges. Furthermore, we propose a novel Attention-Based-Compression KD method that outperforms all investigated mainstream KD frameworks on DNS task., Comment: This paper was accepted to Interspeech 2023 Main Conference

Packet loss concealment (PLC) is challenging in concealing missing contents both plausibly and naturally when there are only limited available context to use. Recently deep-learning based PLC algorithms have demonstrated their superiority over traditional counterparts; but their concealment ability is still mostly limited to a maximum of 120ms loss. Even with strong GAN-based generative models, it is still very challenging to predict long burst losses that could happen within/in-between phonemes. In this paper, we propose to use contrastive learning to learn a loss-robust semantic representation for PLC. A hybrid neural PLC architecture combining the semantic prediction and GAN-based generative model is designed to verify its effectiveness. Results on the blind test set of Interspeech2022 PLC Challenge show its superiority over commonly used UNet-style framework and the one without contrastive learning, especially for the longer burst loss at (120, 220] ms., Comment: 5 pages, ICASSP 2023(Accepted)

Time-variant factors often occur in real-world full-duplex communication applications. Some of them are caused by the complex environment such as non-stationary environmental noises and varying acoustic path while some are caused by the communication system such as the dynamic delay between the far-end and near-end signals. Current end-to-end deep neural network (DNN) based methods usually model the time-variant components implicitly and can hardly handle the unpredictable time-variance in real-time speech enhancement. To explicitly capture the time-variant components, we propose a dynamic kernel generation (DKG) module that can be introduced as a learnable plug-in to a DNN-based end-to-end pipeline. Specifically, the DKG module generates a convolutional kernel regarding to each input audio frame, so that the DNN model is able to dynamically adjust its weights according to the input signal during inference. Experimental results verify that DKG module improves the performance of the model under time-variant scenarios, in the joint acoustic echo cancellation (AEC) and deep noise suppression (DNS) tasks.

For the task of speech separation, previous study usually treats multi-channel and single-channel scenarios as two research tracks with specialized solutions developed respectively. Instead, we propose a simple and unified architecture - DasFormer (Deep alternating spectrogram transFormer) to handle both of them in the challenging reverberant environments. Unlike frame-wise sequence modeling, each TF-bin in the spectrogram is assigned with an embedding encoding spectral and spatial information. With such input, DasFormer is then formed by multiple repetition of simple blocks each of which integrates 1) two multi-head self-attention (MHSA) modules alternately processing within each frequency bin & temporal frame of the spectrogram 2) MBConv before each MHSA for modeling local features on the spectrogram. Experiments show that DasFormer has a powerful ability to model the time-frequency representation, whose performance far exceeds the current SOTA models in multi-channel speech separation, and also achieves single-channel SOTA in the more challenging yet realistic reverberation scenario., Comment: 5 pages, accepted by ICASSP2023

For real-time speech enhancement (SE) including noise suppression, dereverberation and acoustic echo cancellation, the time-variance of the audio signals becomes a severe challenge. The causality and memory usage limit that only the historical information can be used for the system to capture the time-variant characteristics. We propose to adaptively change the receptive field according to the input signal in deep neural network based SE model. Specifically, in an encoder-decoder framework, a dynamic attention span mechanism is introduced to all the attention modules for controlling the size of historical content used for processing the current frame. Experimental results verify that this dynamic mechanism can better track time-variant factors and capture speech-related characteristics, benefiting to both interference removing and speech quality retaining., Comment: ICASSP 2023 (Accepted)

Existing deep learning based speech enhancement (SE) methods either use blind end-to-end training or explicitly incorporate speaker embedding or phonetic information into the SE network to enhance speech quality. In this paper, we perceive speech and noises as different types of sound events and propose an event-based query method for SE. Specifically, representative speech embeddings that can discriminate speech with noises are first pre-trained with the sound event detection (SED) task. The embeddings are then clustered into fixed golden speech queries to assist the SE network to enhance the speech from noisy audio. The golden speech queries can be obtained offline and generalizable to different SE datasets and networks. Therefore, little extra complexity is introduced and no enrollment is needed for each speaker. Experimental results show that the proposed method yields significant gains compared with baselines and the golden queries are well generalized to different datasets., Comment: Accepted by ICASSP2023

Recently end-to-end neural audio/speech coding has shown its great potential to outperform traditional signal analysis based audio codecs. This is mostly achieved by following the VQ-VAE paradigm where blind features are learned, vector-quantized and coded. In this paper, instead of blind end-to-end learning, we propose to learn disentangled features for real-time neural speech coding. Specifically, more global-like speaker identity and local content features are learned with disentanglement to represent speech. Such a compact feature decomposition not only achieves better coding efficiency by exploiting bit allocation among different features but also provides the flexibility to do audio editing in embedding space, such as voice conversion in real-time communications. Both subjective and objective results demonstrate its coding efficiency and we find that the learned disentangled features show comparable performance on any-to-any voice conversion with modern self-supervised speech representation learning models with far less parameters and low latency, showing the potential of our neural coding framework., Comment: ICASSP 2023 (Accepted)

Neural audio/speech coding has recently demonstrated its capability to deliver high quality at much lower bitrates than traditional methods. However, existing neural audio/speech codecs employ either acoustic features or learned blind features with a convolutional neural network for encoding, by which there are still temporal redundancies within encoded features. This paper introduces latent-domain predictive coding into the VQ-VAE framework to fully remove such redundancies and proposes the TF-Codec for low-latency neural speech coding in an end-to-end manner. Specifically, the extracted features are encoded conditioned on a prediction from past quantized latent frames so that temporal correlations are further removed. Moreover, we introduce a learnable compression on the time-frequency input to adaptively adjust the attention paid to main frequencies and details at different bitrates. A differentiable vector quantization scheme based on distance-to-soft mapping and Gumbel-Softmax is proposed to better model the latent distributions with rate constraint. Subjective results on multilingual speech datasets show that, with low latency, the proposed TF-Codec at 1 kbps achieves significantly better quality than Opus at 9 kbps, and TF-Codec at 3 kbps outperforms both EVS at 9.6 kbps and Opus at 12 kbps. Numerous studies are conducted to demonstrate the effectiveness of these techniques., Comment: Accepted by IEEE/ACM TRANSACTIONS ON AUDIO, SPEECH, AND LANGUAGE PROCESSING (TASLP)

Bitrate scalability is a desirable feature for audio coding in real-time communications. Existing neural audio codecs usually enforce a specific bitrate during training, so different models need to be trained for each target bitrate, which increases the memory footprint at the sender and the receiver side and transcoding is often needed to support multiple receivers. In this paper, we introduce a cross-scale scalable vector quantization scheme (CSVQ), in which multi-scale features are encoded progressively with stepwise feature fusion and refinement. In this way, a coarse-level signal is reconstructed if only a portion of the bitstream is received, and progressively improves the quality as more bits are available. The proposed CSVQ scheme can be flexibly applied to any neural audio coding network with a mirrored auto-encoder structure to achieve bitrate scalability. Subjective results show that the proposed scheme outperforms the classical residual VQ (RVQ) with scalability. Moreover, the proposed CSVQ at 3 kbps outperforms Opus at 9 kbps and Lyra at 3kbps and it could provide a graceful quality boost with bitrate increase., Comment: INTERSPEECH 2022(Accepted)

In this paper we propose a multi-modal multi-correlation learning framework targeting at the task of audio-visual speech separation. Although previous efforts have been extensively put on combining audio and visual modalities, most of them solely adopt a straightforward concatenation of audio and visual features. To exploit the real useful information behind these two modalities, we define two key correlations which are: (1) identity correlation (between timbre and facial attributes); (2) phonetic correlation (between phoneme and lip motion). These two correlations together comprise the complete information, which shows a certain superiority in separating target speaker's voice especially in some hard cases, such as the same gender or similar content. For implementation, contrastive learning or adversarial training approach is applied to maximize these two correlations. Both of them work well, while adversarial training shows its advantage by avoiding some limitations of contrastive learning. Compared with previous research, our solution demonstrates clear improvement on experimental metrics without additional complexity. Further analysis reveals the validity of the proposed architecture and its good potential for future extension., Comment: 5 pages, accepted by interspeech2022

Neural audio coding has shown very promising results recently in the literature to largely outperform traditional codecs but limited attention has been paid on its error resilience. Neural codecs trained considering only source coding tend to be extremely sensitive to channel noises, especially in wireless channels with high error rate. In this paper, we investigate how to elevate the error resilience of neural audio codecs for packet losses that often occur during real-time communications. We propose a feature-domain packet loss concealment algorithm (FD-PLC) for real-time neural speech coding. Specifically, we introduce a self-attention-based module on the received latent features to recover lost frames in the feature domain before the decoder. A hybrid segment-level and frame-level frequency-domain discriminator is employed to guide the network to focus on both the generative quality of lost frames and the continuity with neighbouring frames. Experimental results on several error patterns show that the proposed scheme can achieve better robustness compared with the corresponding error-free and error-resilient baselines. We also show that feature-domain concealment is superior to waveform-domain counterpart as post-processing., Comment: 5 pages, Interspeech 2022(Accepted)

Deep-learning based methods have shown their advantages in audio coding over traditional ones but limited attention has been paid on real-time communications (RTC). This paper proposes the TFNet, an end-to-end neural speech codec with low latency for RTC. It takes an encoder-temporal filtering-decoder paradigm that has seldom been investigated in audio coding. An interleaved structure is proposed for temporal filtering to capture both short-term and long-term temporal dependencies. Furthermore, with end-to-end optimization, the TFNet is jointly optimized with speech enhancement and packet loss concealment, yielding a one-for-all network for three tasks. Both subjective and objective results demonstrate the efficiency of the proposed TFNet., Comment: ICASSP 2022 (Accepted)

Existing speech enhancement methods mainly separate speech from noises at the signal level or in the time-frequency domain. They seldom pay attention to the semantic information of a corrupted signal. In this paper, we aim to bridge this gap by extracting phoneme identities to help speech enhancement. Specifically, we propose a phoneme-based distribution regularization (PbDr) for speech enhancement, which incorporates frame-wise phoneme information into speech enhancement network in a conditional manner. As different phonemes always lead to different feature distributions in frequency, we propose to learn a parameter pair, i.e. scale and bias, through a phoneme classification vector to modulate the speech enhancement network. The modulation parameter pair includes not only frame-wise but also frequency-wise conditions, which effectively map features to phoneme-related distributions. In this way, we explicitly regularize speech enhancement features by recognition vectors. Experiments on public datasets demonstrate that the proposed PbDr module can not only boost the perceptual quality for speech enhancement but also the recognition accuracy of an ASR system on the enhanced speech. This PbDr module could be readily incorporated into other speech enhancement networks as well., Comment: ICASSP 2021 (Accepted)

Speech enhancement is challenging because of the diversity of background noise types. Most of the existing methods are focused on modelling the speech rather than the noise. In this paper, we propose a novel idea to model speech and noise simultaneously in a two-branch convolutional neural network, namely SN-Net. In SN-Net, the two branches predict speech and noise, respectively. Instead of information fusion only at the final output layer, interaction modules are introduced at several intermediate feature domains between the two branches to benefit each other. Such an interaction can leverage features learned from one branch to counteract the undesired part and restore the missing component of the other and thus enhance their discrimination capabilities. We also design a feature extraction module, namely residual-convolution-and-attention (RA), to capture the correlations along temporal and frequency dimensions for both the speech and the noises. Evaluations on public datasets show that the interaction module plays a key role in simultaneous modeling and the SN-Net outperforms the state-of-the-art by a large margin on various evaluation metrics. The proposed SN-Net also shows superior performance for speaker separation., Comment: AAAI 2021 (Accepted)

The recent development of CNN-based image dehazing has revealed the effectiveness of end-to-end modeling. However, extending the idea to end-to-end video dehazing has not been explored yet. In this paper, we propose an End-to-End Video Dehazing Network (EVD-Net), to exploit the temporal consistency between consecutive video frames. A thorough study has been conducted over a number of structure options, to identify the best temporal fusion strategy. Furthermore, we build an End-to-End United Video Dehazing and Detection Network(EVDD-Net), which concatenates and jointly trains EVD-Net with a video object detection model. The resulting augmented end-to-end pipeline has demonstrated much more stable and accurate detection results in hazy video.

This paper proposes an image dehazing model built with a convolutional neural network (CNN), called All-in-One Dehazing Network (AOD-Net). It is designed based on a re-formulated atmospheric scattering model. Instead of estimating the transmission matrix and the atmospheric light separately as most previous models did, AOD-Net directly generates the clean image through a light-weight CNN. Such a novel end-to-end design makes it easy to embed AOD-Net into other deep models, e.g., Faster R-CNN, for improving high-level task performance on hazy images. Experimental results on both synthesized and natural hazy image datasets demonstrate our superior performance than the state-of-the-art in terms of PSNR, SSIM and the subjective visual quality. Furthermore, when concatenating AOD-Net with Faster R-CNN and training the joint pipeline from end to end, we witness a large improvement of the object detection performance on hazy images.