No Arabic abstract
Recent advances in neural-network based generative modeling of speech has shown great potential for speech coding. However, the performance of such models drops when the input is not clean speech, e.g., in the presence of background noise, preventing its use in practical applications. In this paper we examine the reason and discuss methods to overcome this issue. Placing a denoising preprocessing stage when extracting features and target clean speech during training is shown to be the best performing strategy.
Recent neural waveform synthesizers such as WaveNet, WaveGlow, and the neural-source-filter (NSF) model have shown good performance in speech synthesis despite their different methods of waveform generation. The similarity between speech and music audio synthesis techniques suggests interesting avenues to explore in terms of the best way to apply speech synthesizers in the music domain. This work compares three neural synthesizers used for musical instrument sounds generation under three scenarios: training from scratch on music data, zero-shot learning from the speech domain, and fine-tuning-based adaptation from the speech to the music domain. The results of a large-scale perceptual test demonstrated that the performance of three synthesizers improved when they were pre-trained on speech data and fine-tuned on music data, which indicates the usefulness of knowledge from speech data for music audio generation. Among the synthesizers, WaveGlow showed the best potential in zero-shot learning while NSF performed best in the other scenarios and could generate samples that were perceptually close to natural audio.
Neural network architectures are at the core of powerful automatic speech recognition systems (ASR). However, while recent researches focus on novel model architectures, the acoustic input features remain almost unchanged. Traditional ASR systems rely on multidimensional acoustic features such as the Mel filter bank energies alongside with the first, and second order derivatives to characterize time-frames that compose the signal sequence. Considering that these components describe three different views of the same element, neural networks have to learn both the internal relations that exist within these features, and external or global dependencies that exist between the time-frames. Quaternion-valued neural networks (QNN), recently received an important interest from researchers to process and learn such relations in multidimensional spaces. Indeed, quaternion numbers and QNNs have shown their efficiency to process multidimensional inputs as entities, to encode internal dependencies, and to solve many tasks with up to four times less learning parameters than real-valued models. We propose to investigate modern quaternion-valued models such as convolutional and recurrent quaternion neural networks in the context of speech recognition with the TIMIT dataset. The experiments show that QNNs always outperform real-valued equivalent models with way less free parameters, leading to a more efficient, compact, and expressive representation of the relevant information.
Statistical signal processing based speech enhancement methods adopt expert knowledge to design the statistical models and linear filters, which is complementary to the deep neural network (DNN) based methods which are data-driven. In this paper, by using expert knowledge from statistical signal processing for network design and optimization, we extend the conventional Kalman filtering (KF) to the supervised learning scheme, and propose the neural Kalman filtering (NKF) for speech enhancement. Two intermediate clean speech estimates are first produced from recurrent neural networks (RNN) and linear Wiener filtering (WF) separately and are then linearly combined by a learned NKF gain to yield the NKF output. Supervised joint training is applied to NKF to learn to automatically trade-off between the instantaneous linear estimation made by the WF and the long-term non-linear estimation made by the RNN. The NKF method can be seen as using expert knowledge from WF to regularize the RNN estimations to improve its generalization ability to the noise conditions unseen in the training. Experiments in different noisy conditions show that the proposed method outperforms the baseline methods both in terms of objective evaluation metrics and automatic speech recognition (ASR) word error rates (WERs).
This paper presents a speech BERT model to extract embedded prosody information in speech segments for improving the prosody of synthesized speech in neural text-to-speech (TTS). As a pre-trained model, it can learn prosody attributes from a large amount of speech data, which can utilize more data than the original training data used by the target TTS. The embedding is extracted from the previous segment of a fixed length in the proposed BERT. The extracted embedding is then used together with the mel-spectrogram to predict the following segment in the TTS decoder. Experimental results obtained by the Transformer TTS show that the proposed BERT can extract fine-grained, segment-level prosody, which is complementary to utterance-level prosody to improve the final prosody of the TTS speech. The objective distortions measured on a single speaker TTS are reduced between the generated speech and original recordings. Subjective listening tests also show that the proposed approach is favorably preferred over the TTS without the BERT prosody embedding module, for both in-domain and out-of-domain applications. For Microsoft professional, single/multiple speakers and the LJ Speaker in the public database, subjective preference is similarly confirmed with the new BERT prosody embedding. TTS demo audio samples are in https://judy44chen.github.io/TTSSpeechBERT/.
Estimating the perceived quality of an audio signal is critical for many multimedia and audio processing systems. Providers strive to offer optimal and reliable services in order to increase the user quality of experience (QoE). In this work, we present an investigation of the applicability of neural networks for non-intrusive audio quality assessment. We propose three neural network-based approaches for mean opinion score (MOS) estimation. We compare our results to three instrumental measures: the perceptual evaluation of speech quality (PESQ), the ITU-T Recommendation P.563, and the speech-to-reverberation energy ratio. Our evaluation uses a speech dataset contaminated with convolutive and additive noise, labeled using a crowd-based QoE evaluation, evaluated with Pearson correlation with MOS labels, and mean-squared-error of the estimated MOS. Our proposed approaches outperform the aforementioned instrumental measures, with a fully connected deep neural network using Mel-frequency features providing the best correlation (0.87) and the lowest mean squared error (0.15)