Creating synthetic voices with found data is challenging, as real-world recordings often contain various types of audio degradation. One way to address this problem is to pre-enhance the speech with an enhancement model and then use the enhanced data for text-to-speech (TTS) model training. Ideally, the enhancement model should be able to tackle multiple types of audio degradation simultaneously. This paper investigates the use of conditional diffusion models for generalized speech enhancement. The enhancement is performed on the log Mel-spectrogram domain to align with the TTS training objective. Text information is introduced as an additional condition to improve the model robustness. Experiments on real-world recordings demonstrate that the synthetic voice built on data enhanced by the proposed model produces higher-quality synthetic speech, compared to those trained on data enhanced by strong baselines. Audio samples are available at \url{https://dmse4tts.github.io/}.
Recent studies on pronunciation scoring have explored the effect of introducing phone embeddings as reference pronunciation, but mostly in an implicit manner, i.e., addition or concatenation of reference phone embedding and actual pronunciation of the target phone as the phone-level pronunciation quality representation. In this paper, we propose to use linguistic-acoustic similarity to explicitly measure the deviation of non-native production from its native reference for pronunciation assessment. Specifically, the deviation is first estimated by the cosine similarity between reference phone embedding and corresponding acoustic embedding. Next, a phone-level Goodness of pronunciation (GOP) pre-training stage is introduced to guide this similarity-based learning for better initialization of the aforementioned two embeddings. Finally, a transformer-based hierarchical pronunciation scorer is used to map a sequence of phone embeddings, acoustic embeddings along with their similarity measures to predict the final utterance-level score. Experimental results on the non-native databases suggest that the proposed system significantly outperforms the baselines, where the acoustic and phone embeddings are simply added or concatenated. A further examination shows that the phone embeddings learned in the proposed approach are able to capture linguistic-acoustic attributes of native pronunciation as reference.
A typical fluency scoring system generally relies on an automatic speech recognition (ASR) system to obtain time stamps in input speech for either the subsequent calculation of fluency-related features or directly modeling speech fluency with an end-to-end approach. This paper describes a novel ASR-free approach for automatic fluency assessment using self-supervised learning (SSL). Specifically, wav2vec2.0 is used to extract frame-level speech features, followed by K-means clustering to assign a pseudo label (cluster index) to each frame. A BLSTM-based model is trained to predict an utterance-level fluency score from frame-level SSL features and the corresponding cluster indexes. Neither speech transcription nor time stamp information is required in the proposed system. It is ASR-free and can potentially avoid the ASR errors effect in practice. Experimental results carried out on non-native English databases show that the proposed approach significantly improves the performance in the "open response" scenario as compared to previous methods and matches the recently reported performance in the "read aloud" scenario.
Very deep models for speaker recognition (SR) have demonstrated remarkable performance improvement in recent research. However, it is impractical to deploy these models for on-device applications with constrained computational resources. On the other hand, light-weight models are highly desired in practice despite their sub-optimal performance. This research aims to improve light-weight SR models through large-scale label-free knowledge distillation (KD). Existing KD approaches for SR typically require speaker labels to learn task-specific knowledge, due to the inefficiency of conventional loss for distillation. To address the inefficiency problem and achieve label-free KD, we propose to employ the contrastive loss from self-supervised learning for distillation. Extensive experiments are conducted on a collection of public speech datasets from diverse sources. Results on light-weight SR models show that the proposed approach of label-free KD with contrastive loss consistently outperforms both conventional distillation methods and self-supervised learning methods by a significant margin.
Probabilistic linear discriminant analysis (PLDA) is commonly used in speaker verification systems to score the similarity of speaker embeddings. Recent studies improved the performance of PLDA in domain-matched conditions by diagonalizing its covariance. We suspect such brutal pruning approach could eliminate its capacity in modeling dimension correlation of speaker embeddings, leading to inadequate performance with domain adaptation. This paper explores two alternative covariance regularization approaches, namely, interpolated PLDA and sparse PLDA, to tackle the problem. The interpolated PLDA incorporates the prior knowledge from cosine scoring to interpolate the covariance of PLDA. The sparse PLDA introduces a sparsity penalty to update the covariance. Experimental results demonstrate that both approaches outperform diagonal regularization noticeably with domain adaptation. In addition, in-domain data can be significantly reduced when training sparse PLDA for domain adaptation.
DNN-based models achieve high performance in the speaker verification (SV) task with substantial computation costs. The model size is an essential concern in applying models on resource-constrained devices, while model compression for SV models has not been studied extensively in previous works. Weight quantization is exploited to compress DNN-based speaker embedding extraction models in this paper. Uniform and Powers-of-Two quantization are utilized in the experiments. The results on VoxCeleb show that the weight quantization can decrease the size of ECAPA-TDNN and ResNet by 4 times with insignificant performance decline. The quantized 4-bit ResNet achieves similar performance to the original model with an 8 times smaller size. We empirically show that the performance of ECAPA-TDNN is more sensitive than ResNet to quantization due to the difference in weight distribution. The experiments on CN-Celeb also demonstrate that quantized models are robust for SV in the language mismatch scenario.