Search

Speech separation, the task of isolating multiple speech sources from a mixed audio signal, remains challenging in noisy environments. In this paper, we propose a generative correction method to enhance the output of a discriminative separator. By leveraging a generative corrector based on a diffusion model, we refine the separation process for single-channel mixture speech by removing noises and perceptually unnatural distortions. Furthermore, we optimize the generative model using a predictive loss to streamline the diffusion model's reverse process into a single step and rectify any associated errors by the reverse process. Our method achieves state-of-the-art performance on the in-domain Libri2Mix noisy dataset, and out-of-domain WSJ with a variety of noises, improving SI-SNR by 22-35% relative to SepFormer, demonstrating robustness and strong generalization capabilities., Comment: Accepted at INTERSPEECH 2024

Adversarial examples have proven to threaten speaker identification systems, and several countermeasures against them have been proposed. In this paper, we propose a method to detect the presence of adversarial examples, i.e., a binary classifier distinguishing between benign and adversarial examples. We build upon and extend previous work on attack type classification by exploring new architectures. Additionally, we introduce a method for identifying the victim model on which the adversarial attack is carried out. To achieve this, we generate a new dataset containing multiple attacks performed against various victim models. We achieve an AUC of 0.982 for attack detection, with no more than a 0.03 drop in performance for unknown attacks. Our attack classification accuracy (excluding benign) reaches 86.48% across eight attack types using our LightResNet34 architecture, while our victim model classification accuracy reaches 72.28% across four victim models.

Time series data is often composed of information at multiple time scales, particularly in biomedical data. While numerous deep learning strategies exist to capture this information, many make networks larger, require more data, are more demanding to compute, and are difficult to interpret. This limits their usefulness in real-world applications facing even modest computational or data constraints and can further complicate their translation into practice. We present a minimal, computationally efficient Time Scale Network combining the translation and dilation sequence used in discrete wavelet transforms with traditional convolutional neural networks and back-propagation. The network simultaneously learns features at many time scales for sequence classification with significantly reduced parameters and operations. We demonstrate advantages in Atrial Dysfunction detection including: superior accuracy-per-parameter and accuracy-per-operation, fast training and inference speeds, and visualization and interpretation of learned patterns in atrial dysfunction detection on ECG signals. We also demonstrate impressive performance in seizure prediction using EEG signals. Our network isolated a few time scales that could be strategically selected to achieve 90.9% accuracy using only 1,133 active parameters and consistently converged on pulsatile waveform shapes. This method does not rest on any constraints or assumptions regarding signal content and could be leveraged in any area of time series analysis dealing with signals containing features at many time scales., Comment: 8 pages, 5 figures, preprint

Common target sound extraction (TSE) approaches primarily relied on discriminative approaches in order to separate the target sound while minimizing interference from the unwanted sources, with varying success in separating the target from the background. This study introduces DPM-TSE, a first generative method based on diffusion probabilistic modeling (DPM) for target sound extraction, to achieve both cleaner target renderings as well as improved separability from unwanted sounds. The technique also tackles common background noise issues with DPM by introducing a correction method for noise schedules and sample steps. This approach is evaluated using both objective and subjective quality metrics on the FSD Kaggle 2018 dataset. The results show that DPM-TSE has a significant improvement in perceived quality in terms of target extraction and purity., Comment: Submitted to ICASSP 2024

Visually grounded speech systems learn from paired images and their spoken captions. Recently, there have been attempts to utilize the visually grounded models trained from images and their corresponding text captions, such as CLIP, to improve speech-based visually grounded models' performance. However, the majority of these models only utilize the pretrained image encoder. Cascaded SpeechCLIP attempted to generate localized word-level information and utilize both the pretrained image and text encoders. Despite using both, they noticed a substantial drop in retrieval performance. We proposed Segmental SpeechCLIP which used a hierarchical segmental speech encoder to generate sequences of word-like units. We used the pretrained CLIP text encoder on top of these word-like unit representations and showed significant improvements over the cascaded variant of SpeechCLIP. Segmental SpeechCLIP directly learns the word embeddings as input to the CLIP text encoder bypassing the vocabulary embeddings. Here, we explore mapping audio to CLIP vocabulary embeddings via regularization and quantization. As our objective is to distill semantic information into the speech encoders, we explore the usage of large unimodal pretrained language models as the text encoders. Our method enables us to bridge image and text encoders e.g. DINO and RoBERTa trained with uni-modal data. Finally, we extend our framework in audio-only settings where only pairs of semantically related audio are available. Experiments show that audio-only systems perform close to the audio-visual system.

We present a novel typical-to-atypical voice conversion approach (DuTa-VC), which (i) can be trained with nonparallel data (ii) first introduces diffusion probabilistic model (iii) preserves the target speaker identity (iv) is aware of the phoneme duration of the target speaker. DuTa-VC consists of three parts: an encoder transforms the source mel-spectrogram into a duration-modified speaker-independent mel-spectrogram, a decoder performs the reverse diffusion to generate the target mel-spectrogram, and a vocoder is applied to reconstruct the waveform. Objective evaluations conducted on the UASpeech show that DuTa-VC is able to capture severity characteristics of dysarthric speech, reserves speaker identity, and significantly improves dysarthric speech recognition as a data augmentation. Subjective evaluations by two expert speech pathologists validate that DuTa-VC can preserve the severity and type of dysarthria of the target speakers in the synthesized speech.

Self-supervised methods such as Contrastive predictive Coding (CPC) have greatly improved the quality of the unsupervised representations. These representations significantly reduce the amount of labeled data needed for downstream task performance, such as automatic speech recognition. CPC learns representations by learning to predict future frames given current frames. Based on the observation that the acoustic information, e.g., phones, changes slower than the feature extraction rate in CPC, we propose regularization techniques that impose slowness constraints on the features. Here we propose two regularization techniques: Self-expressing constraint and Left-or-Right regularization. We evaluate the proposed model on ABX and linear phone classification tasks, acoustic unit discovery, and automatic speech recognition. The regularized CPC trained on 100 hours of unlabeled data matches the performance of the baseline CPC trained on 360 hours of unlabeled data. We also show that our regularization techniques are complementary to data augmentation and can further boost the system's performance. In monolingual, cross-lingual, or multilingual settings, with/without data augmentation, regardless of the amount of data used for training, our regularized models outperformed the baseline CPC models on the ABX task.

The recently proposed Joint Energy-based Model (JEM) interprets discriminatively trained classifier $p(y|x)$ as an energy model, which is also trained as a generative model describing the distribution of the input observations $p(x)$. The JEM training relies on "positive examples" (i.e. examples from the training data set) as well as on "negative examples", which are samples from the modeled distribution $p(x)$ generated by means of Stochastic Gradient Langevin Dynamics (SGLD). Unfortunately, SGLD often fails to deliver negative samples of sufficient quality during the standard JEM training, which causes a very unbalanced contribution from the positive and negative examples when calculating gradients for JEM updates. As a consequence, the standard JEM training is quite unstable requiring careful tuning of hyper-parameters and frequent restarts when the training starts diverging. This makes it difficult to apply JEM to different neural network architectures, modalities, and tasks. In this work, we propose a training procedure that stabilizes SGLD-based JEM training (ST-JEM) by balancing the contribution from the positive and negative examples. We also propose to add an additional "regularization" term to the training objective -- MI between the input observations $x$ and output labels $y$ -- which encourages the JEM classifier to make more certain decisions about output labels. We demonstrate the effectiveness of our approach on the CIFAR10 and CIFAR100 tasks. We also consider the task of classifying phonemes in a speech signal, for which we were not able to train JEM without the proposed stabilization. We show that a convincing speech can be generated from the trained model. Alternatively, corrupted speech can be de-noised by bringing it closer to the modeled speech distribution using a few SGLD iterations. We also propose and discuss additional applications of the trained model.

Speech super-resolution/Bandwidth Extension (BWE) can improve downstream tasks like Automatic Speaker Verification (ASV). We introduce a simple novel technique called Self-FiLM to inject self-supervision into existing BWE models via Feature-wise Linear Modulation. We hypothesize that such information captures domain/environment information, which can give zero-shot generalization. Self-FiLM Conditional GAN (CGAN) gives 18% relative improvement in Equal Error Rate and 8.5% in minimum Decision Cost Function using state-of-the-art ASV system on SRE21 test. We further by 1) deep feature loss from time-domain models and 2) re-training of data2vec 2.0 models on naturalistic wideband (VoxCeleb) and telephone data (SRE Superset etc.). Lastly, we integrate self-supervision with CycleGAN to present a completely unsupervised solution that matches the semi-supervised performance., Comment: Under review

Automatic Speaker Verification (ASV) technology has become commonplace in virtual assistants. However, its performance suffers when there is a mismatch between the train and test domains. Mixed bandwidth training, i.e., pooling training data from both domains, is a preferred choice for developing a universal model that works for both narrowband and wideband domains. We propose complementing this technique by performing neural upsampling of narrowband signals, also known as bandwidth extension. Our main goal is to discover and analyze high-performing time-domain Generative Adversarial Network (GAN) based models to improve our downstream state-of-the-art ASV system. We choose GANs since they (1) are powerful for learning conditional distribution and (2) allow flexible plug-in usage as a pre-processor during the training of downstream task (ASV) with data augmentation. Prior works mainly focus on feature-domain bandwidth extension and limited experimental setups. We address these limitations by 1) using time-domain extension models, 2) reporting results on three real test sets, 2) extending training data, and 3) devising new test-time schemes. We compare supervised (conditional GAN) and unsupervised GANs (CycleGAN) and demonstrate average relative improvement in Equal Error Rate of 8.6% and 7.7%, respectively. For further analysis, we study changes in spectrogram visual quality, audio perceptual quality, t-SNE embeddings, and ASV score distributions. We show that our bandwidth extension leads to phenomena such as a shift of telephone (test) embeddings towards wideband (train) signals, a negative correlation of perceptual quality with downstream performance, and condition-independent score calibration., Comment: Submit to IEEE/ACM Transactions on Audio, Speech, and Language Processing

In recent studies, self-supervised pre-trained models tend to outperform supervised pre-trained models in transfer learning. In particular, self-supervised learning (SSL) of utterance-level speech representation can be used in speech applications that require discriminative representation of consistent attributes within an utterance: speaker, language, emotion, and age. Existing frame-level self-supervised speech representation, e.g., wav2vec, can be used as utterance-level representation with pooling, but the models are usually large. There are also SSL techniques to learn utterance-level representation. One of the most successful is a contrastive method, which requires negative sampling: selecting alternative samples to contrast with the current sample (anchor). However, this does not ensure that all the negative samples belong to classes different from the anchor class without labels. This paper applies a non-contrastive self-supervised method to learn utterance-level embeddings. We adapted DIstillation with NO labels (DINO) from computer vision to speech. Unlike contrastive methods, DINO does not require negative sampling. We compared DINO to x-vector trained in a supervised manner. When transferred to down-stream tasks (speaker verification, speech emotion recognition (SER), and Alzheimer's disease detection), DINO outperformed x-vector. We studied the influence of several aspects during transfer learning such as dividing the fine-tuning process into steps, chunk lengths, or augmentation. During fine-tuning, tuning the last affine layers first and then the whole network surpassed fine-tuning all at once. Using shorter chunk lengths, although they generate more diverse inputs, did not necessarily improve performance, implying speech segments at least with a specific length are required for better performance per application. Augmentation was helpful in SER., Comment: EARLY ACCESS of IEEE JSTSP Special Issue on Self-Supervised Learning for Speech and Audio Processing

Considering the abundance of unlabeled speech data and the high labeling costs, unsupervised learning methods can be essential for better system development. One of the most successful methods is contrastive self-supervised methods, which require negative sampling: sampling alternative samples to contrast with the current sample (anchor). However, it is hard to ensure if all the negative samples belong to classes different from the anchor class without labels. This paper applies a non-contrastive self-supervised learning method on an unlabeled speech corpus to learn utterance-level embeddings. We used DIstillation with NO labels (DINO), proposed in computer vision, and adapted it to the speech domain. Unlike the contrastive methods, DINO does not require negative sampling. These embeddings were evaluated on speaker verification and emotion recognition. In speaker verification, the unsupervised DINO embedding with cosine scoring provided 4.38% EER on the VoxCeleb1 test trial. This outperforms the best contrastive self-supervised method by 40% relative in EER. An iterative pseudo-labeling training pipeline, not requiring speaker labels, further improved the EER to 1.89%. In emotion recognition, the DINO embedding performed 60.87, 79.21, and 56.98% in micro-f1 score on IEMOCAP, Crema-D, and MSP-Podcast, respectively. The results imply the generality of the DINO embedding to different speech applications., Comment: Accepted at Interspeech 2022

Adversarial attacks are a threat to automatic speech recognition (ASR) systems, and it becomes imperative to propose defenses to protect them. In this paper, we perform experiments to show that K2 conformer hybrid ASR is strongly affected by white-box adversarial attacks. We propose three defenses--denoiser pre-processor, adversarially fine-tuning ASR model, and adversarially fine-tuning joint model of ASR and denoiser. Our evaluation shows denoiser pre-processor (trained on offline adversarial examples) fails to defend against adaptive white-box attacks. However, adversarially fine-tuning the denoiser using a tandem model of denoiser and ASR offers more robustness. We evaluate two variants of this defense--one updating parameters of both models and the second keeping ASR frozen. The joint model offers a mean absolute decrease of 19.3\% ground truth (GT) WER with reference to baseline against fast gradient sign method (FGSM) attacks with different $L_\infty$ norms. The joint model with frozen ASR parameters gives the best defense against projected gradient descent (PGD) with 7 iterations, yielding a mean absolute increase of 22.3\% GT WER with reference to baseline; and against PGD with 500 iterations, yielding a mean absolute decrease of 45.08\% GT WER and an increase of 68.05\% adversarial target WER., Comment: Submitted to Interspeech 2022

Adversarial attacks pose a severe security threat to the state-of-the-art speaker identification systems, thereby making it vital to propose countermeasures against them. Building on our previous work that used representation learning to classify and detect adversarial attacks, we propose an improvement to it using AdvEst, a method to estimate adversarial perturbation. First, we prove our claim that training the representation learning network using adversarial perturbations as opposed to adversarial examples (consisting of the combination of clean signal and adversarial perturbation) is beneficial because it eliminates nuisance information. At inference time, we use a time-domain denoiser to estimate the adversarial perturbations from adversarial examples. Using our improved representation learning approach to obtain attack embeddings (signatures), we evaluate their performance for three applications: known attack classification, attack verification, and unknown attack detection. We show that common attacks in the literature (Fast Gradient Sign Method (FGSM), Projected Gradient Descent (PGD), Carlini-Wagner (CW) with different Lp threat models) can be classified with an accuracy of ~96%. We also detect unknown attacks with an equal error rate (EER) of ~9%, which is absolute improvement of ~12% from our previous work., Comment: Submitted to InterSpeech 2022

Speech systems developed for a particular choice of acoustic domain and sampling frequency do not translate easily to others. The usual practice is to learn domain adaptation and bandwidth extension models independently. Contrary to this, we propose to learn both tasks together. Particularly, we learn to map narrowband conversational telephone speech to wideband microphone speech. We developed parallel and non-parallel learning solutions which utilize both paired and unpaired data. First, we first discuss joint and disjoint training of multiple generative models for our tasks. Then, we propose a two-stage learning solution where we use a pre-trained domain adaptation system for pre-processing in bandwidth extension training. We evaluated our schemes on a Speaker Verification downstream task. We used the JHU-MIT experimental setup for NIST SRE21, which comprises SRE16, SRE-CTS Superset and SRE21. Our results provide the first evidence that learning both tasks is better than learning just one. On SRE16, our best system achieves 22% relative improvement in Equal Error Rate w.r.t. a direct learning baseline and 8% w.r.t. a strong bandwidth expansion system., Comment: submitted to Interspeech 2022

The high cost of data acquisition makes Automatic Speech Recognition (ASR) model training problematic for most existing languages, including languages that do not even have a written script, or for which the phone inventories remain unknown. Past works explored multilingual training, transfer learning, as well as zero-shot learning in order to build ASR systems for these low-resource languages. While it has been shown that the pooling of resources from multiple languages is helpful, we have not yet seen a successful application of an ASR model to a language unseen during training. A crucial step in the adaptation of ASR from seen to unseen languages is the creation of the phone inventory of the unseen language. The ultimate goal of our work is to build the phone inventory of a language unseen during training in an unsupervised way without any knowledge about the language. In this paper, we 1) investigate the influence of different factors (i.e., model architecture, phonotactic model, type of speech representation) on phone recognition in an unknown language; 2) provide an analysis of which phones transfer well across languages and which do not in order to understand the limitations of and areas for further improvement for automatic phone inventory creation; and 3) present different methods to build a phone inventory of an unseen language in an unsupervised way. To that end, we conducted mono-, multi-, and crosslingual experiments on a set of 13 phonetically diverse languages and several in-depth analyses. We found a number of universal phone tokens (IPA symbols) that are well-recognized cross-linguistically. Through a detailed analysis of results, we conclude that unique sounds, similar sounds, and tone languages remain a major challenge for phonetic inventory discovery., Comment: Accepted for publication in Computer Speech and Language

The pervasiveness of intra-utterance code-switching (CS) in spoken content requires that speech recognition (ASR) systems handle mixed language. Designing a CS-ASR system has many challenges, mainly due to data scarcity, grammatical structure complexity, and domain mismatch. The most common method for addressing CS is to train an ASR system with the available transcribed CS speech, along with monolingual data. In this work, we propose a zero-shot learning methodology for CS-ASR by augmenting the monolingual data with artificially generating CS text. We based our approach on random lexical replacements and Equivalence Constraint (EC) while exploiting aligned translation pairs to generate random and grammatically valid CS content. Our empirical results show a 65.5% relative reduction in language model perplexity, and 7.7% in ASR WER on two ecologically valid CS test sets. The human evaluation of the generated text using EC suggests that more than 80% is of adequate quality.

Typically, unsupervised segmentation of speech into the phone and word-like units are treated as separate tasks and are often done via different methods which do not fully leverage the inter-dependence of the two tasks. Here, we unify them and propose a technique that can jointly perform both, showing that these two tasks indeed benefit from each other. Recent attempts employ self-supervised learning, such as contrastive predictive coding (CPC), where the next frame is predicted given past context. However, CPC only looks at the audio signal's frame-level structure. We overcome this limitation with a segmental contrastive predictive coding (SCPC) framework to model the signal structure at a higher level, e.g., phone level. A convolutional neural network learns frame-level representation from the raw waveform via noise-contrastive estimation (NCE). A differentiable boundary detector finds variable-length segments, which are then used to optimize a segment encoder via NCE to learn segment representations. The differentiable boundary detector allows us to train frame-level and segment-level encoders jointly. Experiments show that our single model outperforms existing phone and word segmentation methods on TIMIT and Buckeye datasets. We discover that phone class impacts the boundary detection performance, and the boundaries between successive vowels or semivowels are the most difficult to identify. Finally, we use SCPC to extract speech features at the segment level rather than at uniformly spaced frame level (e.g., 10 ms) and produce variable rate representations that change according to the contents of the utterance. We can lower the feature extraction rate from the typical 100 Hz to as low as 14.5 Hz on average while still outperforming the MFCC features on the linear phone classification task., Comment: arXiv admin note: substantial text overlap with arXiv:2106.02170

This technical report describes Johns Hopkins University speaker recognition system submitted to Voxceleb Speaker Recognition Challenge 2021 Track 3: Self-supervised speaker verification (closed). Our overall training process is similar to the proposed one from the first place team in the last year's VoxSRC2020 challenge. The main difference is a recently proposed non-contrastive self-supervised method in computer vision (CV), distillation with no labels (DINO), is used to train our initial model, which outperformed the last year's contrastive learning based on momentum contrast (MoCo). Also, this requires only a few iterations in the iterative clustering stage, where pseudo labels for supervised embedding learning are updated based on the clusters of the embeddings generated from a model that is continually fine-tuned over iterations. In the final stage, Res2Net50 is trained on the final pseudo labels from the iterative clustering stage. This is our best submitted model to the challenge, showing 1.89, 6.50, and 6.89 in EER(%) in voxceleb1 test o, VoxSRC-21 validation, and test trials, respectively.

Speech emotion recognition is the task of recognizing the speaker's emotional state given a recording of their utterance. While most of the current approaches focus on inferring emotion from isolated utterances, we argue that this is not sufficient to achieve conversational emotion recognition (CER) which deals with recognizing emotions in conversations. In this work, we propose several approaches for CER by treating it as a sequence labeling task. We investigated transformer architecture for CER and, compared it with ResNet-34 and BiLSTM architectures in both contextual and context-less scenarios using IEMOCAP corpus. Based on the inner workings of the self-attention mechanism, we proposed DiverseCatAugment (DCA), an augmentation scheme, which improved the transformer model performance by an absolute 3.3% micro-f1 on conversations and 3.6% on isolated utterances. We further enhanced the performance by introducing an interlocutor-aware transformer model where we learn a dictionary of interlocutor index embeddings to exploit diarized conversations., Comment: Accepted for ASRU 2021

Capitalization and punctuation are important cues for comprehending written texts and conversational transcripts. Yet, many ASR systems do not produce punctuated and case-formatted speech transcripts. We propose to use a multi-task system that can exploit the relations between casing and punctuation to improve their prediction performance. Whereas text data for predicting punctuation and truecasing is seemingly abundant, we argue that written text resources are inadequate as training data for conversational models. We quantify the mismatch between written and conversational text domains by comparing the joint distributions of punctuation and word cases, and by testing our model cross-domain. Further, we show that by training the model in the written text domain and then transfer learning to conversations, we can achieve reasonable performance with less data., Comment: Accepted for ASRU 2021

Adversarial attacks have become a major threat for machine learning applications. There is a growing interest in studying these attacks in the audio domain, e.g, speech and speaker recognition; and find defenses against them. In this work, we focus on using representation learning to classify/detect attacks w.r.t. the attack algorithm, threat model or signal-to-adversarial-noise ratio. We found that common attacks in the literature can be classified with accuracies as high as 90%. Also, representations trained to classify attacks against speaker identification can be used also to classify attacks against speaker verification and speech recognition. We also tested an attack verification task, where we need to decide whether two speech utterances contain the same attack. We observed that our models did not generalize well to attack algorithms not included in the attack representation model training. Motivated by this, we evaluated an unknown attack detection task. We were able to detect unknown attacks with equal error rates of about 19%, which is promising., Comment: Accepted at Interspeech 2021

Dialog acts can be interpreted as the atomic units of a conversation, more fine-grained than utterances, characterized by a specific communicative function. The ability to structure a conversational transcript as a sequence of dialog acts -- dialog act recognition, including the segmentation -- is critical for understanding dialog. We apply two pre-trained transformer models, XLNet and Longformer, to this task in English and achieve strong results on Switchboard Dialog Act and Meeting Recorder Dialog Act corpora with dialog act segmentation error rates (DSER) of 8.4% and 14.2%. To understand the key factors affecting dialog act recognition, we perform a comparative analysis of models trained under different conditions. We find that the inclusion of a broader conversational context helps disambiguate many dialog act classes, especially those infrequent in the training data. The presence of punctuation in the transcripts has a massive effect on the models' performance, and a detailed analysis reveals specific segmentation patterns observed in its absence. Finally, we find that the label set specificity does not affect dialog act segmentation performance. These findings have significant practical implications for spoken language understanding applications that depend heavily on a good-quality segmentation being available., Comment: Accepted for publication in Transactions of the Association of Computational Linguistics. This is a pre-MIT Press publication version and it is subject to change

This paper introduces WaveGrad 2, a non-autoregressive generative model for text-to-speech synthesis. WaveGrad 2 is trained to estimate the gradient of the log conditional density of the waveform given a phoneme sequence. The model takes an input phoneme sequence, and through an iterative refinement process, generates an audio waveform. This contrasts to the original WaveGrad vocoder which conditions on mel-spectrogram features, generated by a separate model. The iterative refinement process starts from Gaussian noise, and through a series of refinement steps (e.g., 50 steps), progressively recovers the audio sequence. WaveGrad 2 offers a natural way to trade-off between inference speed and sample quality, through adjusting the number of refinement steps. Experiments show that the model can generate high fidelity audio, approaching the performance of a state-of-the-art neural TTS system. We also report various ablation studies over different model configurations. Audio samples are available at https://wavegrad.github.io/v2., Comment: Proceedings of INTERSPEECH

The success in designing Code-Switching (CS) ASR often depends on the availability of the transcribed CS resources. Such dependency harms the development of ASR in low-resourced languages such as Bengali and Hindi. In this paper, we exploit the transfer learning approach to design End-to-End (E2E) CS ASR systems for the two low-resourced language pairs using different monolingual speech data and a small set of noisy CS data. We trained the CS-ASR, following two steps: (i) building a robust bilingual ASR system using a convolution-augmented transformer (Conformer) based acoustic model and n-gram language model, and (ii) fine-tuned the entire E2E ASR with limited noisy CS data. We tested our method on MUCS 2021 challenge and achieved 3rd place in the CS track. We then tested the proposed method using noisy CS data released for Hindi-English and Bengali-English pairs in Multilingual and Code-Switching ASR Challenges for Low Resource Indian Languages (MUCS 2021) and achieved 3rd place in the CS track. Unlike, the leading two systems that benefited from crawling YouTube and learning transliteration pairs, our proposed transfer learning approach focused on using only the limited CS data with no data-cleaning or data re-segmentation. Our approach achieved 14.1% relative gain in word error rate (WER) in Hindi-English and 27.1% in Bengali-English. We provide detailed guidelines on the steps to finetune the self-attention based model for limited data for ASR. Moreover, we release the code and recipe used in this paper.

Automatic detection of phoneme or word-like units is one of the core objectives in zero-resource speech processing. Recent attempts employ self-supervised training methods, such as contrastive predictive coding (CPC), where the next frame is predicted given past context. However, CPC only looks at the audio signal's frame-level structure. We overcome this limitation with a segmental contrastive predictive coding (SCPC) framework that can model the signal structure at a higher level e.g. at the phoneme level. In this framework, a convolutional neural network learns frame-level representation from the raw waveform via noise-contrastive estimation (NCE). A differentiable boundary detector finds variable-length segments, which are then used to optimize a segment encoder via NCE to learn segment representations. The differentiable boundary detector allows us to train frame-level and segment-level encoders jointly. Typically, phoneme and word segmentation are treated as separate tasks. We unify them and experimentally show that our single model outperforms existing phoneme and word segmentation methods on TIMIT and Buckeye datasets. We analyze the impact of boundary threshold and when is the right time to include the segmental loss in the learning process.

With the increase in the availability of speech from varied domains, it is imperative to use such out-of-domain data to improve existing speech systems. Domain adaptation is a prominent pre-processing approach for this. We investigate it for adapt microphone speech to the telephone domain. Specifically, we explore CycleGAN-based unpaired translation of microphone data to improve the x-vector/speaker embedding network for Telephony Speaker Verification. We first demonstrate the efficacy of this on real challenging data and then, to improve further, we modify the CycleGAN formulation to make the adaptation task-specific. We modify CycleGAN's identity loss, cycle-consistency loss, and adversarial loss to operate in the deep feature space. Deep features of a signal are extracted from an auxiliary (speaker embedding) network and, hence, preserves speaker identity. Our 3D convolution-based Deep Feature Discriminators (DFD) show relative improvements of 5-10% in terms of equal error rate. To dive deeper, we study a challenging scenario of pooling (adapted) microphone and telephone data with data augmentations and telephone codecs. Finally, we highlight the sensitivity of CycleGAN hyper-parameters and introduce a parameter called probability of adaptation.

The ubiquitous presence of machine learning systems in our lives necessitates research into their vulnerabilities and appropriate countermeasures. In particular, we investigate the effectiveness of adversarial attacks and defenses against automatic speech recognition (ASR) systems. We select two ASR models - a thoroughly studied DeepSpeech model and a more recent Espresso framework Transformer encoder-decoder model. We investigate two threat models: a denial-of-service scenario where fast gradient-sign method (FGSM) or weak projected gradient descent (PGD) attacks are used to degrade the model's word error rate (WER); and a targeted scenario where a more potent imperceptible attack forces the system to recognize a specific phrase. We find that the attack transferability across the investigated ASR systems is limited. To defend the model, we use two preprocessing defenses: randomized smoothing and WaveGAN-based vocoder, and find that they significantly improve the model's adversarial robustness. We show that a WaveGAN vocoder can be a useful countermeasure to adversarial attacks on ASR systems - even when it is jointly attacked with the ASR, the target phrases' word error rate is high., Comment: This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible

Adversarial examples to speaker recognition (SR) systems are generated by adding a carefully crafted noise to the speech signal to make the system fail while being imperceptible to humans. Such attacks pose severe security risks, making it vital to deep-dive and understand how much the state-of-the-art SR systems are vulnerable to these attacks. Moreover, it is of greater importance to propose defenses that can protect the systems against these attacks. Addressing these concerns, this paper at first investigates how state-of-the-art x-vector based SR systems are affected by white-box adversarial attacks, i.e., when the adversary has full knowledge of the system. x-Vector based SR systems are evaluated against white-box adversarial attacks common in the literature like fast gradient sign method (FGSM), basic iterative method (BIM)--a.k.a. iterative-FGSM--, projected gradient descent (PGD), and Carlini-Wagner (CW) attack. To mitigate against these attacks, the paper proposes four pre-processing defenses. It evaluates them against powerful adaptive white-box adversarial attacks, i.e., when the adversary has full knowledge of the system, including the defense. The four pre-processing defenses--viz. randomized smoothing, DefenseGAN, variational autoencoder (VAE), and Parallel WaveGAN vocoder (PWG) are compared against the baseline defense of adversarial training. Conclusions indicate that SR systems were extremely vulnerable under BIM, PGD, and CW attacks. Among the proposed pre-processing defenses, PWG combined with randomized smoothing offers the most protection against the attacks, with accuracy averaging 93% compared to 52% in the undefended system and an absolute improvement >90% for BIM attacks with $L_\infty>0.001$ and CW attack., Comment: This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible

This paper introduces a novel method to diagnose the source-target attention in state-of-the-art end-to-end speech recognition models with joint connectionist temporal classification (CTC) and attention training. Our method is based on the fact that both, CTC and source-target attention, are acting on the same encoder representations. To understand the functionality of the attention, CTC is applied to compute the token posteriors given the attention outputs. We found that the source-target attention heads are able to predict several tokens ahead of the current one. Inspired by the observation, a new regularization method is proposed which leverages CTC to make source-target attention more focused on the frames corresponding to the output token being predicted by the decoder. Experiments reveal stable improvements up to 7\% and 13\% relatively with the proposed regularization on TED-LIUM 2 and LibriSpeech., Comment: submitted to ICASSP2021. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works

Data augmentation is a widely used strategy for training robust machine learning models. It partially alleviates the problem of limited data for tasks like speech emotion recognition (SER), where collecting data is expensive and challenging. This study proposes CopyPaste, a perceptually motivated novel augmentation procedure for SER. Assuming that the presence of emotions other than neutral dictates a speaker's overall perceived emotion in a recording, concatenation of an emotional (emotion E) and a neutral utterance can still be labeled with emotion E. We hypothesize that SER performance can be improved using these concatenated utterances in model training. To verify this, three CopyPaste schemes are tested on two deep learning models: one trained independently and another using transfer learning from an x-vector model, a speaker recognition model. We observed that all three CopyPaste schemes improve SER performance on all the three datasets considered: MSP-Podcast, Crema-D, and IEMOCAP. Additionally, CopyPaste performs better than noise augmentation and, using them together improves the SER performance further. Our experiments on noisy test sets suggested that CopyPaste is effective even in noisy test conditions., Comment: Accepted at ICASSP2021

Deep learning based speech denoising still suffers from the challenge of improving perceptual quality of enhanced signals. We introduce a generalized framework called Perceptual Ensemble Regularization Loss (PERL) built on the idea of perceptual losses. Perceptual loss discourages distortion to certain speech properties and we analyze it using six large-scale pre-trained models: speaker classification, acoustic model, speaker embedding, emotion classification, and two self-supervised speech encoders (PASE+, wav2vec 2.0). We first build a strong baseline (w/o PERL) using Conformer Transformer Networks on the popular enhancement benchmark called VCTK-DEMAND. Using auxiliary models one at a time, we find acoustic event and self-supervised model PASE+ to be most effective. Our best model (PERL-AE) only uses acoustic event model (utilizing AudioSet) to outperform state-of-the-art methods on major perceptual metrics. To explore if denoising can leverage full framework, we use all networks but find that our seven-loss formulation suffers from the challenges of Multi-Task Learning. Finally, we report a critical observation that state-of-the-art Multi-Task weight learning methods cannot outperform hand tuning, perhaps due to challenges of domain mismatch and weak complementarity of losses.

The idea of combining multiple languages' recordings to train a single automatic speech recognition (ASR) model brings the promise of the emergence of universal speech representation. Recently, a Transformer encoder-decoder model has been shown to leverage multilingual data well in IPA transcriptions of languages presented during training. However, the representations it learned were not successful in zero-shot transfer to unseen languages. Because that model lacks an explicit factorization of the acoustic model (AM) and language model (LM), it is unclear to what degree the performance suffered from differences in pronunciation or the mismatch in phonotactics. To gain more insight into the factors limiting zero-shot ASR transfer, we replace the encoder-decoder with a hybrid ASR system consisting of a separate AM and LM. Then, we perform an extensive evaluation of monolingual, multilingual, and crosslingual (zero-shot) acoustic and language models on a set of 13 phonetically diverse languages. We show that the gain from modeling crosslingual phonotactics is limited, and imposing a too strong model can hurt the zero-shot transfer. Furthermore, we find that a multilingual LM hurts a multilingual ASR system's performance, and retaining only the target language's phonotactic data in LM training is preferable., Comment: Accepted for publication in IEEE ICASSP 2021. The first 2 authors contributed equally to this work

Zero-shot multi-speaker Text-to-Speech (TTS) generates target speaker voices given an input text and the corresponding speaker embedding. In this work, we investigate the effectiveness of the TTS reconstruction objective to improve representation learning for speaker verification. We jointly trained end-to-end Tacotron 2 TTS and speaker embedding networks in a self-supervised fashion. We hypothesize that the embeddings will contain minimal phonetic information since the TTS decoder will obtain that information from the textual input. TTS reconstruction can also be combined with speaker classification to enhance these embeddings further. Once trained, the speaker encoder computes representations for the speaker verification task, while the rest of the TTS blocks are discarded. We investigated training TTS from either manual or ASR-generated transcripts. The latter allows us to train embeddings on datasets without manual transcripts. We compared ASR transcripts and Kaldi phone alignments as TTS inputs, showing that the latter performed better due to their finer resolution. Unsupervised TTS embeddings improved EER by 2.06\% absolute with regard to i-vectors for the LibriTTS dataset. TTS with speaker classification loss improved EER by 0.28\% and 0.73\% absolutely from a model using only speaker classification loss in LibriTTS and Voxceleb1 respectively.

Unsupervised spoken term discovery consists of two tasks: finding the acoustic segment boundaries and labeling acoustically similar segments with the same labels. We perform segmentation based on the assumption that the frame feature vectors are more similar within a segment than across the segments. Therefore, for strong segmentation performance, it is crucial that the features represent the phonetic properties of a frame more than other factors of variability. We achieve this via a self-expressing autoencoder framework. It consists of a single encoder and two decoders with shared weights. The encoder projects the input features into a latent representation. One of the decoders tries to reconstruct the input from these latent representations and the other from the self-expressed version of them. We use the obtained features to segment and cluster the speech data. We evaluate the performance of the proposed method in the Zero Resource 2020 challenge unit discovery task. The proposed system consistently outperforms the baseline, demonstrating the usefulness of the method in learning representations.

We investigated an enhancement and a domain adaptation approach to make speaker verification systems robust to perturbations of far-field speech. In the enhancement approach, using paired (parallel) reverberant-clean speech, we trained a supervised Generative Adversarial Network (GAN) along with a feature mapping loss. For the domain adaptation approach, we trained a Cycle Consistent Generative Adversarial Network (CycleGAN), which maps features from far-field domain to the speaker embedding training domain. This was trained on unpaired data in an unsupervised manner. Both networks, termed Supervised Enhancement Network (SEN) and Domain Adaptation Network (DAN) respectively, were trained with multi-task objectives in (filter-bank) feature domain. On a simulated test setup, we first note the benefit of using feature mapping (FM) loss along with adversarial loss in SEN. Then, we tested both supervised and unsupervised approaches on several real noisy datasets. We observed relative improvements ranging from 2% to 31% in terms of DCF. Using three training schemes, we also establish the effectiveness of the novel DAN approach., Comment: submitted to INTERSPEECH 2020

Only a handful of the world's languages are abundant with the resources that enable practical applications of speech processing technologies. One of the methods to overcome this problem is to use the resources existing in other languages to train a multilingual automatic speech recognition (ASR) model, which, intuitively, should learn some universal phonetic representations. In this work, we focus on gaining a deeper understanding of how general these representations might be, and how individual phones are getting improved in a multilingual setting. To that end, we select a phonetically diverse set of languages, and perform a series of monolingual, multilingual and crosslingual (zero-shot) experiments. The ASR is trained to recognize the International Phonetic Alphabet (IPA) token sequences. We observe significant improvements across all languages in the multilingual setting, and stark degradation in the crosslingual setting, where the model, among other errors, considers Javanese as a tone language. Notably, as little as 10 hours of the target language training data tremendously reduces ASR error rates. Our analysis uncovered that even the phones that are unique to a single language can benefit greatly from adding training data from other languages - an encouraging result for the low-resource speech community., Comment: Submitted to Interspeech 2020. For some reason, the ArXiv Latex engine rendered it in more than 4 pages

Automatic Speech Recognition (ASR) systems introduce word errors, which often confuse punctuation prediction models, turning punctuation restoration into a challenging task. These errors usually take the form of homonyms. We show how retrofitting of the word embeddings on the domain-specific data can mitigate ASR errors. Our main contribution is a method for better alignment of homonym embeddings and the validation of the presented method on the punctuation prediction task. We record the absolute improvement in punctuation prediction accuracy between 6.2% (for question marks) to 9% (for periods) when compared with the state-of-the-art model., Comment: submitted to INTERSPEECH'20

In this work, we explore the dependencies between speaker recognition and emotion recognition. We first show that knowledge learned for speaker recognition can be reused for emotion recognition through transfer learning. Then, we show the effect of emotion on speaker recognition. For emotion recognition, we show that using a simple linear model is enough to obtain good performance on the features extracted from pre-trained models such as the x-vector model. Then, we improve emotion recognition performance by fine-tuning for emotion classification. We evaluated our experiments on three different types of datasets: IEMOCAP, MSP-Podcast, and Crema-D. By fine-tuning, we obtained 30.40%, 7.99%, and 8.61% absolute improvement on IEMOCAP, MSP-Podcast, and Crema-D respectively over baseline model with no pre-training. Finally, we present results on the effect of emotion on speaker verification. We observed that speaker verification performance is prone to changes in test speaker emotions. We found that trials with angry utterances performed worst in all three datasets. We hope our analysis will initiate a new line of research in the speaker recognition community., Comment: 45th International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2020

Data augmentation is conventionally used to inject robustness in Speaker Verification systems. Several recently organized challenges focus on handling novel acoustic environments. Deep learning based speech enhancement is a modern solution for this. Recently, a study proposed to optimize the enhancement network in the activation space of a pre-trained auxiliary network. This methodology, called deep feature loss, greatly improved over the state-of-the-art conventional x-vector based system on a children speech dataset called BabyTrain. This work analyzes various facets of that approach and asks few novel questions in that context. We first search for optimal number of auxiliary network activations, training data, and enhancement feature dimension. Experiments reveal the importance of Signal-to-Noise Ratio filtering that we employ to create a large, clean, and naturalistic corpus for enhancement network training. To counter the "mismatch" problem in enhancement, we find enhancing front-end (x-vector network) data helpful while harmful for the back-end (Probabilistic Linear Discriminant Analysis (PLDA)). Importantly, we find enhanced signals contain complementary information to original. Established by combining them in front-end, this gives ~40% relative improvement over the baseline. We also do an ablation study to remove a noise class from x-vector data augmentation and, for such systems, we establish the utility of enhancement regardless of whether it has seen that noise class itself during training. Finally, we design several dereverberation schemes to conclude ineffectiveness of deep feature loss enhancement scheme for this task., Comment: 8 pages; accepted in Odyssey2020 workshop

This paper presents the problems and solutions addressed at the JSALT workshop when using a single microphone for speaker detection in adverse scenarios. The main focus was to tackle a wide range of conditions that go from meetings to wild speech. We describe the research threads we explored and a set of modules that was successful for these scenarios. The ultimate goal was to explore speaker detection; but our first finding was that an effective diarization improves detection, and not having a diarization stage impoverishes the performance. All the different configurations of our research agree on this fact and follow a main backbone that includes diarization as a previous stage. With this backbone, we analyzed the following problems: voice activity detection, how to deal with noisy signals, domain mismatch, how to improve the clustering; and the overall impact of previous stages in the final speaker detection. In this paper, we show partial results for speaker diarizarion to have a better understanding of the problem and we present the final results for speaker detection., Comment: Submitted to ICASSP 2020

Recently very deep transformers have outperformed conventional bi-directional long short-term memory networks by a large margin in speech recognition. However, to put it into production usage, inference computation cost is still a serious concern in real scenarios. In this paper, we study two different non-autoregressive transformer structure for automatic speech recognition (ASR): A-CMLM and A-FMLM. During training, for both frameworks, input tokens fed to the decoder are randomly replaced by special mask tokens. The network is required to predict the tokens corresponding to those mask tokens by taking both unmasked context and input speech into consideration. During inference, we start from all mask tokens and the network iteratively predicts missing tokens based on partial results. We show that this framework can support different decoding strategies, including traditional left-to-right. A new decoding strategy is proposed as an example, which starts from the easiest predictions to the most difficult ones. Results on Mandarin (Aishell) and Japanese (CSJ) ASR benchmarks show the possibility to train such a non-autoregressive network for ASR. Especially in Aishell, the proposed method outperformed the Kaldi ASR system and it matches the performance of the state-of-the-art autoregressive transformer with 7x speedup. Pretrained models and code will be made available after publication.

In this paper, we propose to improve emotion recognition by combining acoustic information and conversation transcripts. On the one hand, an LSTM network was used to detect emotion from acoustic features like f0, shimmer, jitter, MFCC, etc. On the other hand, a multi-resolution CNN was used to detect emotion from word sequences. This CNN consists of several parallel convolutions with different kernel sizes to exploit contextual information at different levels. A temporal pooling layer aggregates the hidden representations of different words into a unique sequence level embedding, from which we computed the emotion posteriors. We optimized a weighted sum of classification and verification losses. The verification loss tries to bring embeddings from the same emotions closer while separating embeddings from different emotions. We also compared our CNN with state-of-the-art text-based hand-crafted features (e-vector). We evaluated our approach on the USC-IEMOCAP dataset as well as the dataset consisting of US English telephone speech. In the former, we used human-annotated transcripts while in the latter, we used ASR transcripts. The results showed fusing audio and transcript information improved unweighted accuracy by relative 24% for IEMOCAP and relative 3.4% for the telephone data compared to a single acoustic system.

Current speaker recognition technology provides great performance with the x-vector approach. However, performance decreases when the evaluation domain is different from the training domain, an issue usually addressed with domain adaptation approaches. Recently, unsupervised domain adaptation using cycle-consistent Generative Adversarial Netorks (CycleGAN) has received a lot of attention. CycleGAN learn mappings between features of two domains given non-parallel data. We investigate their effectiveness in low resource scenario i.e. when limited amount of target domain data is available for adaptation, a case unexplored in previous works. We experiment with two adaptation tasks: microphone to telephone and a novel reverberant to clean adaptation with the end goal of improving speaker recognition performance. Number of speakers present in source and target domains are 7000 and 191 respectively. By adding noise to the target domain during CycleGAN training, we were able to achieve better performance compared to the adaptation system whose CycleGAN was trained on a larger target data. On reverberant to clean adaptation task, our models improved EER by 18.3% relative on VOiCES dataset compared to a system trained on clean data. They also slightly improved over the state-of-the-art Weighted Prediction Error (WPE) de-reverberation algorithm., Comment: 8 pages, accepted to ASRU 2019

Speaker Verification still suffers from the challenge of generalization to novel adverse environments. We leverage on the recent advancements made by deep learning based speech enhancement and propose a feature-domain supervised denoising based solution. We propose to use Deep Feature Loss which optimizes the enhancement network in the hidden activation space of a pre-trained auxiliary speaker embedding network. We experimentally verify the approach on simulated and real data. A simulated testing setup is created using various noise types at different SNR levels. For evaluation on real data, we choose BabyTrain corpus which consists of children recordings in uncontrolled environments. We observe consistent gains in every condition over the state-of-the-art augmented Factorized-TDNN x-vector system. On BabyTrain corpus, we observe relative gains of 10.38% and 12.40% in minDCF and EER respectively., Comment: 5 pages, accepted in ICASSP 2020

The task of making speaker verification systems robust to adverse scenarios remain a challenging and an active area of research. We developed an unsupervised feature enhancement approach in log-filter bank domain with the end goal of improving speaker verification performance. We experimented with using both real speech recorded in adverse environments and degraded speech obtained by simulation to train the enhancement systems. The effectiveness of the approach was shown by testing on several real, simulated noisy, and reverberant test sets. The approach yielded significant improvements on both real and simulated sets when data augmentation was not used in speaker verification pipeline or augmentation was used only during x-vector training. When data augmentation was used for x-vector and PLDA training, our enhancement approach yielded slight improvements., Comment: 5 pages; accepted in ICASSP 2020

BERT, which stands for Bidirectional Encoder Representations from Transformers, is a recently introduced language representation model based upon the transfer learning paradigm. We extend its fine-tuning procedure to address one of its major limitations - applicability to inputs longer than a few hundred words, such as transcripts of human call conversations. Our method is conceptually simple. We segment the input into smaller chunks and feed each of them into the base model. Then, we propagate each output through a single recurrent layer, or another transformer, followed by a softmax activation. We obtain the final classification decision after the last segment has been consumed. We show that both BERT extensions are quick to fine-tune and converge after as little as 1 epoch of training on a small, domain-specific data set. We successfully apply them in three different tasks involving customer call satisfaction prediction and topic classification, and obtain a significant improvement over the baseline models in two of them., Comment: 4 figures, 7 pages

This paper presents an unsupervised segment-based method for robust voice activity detection (rVAD). The method consists of two passes of denoising followed by a voice activity detection (VAD) stage. In the first pass, high-energy segments in a speech signal are detected by using a posteriori signal-to-noise ratio (SNR) weighted energy difference and if no pitch is detected within a segment, the segment is considered as a high-energy noise segment and set to zero. In the second pass, the speech signal is denoised by a speech enhancement method, for which several methods are explored. Next, neighbouring frames with pitch are grouped together to form pitch segments, and based on speech statistics, the pitch segments are further extended from both ends in order to include both voiced and unvoiced sounds and likely non-speech parts as well. In the end, a posteriori SNR weighted energy difference is applied to the extended pitch segments of the denoised speech signal for detecting voice activity. We evaluate the VAD performance of the proposed method using two databases, RATS and Aurora-2, which contain a large variety of noise conditions. The rVAD method is further evaluated, in terms of speaker verification performance, on the RedDots 2016 challenge database and its noise-corrupted versions. Experiment results show that rVAD is compared favourably with a number of existing methods. In addition, we present a modified version of rVAD where computationally intensive pitch extraction is replaced by computationally efficient spectral flatness calculation. The modified version significantly reduces the computational complexity at the cost of moderately inferior VAD performance, which is an advantage when processing a large amount of data and running on low resource devices. The source code of rVAD is made publicly available.