Modelling prosody variation is critical for synthesizing natural and expressive speech in end-to-end text-to-speech (TTS) systems. In this paper, a cross-utterance conditional VAE (CUC-VAE) is proposed to estimate a posterior probability distribution of the latent prosody features for each phoneme by conditioning on acoustic features, speaker information, and text features obtained from both past and future sentences. At inference time, instead of the standard Gaussian distribution used by VAE, CUC-VAE allows sampling from an utterance-specific prior distribution conditioned on cross-utterance information, which allows the prosody features generated by the TTS system to be related to the context and is more similar to how humans naturally produce prosody. The performance of CUC-VAE is evaluated via a qualitative listening test for naturalness, intelligibility and quantitative measurements, including word error rates and the standard deviation of prosody attributes. Experimental results on LJ-Speech and LibriTTS data show that the proposed CUC-VAE TTS system improves naturalness and prosody diversity with clear margins.
One of the most challenging scenarios for smart speakers is multi-talker, when target speech from the desired speaker is mixed with interfering speech from one or more speakers. A smart assistant needs to determine which voice to recognize and which to ignore and it needs to do so in a streaming, low-latency manner. This work presents two multi-microphone speech enhancement algorithms targeted at this scenario. Targeting on-device use-cases, we assume that the algorithm has access to the signal before the hotword, which is referred to as the noise context. First is the Context Aware Beamformer which uses the noise context and detected hotword to determine how to target the desired speaker. The second is an adaptive noise cancellation algorithm called Speech Cleaner which trains a filter using the noise context. It is demonstrated that the two algorithms are complementary in the signal-to-noise ratio conditions under which they work well. We also propose an algorithm to select which one to use based on estimated SNR. When using 3 microphone channels, the final system achieves a relative word error rate reduction of 55% at -12dB, and 43\% at 12dB.
Speech systems are sensitive to accent variations. This is especially challenging in the Indian context, with an abundance of languages but a dearth of linguistic studies characterising pronunciation variations. The growing number of L2 English speakers in India reinforces the need to study accents and L1-L2 interactions. We investigate the accents of Indian English (IE) speakers and report in detail our observations, both specific and common to all regions. In particular, we observe the phonemic variations and phonotactics occurring in the speakers' native languages and apply this to their English pronunciations. We demonstrate the influence of 18 Indian languages on IE by comparing the native language pronunciations with IE pronunciations obtained jointly from existing literature studies and phonetically annotated speech of 80 speakers. Consequently, we are able to validate the intuitions of Indian language influences on IE pronunciations by justifying pronunciation rules from the perspective of Indian language phonology. We obtain a comprehensive description in terms of universal and region-specific characteristics of IE, which facilitates accent conversion and adaptation of existing ASR and TTS systems to different Indian accents.
Several deep neural networks have recently been shown to generate activations similar to those of the brain in response to the same input. These algorithms, however, remain largely implausible: they require (1) extraordinarily large amounts of data, (2) unobtainable supervised labels, (3) textual rather than raw sensory input, and / or (4) implausibly large memory (e.g. thousands of contextual words). These elements highlight the need to identify algorithms that, under these limitations, would suffice to account for both behavioral and brain responses. Focusing on the issue of speech processing, we here hypothesize that self-supervised algorithms trained on the raw waveform constitute a promising candidate. Specifically, we compare a recent self-supervised architecture, Wav2Vec 2.0, to the brain activity of 412 English, French, and Mandarin individuals recorded with functional Magnetic Resonance Imaging (fMRI), while they listened to ~1h of audio books. Our results are four-fold. First, we show that this algorithm learns brain-like representations with as little as 600 hours of unlabelled speech -- a quantity comparable to what infants can be exposed to during language acquisition. Second, its functional hierarchy aligns with the cortical hierarchy of speech processing. Third, different training regimes reveal a functional specialization akin to the cortex: Wav2Vec 2.0 learns sound-generic, speech-specific and language-specific representations similar to those of the prefrontal and temporal cortices. Fourth, we confirm the similarity of this specialization with the behavior of 386 additional participants. These elements, resulting from the largest neuroimaging benchmark to date, show how self-supervised learning can account for a rich organization of speech processing in the brain, and thus delineate a path to identify the laws of language acquisition which shape the human brain.
The use of Automatic speech recognition (ASR) interfaces have become increasingly popular in daily life for use in interaction and control of electronic devices. The interfaces currently being used are not feasible for a variety of users such as those suffering from a speech disorder, locked-in syndrome, paralysis or people with utmost privacy requirements. In such cases, an interface that can identify envisioned speech using electroencephalogram (EEG) signals can be of great benefit. Various works targeting this problem have been done in the past. However, there has been limited work in identifying the frequency bands ($\delta, \theta, \alpha, \beta, \gamma$) of the EEG signal that contribute towards envisioned speech recognition. Therefore, in this work, we aim to analyze the significance of different EEG frequency bands and signals obtained from different lobes of the brain and their contribution towards recognizing envisioned speech. Signals obtained from different lobes and bandpass filtered for different frequency bands are fed to a spatio-temporal deep learning architecture with Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM). The performance is evaluated on a publicly available dataset comprising of three classification tasks - digit, character and images. We obtain a classification accuracy of $85.93\%$, $87.27\%$ and $87.51\%$ for the three tasks respectively. The code for the implementation has been made available at https://github.com/ayushayt/ImaginedSpeechRecognition.
Training speech translation (ST) models requires large and high-quality datasets. MuST-C is one of the most widely used ST benchmark datasets. It contains around 400 hours of speech-transcript-translation data for each of the eight translation directions. This dataset passes several quality-control filters during creation. However, we find that MuST-C still suffers from three major quality issues: audio-text misalignment, inaccurate translation, and unnecessary speaker's name. What are the impacts of these data quality issues for model development and evaluation? In this paper, we propose an automatic method to fix or filter the above quality issues, using English-German (En-De) translation as an example. Our experiments show that ST models perform better on clean test sets, and the rank of proposed models remains consistent across different test sets. Besides, simply removing misaligned data points from the training set does not lead to a better ST model.
In this paper, we investigate several existing and a new state-of-the-art generative adversarial network-based (GAN) voice conversion method for enhancing dysarthric speech for improved dysarthric speech recognition. We compare key components of existing methods as part of a rigorous ablation study to find the most effective solution to improve dysarthric speech recognition. We find that straightforward signal processing methods such as stationary noise removal and vocoder-based time stretching lead to dysarthric speech recognition results comparable to those obtained when using state-of-the-art GAN-based voice conversion methods as measured using a phoneme recognition task. Additionally, our proposed solution of a combination of MaskCycleGAN-VC and time stretched enhancement is able to improve the phoneme recognition results for certain dysarthric speakers compared to our time stretched baseline.
We present a neural text-to-speech (TTS) method that models natural vocal effort variation to improve the intelligibility of synthetic speech in the presence of noise. The method consists of first measuring the spectral tilt of unlabeled conventional speech data, and then conditioning a neural TTS model with normalized spectral tilt among other prosodic factors. Changing the spectral tilt parameter and keeping other prosodic factors unchanged enables effective vocal effort control at synthesis time independent of other prosodic factors. By extrapolation of the spectral tilt values beyond what has been seen in the original data, we can generate speech with high vocal effort levels, thus improving the intelligibility of speech in the presence of masking noise. We evaluate the intelligibility and quality of normal speech and speech with increased vocal effort in the presence of various masking noise conditions, and compare these to well-known speech intelligibility-enhancing algorithms. The evaluations show that the proposed method can improve the intelligibility of synthetic speech with little loss in speech quality.
Despite recent advancements in deep learning technologies, Child Speech Recognition remains a challenging task. Current Automatic Speech Recognition (ASR) models required substantial amounts of annotated data for training, which is scarce. In this work, we explore using the ASR model, wav2vec2, with different pretraining and finetuning configurations for self supervised learning (SSL) towards improving automatic child speech recognition. The pretrained wav2vec2 models were finetuned using different amounts of child speech training data to discover the optimum amount of data required to finetune the model for the task of child ASR. Our trained model receives the best word error rate (WER) of 8.37 on the in domain MyST dataset and WER of 10.38 on the out of domain PFSTAR dataset. We do not use any Language Models (LM) in our experiments.
Incremental learning is one paradigm to enable model building and updating at scale with streaming data. For end-to-end automatic speech recognition (ASR) tasks, the absence of human annotated labels along with the need for privacy preserving policies for model building makes it a daunting challenge. Motivated by these challenges, in this paper we use a cloud based framework for production systems to demonstrate insights from privacy preserving incremental learning for automatic speech recognition (ILASR). By privacy preserving, we mean, usage of ephemeral data which are not human annotated. This system is a step forward for production levelASR models for incremental/continual learning that offers near real-time test-bed for experimentation in the cloud for end-to-end ASR, while adhering to privacy-preserving policies. We show that the proposed system can improve the production models significantly(3%) over a new time period of six months even in the absence of human annotated labels with varying levels of weak supervision and large batch sizes in incremental learning. This improvement is 20% over test sets with new words and phrases in the new time period. We demonstrate the effectiveness of model building in a privacy-preserving incremental fashion for ASR while further exploring the utility of having an effective teacher model and use of large batch sizes.