End-to-end modeling (E2E) of automatic speech recognition (ASR) blends all the components of a traditional speech recognition system into a unified model. Although it simplifies training and decoding pipelines, the unified model is hard to adapt when mismatch exists between training and test data. In this work, we focus on contextual speech recognition, which is particularly challenging for E2E models because it introduces significant mismatch between training and test data. To improve the performance in the presence of complex contextual information, we propose to use class-based language models(CLM) that can populate the classes with contextdependent information in real-time. To enable this approach to scale to a large number of class members and minimize search errors, we propose a token passing decoder with efficient token recombination for E2E systems for the first time. We evaluate the proposed system on general and contextual ASR, and achieve relative 62% Word Error Rate(WER) reduction for contextual ASR without hurting performance for general ASR. We show that the proposed method performs well without modification of the decoding hyper-parameters across tasks, making it a general solution for E2E ASR.
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.
The INTERSPEECH 2020 Deep Noise Suppression Challenge is intended to promote collaborative research in real-time single-channel Speech Enhancement aimed to maximize the subjective (perceptual) quality of the enhanced speech. A typical approach to evaluate the noise suppression methods is to use objective metrics on the test set obtained by splitting the original dataset. Many publications report reasonable performance on the synthetic test set drawn from the same distribution as that of the training set. However, often the model performance degrades significantly on real recordings. Also, most of the conventional objective metrics do not correlate well with subjective tests and lab subjective tests are not scalable for a large test set. In this challenge, we open-source a large clean speech and noise corpus for training the noise suppression models and a representative test set to real-world scenarios consisting of both synthetic and real recordings. We also open source an online subjective test framework based on ITU-T P.808 for researchers to quickly test their developments. The winners of this challenge will be selected based on subjective evaluation on a representative test set using P.808 framework.
This paper presents a novel zero-shot learning approach towards personalized speech enhancement through the use of a sparsely active ensemble model. Optimizing speech denoising systems towards a particular test-time speaker can improve performance and reduce run-time complexity. However, test-time model adaptation may be challenging if collecting data from the test-time speaker is not possible. To this end, we propose using an ensemble model wherein each specialist module denoises noisy utterances from a distinct partition of training set speakers. The gating module inexpensively estimates test-time speaker characteristics in the form of an embedding vector and selects the most appropriate specialist module for denoising the test signal. Grouping the training set speakers into non-overlapping semantically similar groups is non-trivial and ill-defined. To do this, we first train a Siamese network using noisy speech pairs to maximize or minimize the similarity of its output vectors depending on whether the utterances derive from the same speaker or not. Next, we perform k-means clustering on the latent space formed by the averaged embedding vectors per training set speaker. In this way, we designate speaker groups and train specialist modules optimized around partitions of the complete training set. Our experiments show that ensemble models made up of low-capacity specialists can outperform high-capacity generalist models with greater efficiency and improved adaptation towards unseen test-time speakers.
Automatic speech recognition (ASR) is a capability which enables a program to process human speech into a written form. Recent developments in artificial intelligence (AI) have led to high-accuracy ASR systems based on deep neural networks, such as the recurrent neural network transducer (RNN-T). However, the core components and the performed operations of these approaches depart from the powerful biological counterpart, i.e., the human brain. On the other hand, the current developments in biologically-inspired ASR models, based on spiking neural networks (SNNs), lag behind in terms of accuracy and focus primarily on small scale applications. In this work, we revisit the incorporation of biologically-plausible models into deep learning and we substantially enhance their capabilities, by taking inspiration from the diverse neural and synaptic dynamics found in the brain. In particular, we introduce neural connectivity concepts emulating the axo-somatic and the axo-axonic synapses. Based on this, we propose novel deep learning units with enriched neuro-synaptic dynamics and integrate them into the RNN-T architecture. We demonstrate for the first time, that a biologically realistic implementation of a large-scale ASR model can yield competitive performance levels compared to the existing deep learning models. Specifically, we show that such an implementation bears several advantages, such as a reduced computational cost and a lower latency, which are critical for speech recognition applications.
This paper investigates the use of target-speaker automatic speech recognition (TS-ASR) for simultaneous speech recognition and speaker diarization of single-channel dialogue recordings. TS-ASR is a technique to automatically extract and recognize only the speech of a target speaker given a short sample utterance of that speaker. One obvious drawback of TS-ASR is that it cannot be used when the speakers in the recordings are unknown because it requires a sample of the target speakers in advance of decoding. To remove this limitation, we propose an iterative method, in which (i) the estimation of speaker embeddings and (ii) TS-ASR based on the estimated speaker embeddings are alternately executed. We evaluated the proposed method by using very challenging dialogue recordings in which the speaker overlap ratio was over 20%. We confirmed that the proposed method significantly reduced both the word error rate (WER) and diarization error rate (DER). Our proposed method combined with i-vector speaker embeddings ultimately achieved a WER that differed by only 2.1 % from that of TS-ASR given oracle speaker embeddings. Furthermore, our method can solve speaker diarization simultaneously as a by-product and achieved better DER than that of the conventional clustering-based speaker diarization method based on i-vector.
Zero-resource speech technology is a growing research area that aims to develop methods for speech processing in the absence of transcriptions, lexicons, or language modelling text. Early term discovery systems focused on identifying isolated recurring patterns in a corpus, while more recent full-coverage systems attempt to completely segment and cluster the audio into word-like units---effectively performing unsupervised speech recognition. This article presents the first attempt we are aware of to apply such a system to large-vocabulary multi-speaker data. Our system uses a Bayesian modelling framework with segmental word representations: each word segment is represented as a fixed-dimensional acoustic embedding obtained by mapping the sequence of feature frames to a single embedding vector. We compare our system on English and Xitsonga datasets to state-of-the-art baselines, using a variety of measures including word error rate (obtained by mapping the unsupervised output to ground truth transcriptions). Very high word error rates are reported---in the order of 70--80% for speaker-dependent and 80--95% for speaker-independent systems---highlighting the difficulty of this task. Nevertheless, in terms of cluster quality and word segmentation metrics, we show that by imposing a consistent top-down segmentation while also using bottom-up knowledge from detected syllable boundaries, both single-speaker and multi-speaker versions of our system outperform a purely bottom-up single-speaker syllable-based approach. We also show that the discovered clusters can be made less speaker- and gender-specific by using an unsupervised autoencoder-like feature extractor to learn better frame-level features (prior to embedding). Our system's discovered clusters are still less pure than those of unsupervised term discovery systems, but provide far greater coverage.
In this paper, we present an end-to-end training framework for building state-of-the-art end-to-end speech recognition systems. Our training system utilizes a cluster of Central Processing Units(CPUs) and Graphics Processing Units (GPUs). The entire data reading, large scale data augmentation, neural network parameter updates are all performed "on-the-fly". We use vocal tract length perturbation  and an acoustic simulator  for data augmentation. The processed features and labels are sent to the GPU cluster. The Horovod allreduce approach is employed to train neural network parameters. We evaluated the effectiveness of our system on the standard Librispeech corpus  and the 10,000-hr anonymized Bixby English dataset. Our end-to-end speech recognition system built using this training infrastructure showed a 2.44 % WER on test-clean of the LibriSpeech test set after applying shallow fusion with a Transformer language model (LM). For the proprietary English Bixby open domain test set, we obtained a WER of 7.92 % using a Bidirectional Full Attention (BFA) end-to-end model after applying shallow fusion with an RNN-LM. When the monotonic chunckwise attention (MoCha) based approach is employed for streaming speech recognition, we obtained a WER of 9.95 % on the same Bixby open domain test set.
The Mandarin Chinese language is known to be strongly influenced by a rich set of regional accents, while Mandarin speech with each accent is quite low resource. Hence, an important task in Mandarin speech recognition is to appropriately model the acoustic variabilities imposed by accents. In this paper, an investigation of implicit and explicit use of accent information on a range of deep neural network (DNN) based acoustic modelling techniques is conducted. Meanwhile, approaches of multi-accent modelling including multi-style training, multi-accent decision tree state tying, DNN tandem and multi-level adaptive network (MLAN) tandem hidden Markov model (HMM) modelling are combined and compared in this paper. On a low resource accented Mandarin speech recognition task consisting of four regional accents, an improved MLAN tandem HMM systems explicitly leveraging the accent information was proposed and significantly outperformed the baseline accent independent DNN tandem systems by 0.8%-1.5% absolute (6%-9% relative) in character error rate after sequence level discriminative training and adaptation.
This paper describes a variational auto-encoder based non-autoregressive text-to-speech (VAENAR-TTS) model. The autoregressive TTS (AR-TTS) models based on the sequence-to-sequence architecture can generate high-quality speech, but their sequential decoding process can be time-consuming. Recently, non-autoregressive TTS (NAR-TTS) models have been shown to be more efficient with the parallel decoding process. However, these NAR-TTS models rely on phoneme-level durations to generate a hard alignment between the text and the spectrogram. Obtaining duration labels, either through forced alignment or knowledge distillation, is cumbersome. Furthermore, hard alignment based on phoneme expansion can degrade the naturalness of the synthesized speech. In contrast, the proposed model of VAENAR-TTS is an end-to-end approach that does not require phoneme-level durations. The VAENAR-TTS model does not contain recurrent structures and is completely non-autoregressive in both the training and inference phases. Based on the VAE architecture, the alignment information is encoded in the latent variable, and attention-based soft alignment between the text and the latent variable is used in the decoder to reconstruct the spectrogram. Experiments show that VAENAR-TTS achieves state-of-the-art synthesis quality, while the synthesis speed is comparable with other NAR-TTS models.