The recent wave of audio foundation models (FMs) could provide new capabilities for conversational modeling. However, there have been limited efforts to evaluate these audio FMs comprehensively on their ability to have natural and interactive conversations. To engage in meaningful conversation with the end user, we would want the FMs to additionally perform a fluent succession of turns without too much overlapping speech or long stretches of silence. Inspired by this, we ask whether the recently proposed audio FMs can understand, predict, and perform turn-taking events? To answer this, we propose a novel evaluation protocol that can assess spoken dialog system's turn-taking capabilities using a supervised model as a judge that has been trained to predict turn-taking events in human-human conversations. Using this protocol, we present the first comprehensive user study that evaluates existing spoken dialogue systems on their ability to perform turn-taking events and reveal many interesting insights, such as they sometimes do not understand when to speak up, can interrupt too aggressively and rarely backchannel. We further evaluate multiple open-source and proprietary audio FMs accessible through APIs on carefully curated test benchmarks from Switchboard to measure their ability to understand and predict turn-taking events and identify significant room for improvement. We will open source our evaluation platform to promote the development of advanced conversational AI systems.

Direct Preference Optimization (DPO) has been widely adopted for preference alignment of Large Language Models (LLMs) due to its simplicity and effectiveness. However, DPO is derived as a bandit problem in which the whole response is treated as a single arm, ignoring the importance differences between tokens, which may affect optimization efficiency and make it difficult to achieve optimal results. In this work, we propose that the optimal data for DPO has equal expected rewards for each token in winning and losing responses, as there is no difference in token importance. However, since the optimal dataset is unavailable in practice, we propose using the original dataset for importance sampling to achieve unbiased optimization. Accordingly, we propose a token-level importance sampling DPO objective named TIS-DPO that assigns importance weights to each token based on its reward. Inspired by previous works, we estimate the token importance weights using the difference in prediction probabilities from a pair of contrastive LLMs. We explore three methods to construct these contrastive LLMs: (1) guiding the original LLM with contrastive prompts, (2) training two separate LLMs using winning and losing responses, and (3) performing forward and reverse DPO training with winning and losing responses. Experiments show that TIS-DPO significantly outperforms various baseline methods on harmlessness and helpfulness alignment and summarization tasks. We also visualize the estimated weights, demonstrating their ability to identify key token positions.

Aiwei Liu, Haoping Bai, Zhiyun Lu, Xiang Kong, Xiaoming Wang, Jiulong Shan, Meng Cao, Lijie Wen. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

The recurrent neural network transducer (RNN-T) has recently become the mainstream end-to-end approach for streaming automatic speech recognition (ASR). To estimate the output distributions over subword units, RNN-T uses a fully connected layer as the joint network to fuse the acoustic representations extracted using the acoustic encoder with the text representations obtained using the prediction network based on the previous subword units. In this paper, we propose to use gating, bilinear pooling, and a combination of them in the joint network to produce more expressive representations to feed into the output layer. A regularisation method is also proposed to enable better acoustic encoder training by reducing the gradients back-propagated into the prediction network at the beginning of RNN-T training. Experimental results on a multilingual ASR setting for voice search over nine languages show that the joint use of the proposed methods can result in 4%–5% relative word error rate reductions with only a few million extra parameters.

Improving the performance of end-to-end ASR models on long utterances ranging from minutes to hours in length is an ongoing challenge in speech recognition. A common solution is to segment the audio in advance using a separate voice activity detector (VAD) that decides segment boundary locations based purely on acoustic speech/non-speech information. VAD segmenters, however, may be sub-optimal for real-world speech where, e.g., a complete sentence that should be taken as a whole may contain hesitations in the middle ("set an alarm for... 5 o'clock"). We propose to replace the VAD with an end-to-end ASR model capable of predicting segment boundaries in a streaming fashion, allowing the segmentation decision to be conditioned not only on better acoustic features but also on semantic features from the decoded text with negligible extra computation. In experiments on real world long-form audio (YouTube) with lengths of up to 30 minutes, we demonstrate 8.5% relative WER improvement and 250 ms reduction in median end-of-segment latency compared to the VAD segmenter baseline on a state-of-the-art Conformer RNN-T model.

Self-supervised learning of speech representations has achieved impressive results in improving automatic speech recognition (ASR).In this paper, we show that data selection is important for self-supervised learning.We propose a simple and effective unsupervised data selection method which selects acoustically similar speech to a target domain.It takes the discrete speech representation available in common self-supervised learning frameworks as input, and applies a contrastive data selection method on the discrete tokens.Through extensive empirical studies we show that our proposed method reduces the amount of required pre-training data and improves the downstream ASR performance.Pre-training on a selected subset of 6% of the general data pool results in 11.8% relative improvements in LibriSpeech test-other compared to pre-training on the full set.On Multilingual LibriSpeech French, German, and Spanish test sets, selecting 6% data for pre-training reduces word error rate by more than 15% relatively compared to the full set, and achieves competitive results compared to current state-of-theart performances.

Streaming end-to-end automatic speech recognition (ASR) models are widely used on smart speakers and on-device applications. Since these models are expected to transcribe speech with minimal latency, they are constrained to be causal with no future context, compared to their non-streaming counterparts. Consequently, streaming models usually perform worse than non-streaming models. We propose a novel and effective learning method by leveraging a non-streaming ASR model as a teacher to generate transcripts on an arbitrarily large data set, which is then used to distill knowledge into streaming ASR models. This way, we scale the training of streaming models to up to 3 million hours of YouTube audio. Experiments show that our approach can significantly reduce the word error rate (WER) of RNN-T models not only on LibriSpeech but also on YouTube data in four languages. For example, in French, we are able to reduce the WER by 16.4% relatively to a baseline streaming model by leveraging a non-streaming teacher model trained on the same amount of labeled data as the baseline.

Although end-to-end automatic speech recognition (e2e ASR) models are widely deployed in many applications, there have been very few studies to understand models' robustness against adversarial perturbations.In this paper, we explore whether a targeted universal perturbation vector exists for e2e ASR models.Our goal is to find perturbations that can mislead the models to predict the given targeted transcript such as "thank you" or empty string on any input utterance.We study two different attacks, namely additive and prepending perturbations, and their performances on the state-of-the-art LAS, CTC and RNN-T models.We find that LAS is the most vulnerable to perturbations among the three models.RNN-T is more robust against additive perturbations, especially on long utterances.And CTC is robust against both additive and prepending perturbations.To attack RNN-T, we find prepending perturbation is more effective than the additive perturbation, and can mislead the models to predict the same short target on utterances of arbitrary length.

In this paper, we propose to use pre-trained features from end-to-end ASR models to solve speech sentiment analysis as a down-stream task. We show that end-to-end ASR features, which integrate both acoustic and text information from speech, achieve promising results. We use RNN with self-attention as the sentiment classifier, which also provides an easy visualization through attention weights to help interpret model predictions. We use well benchmarked IEMOCAP dataset and a new large-scale speech sentiment dataset SWBD-sentiment for evaluation. Our approach improves the-state-of-the-art accuracy on IEMOCAP from 66.6% to 71.7%, and achieves an accuracy of 70.10% on SWBD-sentiment with more than 49,500 utterances.

We present a multimodal corpus for sentiment analysis based on the existing Switchboard-1 Telephone Speech Corpus released by the Linguistic Data Consortium. This corpus extends the Switchboard-1 Telephone Speech Corpus by adding sentiment labels from 3 different human annotators for every transcript segment. Each sentiment label can be one of three options: positive, negative, and neutral. Annotators are recruited using Google Cloud’s data labeling service and the labeling task was conducted over the internet. The corpus contains a total of 49500 labeled speech segments covering 140 hours of audio. To the best of our knowledge, this is the largest multimodal Corpus for sentiment analysis that includes both speech and text features.

Hyper-parameter tuning is of crucial importance for real-world machine learning applications. While existing works mainly focus on speeding up the tuning process, we propose to study the problem of hyper-parameter tuning under a budget constraint, which is a more realistic scenario in developing large-scale systems. We formulate the task into a sequential decision making problem and propose a solution, which uses a Bayesian belief model to predict future performances, and an action-value function to plan and select the next configuration to run. With long term prediction and planning capability, our method is able to early stop unpromising configurations, and adapt the tuning behaviors to different constraints. Experiment results show that our method outperforms existing algorithms, including the-state-of-the-art one, on real-world tuning tasks across a range of different budgets.

We study the performance of kernel methods on the acoustic modeling task for automatic speech recognition, and compare their performance to deep neural networks (DNNs). To scale the kernel methods to large data sets, we use the random Fourier feature method of Rahimi and Recht (2007). We propose two novel techniques for improving the performance of kernel acoustic models. First, we propose a simple but effective feature selection method which reduces the number of random features required to attain a fixed level of performance. Second, we present a number of metrics which correlate strongly with speech recognition performance when computed on the heldout set; we attain improved performance by using these metrics to decide when to stop training. Additionally, we show that the linear bottleneck method of Sainath et al. (2013a) improves the performance of our kernel models significantly, in addition to speeding up training and making the models more compact. Leveraging these three methods, the kernel methods attain token error rates between $0.5\%$ better and $0.1\%$ worse than fully-connected DNNs across four speech recognition data sets, including the TIMIT and Broadcast News benchmark tasks.

We study large-scale kernel methods for acoustic modeling and compare to DNNs on performance metrics related to both acoustic modeling and recognition. Measuring perplexity and frame-level classification accuracy, kernel-based acoustic models are as effective as their DNN counterparts. However, on token-error-rates DNN models can be significantly better. We have discovered that this might be attributed to DNN's unique strength in reducing both the perplexity and the entropy of the predicted posterior probabilities. Motivated by our findings, we propose a new technique, entropy regularized perplexity, for model selection. This technique can noticeably improve the recognition performance of both types of models, and reduces the gap between them. While effective on Broadcast News, this technique could be also applicable to other tasks.

Recurrent neural networks (RNNs), including long short-term memory (LSTM) RNNs, have produced state-of-the-art results on a variety of speech recognition tasks. However, these models are often too large in size for deployment on mobile devices with memory and latency constraints. In this work, we study mechanisms for learning compact RNNs and LSTMs via low-rank factorizations and parameter sharing schemes. Our goal is to investigate redundancies in recurrent architectures where compression can be admitted without losing performance. A hybrid strategy of using structured matrices in the bottom layers and shared low-rank factors on the top layers is found to be particularly effective, reducing the parameters of a standard LSTM by 75%, at a small cost of 0.3% increase in WER, on a 2,000-hr English Voice Search task.