Apple is attendingInterspeech 2019, the world’s largest conference on the science and technology of spoken language processing. The conference, of which Apple is a Platinum Sponsor, will take place in Graz, Austria from September (th to) th. For Interspeech attendees, join the authors of our accepted papers at our booth to learn more about the great speech research happening at Apple.
Apple continues to build cutting-edge technology in the space of machine hearing, speech recognition, natural language processing, machine translation, text-to-speech, and artificial intelligence, improving the lives of millions of customers every day. If you’re interested in opportunities to make an impact on Apple products through machine learning research and development, check out our teams atJobs at Apple.
Vikramjit Mitra, Sue Booker, Erik Marchi, David Scott Farrar, Ute Dorothea Peitz, Bridget Cheng, Ermine Teves, Anuj Mehta, Devang Naik
Millions of people reach out to digital assistants such as Siri every day, asking for information, making phone calls, seeking assistance, and much more. The expectation is that such assistants should understand the intent of the user’s query. Detecting the intent of a query from a short, isolated utterance is a difficult task. Intent cannot always be obtained from speech-recognized transcriptions. A transcription-driven approach can interpret what has been said but fails to acknowledge how it has been said, and as a consequence, may ignore the expression present in the voice. Our work investigates whether a system can reliably detect vocal expression in queries using acoustic and paralinguistic embedding. Results show that the proposed method offers a relative equal error rate (EER) decrease of 60% compared to a bag-of-word based system, corroborating that expression is significantly represented by vocal attributes, rather than being purely lexical. Addition of emotion embedding helped to reduce the EER by 30% relative to the acoustic embedding, demonstrating the relevance of emotion in expressive voice.
Gautam Mantena, Ozlem Kalinli, Ossama Abdel-Hamid, Don McAllaster
In this paper, we tackle the problem of handling narrowband and wideband speech by building a single acoustic model (AM), also called mixed bandwidth AM. In the proposed approach, an auxiliary input feature is used to provide the bandwidth information to the model, and bandwidth embeddings are jointly learned as part of acoustic model training. Experimental evaluations show that using bandwidth embeddings helps the model to handle the variability of the narrow and wideband speech, and makes it possible to train a mixed-bandwidth AM. Furthermore, we propose to use parallel convolutional layers to handle the mismatch between the narrow and wideband speech better, where separate convolution layers are used for each type of input speech signal. Our best system achieves 13% relative improvement on narrowband speech, while not degrading on wideband speech.
Xizi Wei, Melvyn Hunt, Adrian Skilling
A deep neural network (DNN) -based model has been developed to predict non-parametric distributions of durations of phonemes in specified phonetic contexts and used to explore which factors influence durations most. Major factors in US English are pre-pausal lengthening, lexical stress, and speaking rate. The model can be used to check that text-to-speech (TTS) training speech follows the script and words are pronounced as expected. Duration prediction is poorer with training speech for automatic speech recognition (ASR) because the training corpus typically consists of single utterances from many speakers and is often noisy or casually spoken. Low probability durations in ASR training material nevertheless mostly correspond to non-standard speech, with some having disfluencies. Children’s speech is disproportionately present in these utterances, since children show much more variation in timing.
Ernest Pusateri, Christophe Van Gysel, Rami Botros, Sameer Badaskar, Mirko Hannemann, Youssef Oualil, Ilya Oparin*****
In this work, we uncover a theoretical connection between two language model interpolation techniques, count merging and Bayesian interpolation. We compare these techniques as well as linear interpolation in three scenarios with abundant training data per component model. Consistent with prior work, we show that both count merging and Bayesian interpolation outperform linear interpolation. We include the first (to our knowledge) published comparison of count merging and Bayesian interpolation, showing that the two techniques perform similarly. Finally, we argue that other considerations will make Bayesian interpolation the preferred approach in most circumstances.
Xi Chen, Adithya Sagar, Justine Kao, Tony Li, Christopher Klein, Stephen Pulman, Ashish Garg, Jason Williams
We describe a method for selecting relevant new training data for the LSTM-based domain selection component of our personal assistant system. Adding more annotated training data for any ML system typically improves accuracy, but only if it provides examples not already adequately covered in the existing data. However, obtaining, selecting, and labeling relevant data is expensive. This work presents a simple technique that automatically identifies new helpful examples suitable for human annotation. Our experimental results show that the proposed method, compared with random-selection and entropy-based methods, leads to higher accuracy improvements given a fixed annotation budget. Although developed and tested in the setting of a commercial intelligent assistant, the technique is of wider applicability.
Katherine Metcalf, Barry-John Theobald, Garrett Weinberg, Robert Lee, Ing-Marie Jonsson, Russ Webb, Nicholas Apostoloff
We describe experiments towards building a conversational digital assistant that considers the preferred conversational style of the user. In particular, these experiments are designed to measure whether users prefer and trust an assistant whose conversational style matches their own. To this end we conducted a user study where subjects interacted with a digital assistant whose response either matched their conversational style, or did not. We found that people prefer a digital assistant that mirrors their “chattiness”, and this preference can reliably be detected.
Jian Yao, Ahmad Al-Dahle
In this paper, we propose the coarse-to-fine optimization for the task of speech enhancement. Cosine similarity loss has proven to be an effective metric to measure similarity of speech signals. However, due to the large variance of the enhanced speech with even the same cosine similarity loss in high dimensional space, a deep neural network learnt with this loss might not be able to predict enhanced speech with good quality. Our coarse-to-fine strategy optimizes the cosine similarity loss for different granularities so that more constraints are added to the prediction from high dimension to relatively low dimension. In this way, the enhanced speech will better resemble the clean speech. Experimental results show the effectiveness of our proposed coarse-to-fine optimization in both discriminative models and generative models. Moreover, we apply the coarse-to-fine strategy to the adversarial loss in generative adversarial network (GAN) and propose dynamic perceptual loss, which dynamically computes the adversarial loss from coarse resolution to fine resolution. Dynamic perceptual loss further improves the accuracy and achieves state-of-the-art results compared to other generative models.