Speculative Streaming: Fast LLM Inference without Auxiliary Models

Speculative decoding is a prominent technique to speed up the inference of a large target language model based on predictions of an auxiliary draft model. While effective, in application-specific settings, it often involves fine-tuning both draft and target models to achieve high acceptance rates. As the number of downstream tasks grows, these draft models add significant complexity to inference systems. We propose Speculative Streaming, a single-model speculative decoding method that fuses drafting into the target model by changing the fine-tuning objective from next token prediction to future n-gram prediction. Speculative Streaming speeds up decoding by 1.8 - 3.1X in a diverse set of tasks, such as Summarization, Structured Queries, and Meaning Representation, without sacrificing generation quality. Additionally, Speculative Streaming is parameter-efficient. It achieves on-par/higher speed-ups than Medusa-style architectures while using ~10000X fewer extra parameters, making it well-suited for resource-constrained devices.

Conformer-Based Speech Recognition On Extreme Edge-Computing Devices

With increasingly more powerful compute capabilities and resources in today's devices, traditionally compute-intensive automatic speech recognition (ASR) has been moving from the cloud to devices to better protect user privacy. However, it is still challenging to implement on-device ASR on resource-constrained devices, such as smartphones, smart wearables, and other small home automation devices. In this paper, we propose a series of model architecture adaptions, neural network graph transformations, and numerical optimizations to fit an advanced Conformer based end-to-end streaming ASR system on resource-constrained devices without accuracy degradation. We achieve over 5.26 times faster than realtime (0.19 RTF) speech recognition on small wearables while minimizing energy consumption and achieving state-of-the-art accuracy. The proposed methods are widely applicable to other transformer-based server-free AI applications. In addition, we provide a complete theory on optimal pre-normalizers that numerically stabilize layer normalization in any Lp-norm using any floating point precision.

Variable Attention Masking for Configurable Transformer Transducer Speech Recognition

This work studies the use of attention masking in transformer transducer based speech recognition for building a single configurable model for different deployment scenarios. We present a comprehensive set of experiments comparing fixed masking, where the same attention mask is applied at every frame, with chunked masking, where the attention mask for each frame is determined by chunk boundaries, in terms of recognition accuracy and latency. We then explore the use of variable masking, where the attention masks are sampled from a target distribution at training time, to build models that can work in different configurations. Finally, we investigate how a single configurable model can be used to perform both first pass streaming recognition and second pass acoustic rescoring. Experiments show that chunked masking achieves a better accuracy vs latency trade-off compared to fixed masking, both with and without FastEmit. We also show that variable masking improves the accuracy by up to 8% relative in the acoustic re-scoring scenario.

Federated Evaluation and Tuning for On-Device Personalization: System Design & Applications

We describe the design of our federated task processing system. Originally, the system was created to support two specific federated tasks: evaluation and tuning of on-device ML systems, primarily for the purpose of personalizing these systems. In recent years, support for an additional federated task has been added: federated learning (FL) of deep neural networks. To our knowledge, only one other system has been described in literature that supports FL at scale. We include comparisons to that system to help discuss design decisions and attached trade-offs. Finally, we describe two specific large scale personalization use cases in detail to showcase the applicability of federated tuning to on-device personalization and to highlight application specific solutions.

Voice trigger detection from LVCSR hypothesis lattices using bidirectional lattice recurrent neural networks

We propose a method to reduce false voice triggers of a speech-enabled personal assistant by post-processing the hypothesis lattice of a server-side large-vocabulary continuous speech recognizer (LVCSR) via a neural network. We first discuss how an estimate of the posterior probability of the trigger phrase can be obtained from the hypothesis lattice using known techniques to perform detection, then investigate a statistical model that processes the lattice in a more explicitly data-driven, discriminative manner. We propose using a Bidirectional Lattice Recurrent Neural Network (LatticeRNN) for the task, and show that it can significantly improve detection accuracy over using the 1-best result or the posterior.

SNDCNN: Self-normalizing deep CNNs with scaled exponential linear units for speech recognition

Very deep CNNs achieve state-of-the-art results in both computer vision and speech recognition, but are difficult to train. The most popular way to train very deep CNNs is to use shortcut connections (SC) together with batch normalization (BN). Inspired by Self-Normalizing Neural Networks, we propose the self-normalizing deep CNN (SNDCNN) based acoustic model topology, by removing the SC/BN and replacing the typical RELU activations with scaled exponential linear unit (SELU) in ResNet-50. SELU activations make the network self-normalizing and remove the need for both shortcut connections and batch normalization. Compared to ResNet-50, we can achieve the same or lower word error rate (WER) while at the same time improving both training and inference speed by 60%-80%. We also explore other model inference optimizations to further reduce latency for production use.