Learnings from curating a trustworthy, well-annotated, and useful dataset of disordered English speech

Project Euphonia, a Google initiative, is dedicated to improving automatic speech recognition (ASR) of disordered speech. A central objective of the project is to create a large, high-quality, and diverse speech corpus. This report describes the project's latest advancements in data collection and annotation methodologies, such as expanding speaker diversity in the database, adding human-reviewed transcript corrections and audio quality tags to 350K (of the 1.2M total) audio recordings, and amassing a comprehensive set of metadata (including more than 40 speech characteristic labels) for over 75\% of the speakers in the database. We report on the impact of transcript corrections on our machine-learning (ML) research, inter-rater variability of assessments of disordered speech patterns, and our rationale for gathering speech metadata. We also consider the limitations of using automated off-the-shelf annotation methods for assessing disordered speech.

Detecting Hallucination and Coverage Errors in Retrieval Augmented Generation for Controversial Topics

We explore a strategy to handle controversial topics in LLM-based chatbots based on Wikipedia's Neutral Point of View (NPOV) principle: acknowledge the absence of a single true answer and surface multiple perspectives. We frame this as retrieval augmented generation, where perspectives are retrieved from a knowledge base and the LLM is tasked with generating a fluent and faithful response from the given perspectives. As a starting point, we use a deterministic retrieval system and then focus on common LLM failure modes that arise during this approach to text generation, namely hallucination and coverage errors. We propose and evaluate three methods to detect such errors based on (1) word-overlap, (2) salience, and (3) LLM-based classifiers. Our results demonstrate that LLM-based classifiers, even when trained only on synthetic errors, achieve high error detection performance, with ROC AUC scores of 95.3% for hallucination and 90.5% for coverage error detection on unambiguous error cases. We show that when no training data is available, our other methods still yield good results on hallucination (84.0%) and coverage error (85.2%) detection.

Parameter Efficient Tuning Allows Scalable Personalization of LLMs for Text Entry: A Case Study on Abbreviation Expansion

Abbreviation expansion is a strategy used to speed up communication by limiting the amount of typing and using a language model to suggest expansions. Here we look at personalizing a Large Language Model's (LLM) suggestions based on prior conversations to enhance the relevance of predictions, particularly when the user data is small (~1000 samples). Specifically, we compare fine-tuning, prompt-tuning, and retrieval augmented generation of expanded text suggestions for abbreviated inputs. Our case study with a deployed 8B parameter LLM on a real user living with ALS, and experiments on movie character personalization indicates that (1) customization may be necessary in some scenarios and prompt-tuning generalizes well to those, (2) fine-tuning on in-domain data (with as few as 600 samples) still shows some gains, however (3) retrieval augmented few-shot selection also outperforms fine-tuning. (4) Parameter efficient tuning allows for efficient and scalable personalization. For prompt-tuning, we also find that initializing the learned "soft-prompts" to user relevant concept tokens leads to higher accuracy than random initialization.

Using Large Language Models to Accelerate Communication for Users with Severe Motor Impairments

Finding ways to accelerate text input for individuals with profound motor impairments has been a long-standing area of research. Closing the speed gap for augmentative and alternative communication (AAC) devices such as eye-tracking keyboards is important for improving the quality of life for such individuals. Recent advances in neural networks of natural language pose new opportunities for re-thinking strategies and user interfaces for enhanced text-entry for AAC users. In this paper, we present SpeakFaster, consisting of large language models (LLMs) and a co-designed user interface for text entry in a highly-abbreviated form, allowing saving 57% more motor actions than traditional predictive keyboards in offline simulation. A pilot study with 19 non-AAC participants typing on a mobile device by hand demonstrated gains in motor savings in line with the offline simulation, while introducing relatively small effects on overall typing speed. Lab and field testing on two eye-gaze typing users with amyotrophic lateral sclerosis (ALS) demonstrated text-entry rates 29-60% faster than traditional baselines, due to significant saving of expensive keystrokes achieved through phrase and word predictions from context-aware LLMs. These findings provide a strong foundation for further exploration of substantially-accelerated text communication for motor-impaired users and demonstrate a direction for applying LLMs to text-based user interfaces.

Towards Agile Text Classifiers for Everyone

Text-based safety classifiers are widely used for content moderation and increasingly to tune generative language model behavior - a topic of growing concern for the safety of digital assistants and chatbots. However, different policies require different classifiers, and safety policies themselves improve from iteration and adaptation. This paper introduces and evaluates methods for agile text classification, whereby classifiers are trained using small, targeted datasets that can be quickly developed for a particular policy. Experimenting with 7 datasets from three safety-related domains, comprising 15 annotation schemes, led to our key finding: prompt-tuning large language models, like PaLM 62B, with a labeled dataset of as few as 80 examples can achieve state-of-the-art performance. We argue that this enables a paradigm shift for text classification, especially for models supporting safer online discourse. Instead of collecting millions of examples to attempt to create universal safety classifiers over months or years, classifiers could be tuned using small datasets, created by individuals or small organizations, tailored for specific use cases, and iterated on and adapted in the time-span of a day.

An analysis of degenerating speech due to progressive dysarthria on ASR performance

Although personalized automatic speech recognition (ASR) models have recently been designed to recognize even severely impaired speech, model performance may degrade over time for persons with degenerating speech. The aims of this study were to (1) analyze the change of performance of ASR over time in individuals with degrading speech, and (2) explore mitigation strategies to optimize recognition throughout disease progression. Speech was recorded by four individuals with degrading speech due to amyotrophic lateral sclerosis (ALS). Word error rates (WER) across recording sessions were computed for three ASR models: Unadapted Speaker Independent (U-SI), Adapted Speaker Independent (A-SI), and Adapted Speaker Dependent (A-SD or personalized). The performance of all three models degraded significantly over time as speech became more impaired, but the performance of the A-SD model improved markedly when it was updated with recordings from the severe stages of speech progression. Recording additional utterances early in the disease before speech degraded significantly did not improve the performance of A-SD models. Overall, our findings emphasize the importance of continuous recording (and model retraining) when providing personalized models for individuals with progressive speech impairments.

Assessing ASR Model Quality on Disordered Speech using BERTScore

Word Error Rate (WER) is the primary metric used to assess automatic speech recognition (ASR) model quality. It has been shown that ASR models tend to have much higher WER on speakers with speech impairments than typical English speakers. It is hard to determine if models can be be useful at such high error rates. This study investigates the use of BERTScore, an evaluation metric for text generation, to provide a more informative measure of ASR model quality and usefulness. Both BERTScore and WER were compared to prediction errors manually annotated by Speech Language Pathologists for error type and assessment. BERTScore was found to be more correlated with human assessment of error type and assessment. BERTScore was specifically more robust to orthographic changes (contraction and normalization errors) where meaning was preserved. Furthermore, BERTScore was a better fit of error assessment than WER, as measured using an ordinal logistic regression and the Akaike's Information Criterion (AIC). Overall, our findings suggest that BERTScore can complement WER when assessing ASR model performance from a practical perspective, especially for accessibility applications where models are useful even at lower accuracy than for typical speech.

Context-Aware Abbreviation Expansion Using Large Language Models

Motivated by the need for accelerating text entry in augmentative and alternative communication (AAC) for people with severe motor impairments, we propose a paradigm in which phrases are abbreviated aggressively as primarily word-initial letters. Our approach is to expand the abbreviations into full-phrase options by leveraging conversation context with the power of pretrained large language models (LLMs). Through zero-shot, few-shot, and fine-tuning experiments on four public conversation datasets, we show that for replies to the initial turn of a dialog, an LLM with 64B parameters is able to exactly expand over 70% of phrases with abbreviation length up to 10, leading to an effective keystroke saving rate of up to about 77% on these exact expansions. Including a small amount of context in the form of a single conversation turn more than doubles abbreviation expansion accuracies compared to having no context, an effect that is more pronounced for longer phrases. Additionally, the robustness of models against typo noise can be enhanced through fine-tuning on noisy data.

Personalized Automatic Speech Recognition Trained on Small Disordered Speech Datasets

This study investigates the performance of personalized automatic speech recognition (ASR) for recognizing disordered speech using small amounts of per-speaker adaptation data. We trained personalized models for 195 individuals with different types and severities of speech impairment with training sets ranging in size from <1 minute to 18-20 minutes of speech data. Word error rate (WER) thresholds were selected to determine Success Percentage (the percentage of personalized models reaching the target WER) in different application scenarios. For the home automation scenario, 79% of speakers reached the target WER with 18-20 minutes of speech; but even with only 3-4 minutes of speech, 63% of speakers reached the target WER. Further evaluation found similar improvement on test sets with conversational and out-of-domain, unprompted phrases. Our results demonstrate that with only a few minutes of recordings, individuals with disordered speech could benefit from personalized ASR.

Residual Adapters for Parameter-Efficient ASR Adaptation to Atypical and Accented Speech

Automatic Speech Recognition (ASR) systems are often optimized to work best for speakers with canonical speech patterns. Unfortunately, these systems perform poorly when tested on atypical speech and heavily accented speech. It has previously been shown that personalization through model fine-tuning substantially improves performance. However, maintaining such large models per speaker is costly and difficult to scale. We show that by adding a relatively small number of extra parameters to the encoder layers via so-called residual adapter, we can achieve similar adaptation gains compared to model fine-tuning, while only updating a tiny fraction (less than 0.5%) of the model parameters. We demonstrate this on two speech adaptation tasks (atypical and accented speech) and for two state-of-the-art ASR architectures.