Search for: All records

While Speech Foundation Models (SFMs) excel in various speech tasks, their performance for low-resource tasks such as child Automatic Speech Recognition (ASR) is hampered by limited pretraining data. To address this, we explore different model merging techniques to leverage knowledge from models trained on larger, more diverse speech corpora. This paper also introduces Selective Attention (SA) Merge, a novel method that selectively merges task vectors from attention matrices to enhance SFM performance on low-resource tasks. Experiments on the MyST database show significant reductions in relative word error rate of up to 14%, outperforming existing model merging and data augmentation techniques. By combining data augmentation techniques with SA Merge, we achieve a new state-of-the-art WER of 8.69 on the MyST database for the Whisper-small model, highlighting the potential of SA Merge for improving low-resource ASR. 

Recently, speech foundation models have gained popularity due to their superiority in finetuning downstream ASR tasks. However, models finetuned on certain domains, such as LibriSpeech (adult read speech), behave poorly on other domains (child or noisy speech). One solution could be collecting as much labeled and diverse data as possible for joint finetuning on various domains. However, collecting target domain speech-text paired data and retraining the model is often costly and computationally expensive. In this paper, we introduce a simple yet effective method, speech only adaptation (SOA), based on speech foundation models (Wav2vec 2.0), which requires only speech input data from the target domain. Specifically, the Wav2vec 2.0 feature encoder is continually pretrained with the Wav2vec 2.0 loss on both the source and target domain data for domain adaptation, while the contextual encoder is frozen. Compared to a source domain finetuned model with the feature encoder being frozen during training, we find that replacing the frozen feature encoder with the adapted one provides significant WER improvements to the target domain while preserving the performance of the source domain. The effectiveness of SOA is examined on various low resource or domain mismatched ASR settings, including adult-child and clean-noisy speech. 

This paper evaluates an innovative framework for spoken dialect density prediction on children's and adults' African American English. A speaker's dialect density is defined as the frequency with which dialect-specific language characteristics occur in their speech. Rather than treating the presence or absence of a target dialect in a user's speech as a binary decision, instead, a classifier is trained to predict the level of dialect density to provide a higher degree of specificity in downstream tasks. For this, self-supervised learning representations from HuBERT, handcrafted grammar-based features extracted from ASR transcripts, prosodic features, and other feature sets are experimented with as the input to an XGBoost classifier. Then, the classifier is trained to assign dialect density labels to short recorded utterances. High dialect density level classification accuracy is achieved for child and adult speech and demonstrated robust performance across age and regional varieties of dialect. Additionally, this work is used as a basis for analyzing which acoustic and grammatical cues affect machine perception of dialect. 

This paper presents a novel dataset (CORAAL QA) and framework for audio question-answering from long audio recordings contain- ing spontaneous speech. The dataset introduced here provides sets of questions that can be factually answered from short spans of a long audio files (typically 30min to 1hr) from the Corpus of Re- gional African American Language. Using this dataset, we divide the audio recordings into 60 second segments, automatically tran- scribe each segment, and use PLDA scoring of BERT-based seman- tic embeddings to rank the relevance of ASR transcript segments in answering the target question. In order to improve this framework through data augmentation, we use large language models including ChatGPT and Llama 2 to automatically generate further training ex- amples and show how prompt engineering can be optimized for this process. By creatively leveraging knowledge from large-language models, we achieve state-of-the-art question-answering performance in this information retrieval task. 

Non-autoregressive automatic speech recognition (NASR) models have gained attention due to their parallelism and fast inference. The encoder-based NASR, e.g. connectionist temporal classification (CTC), can be initialized from the speech foundation models (SFM) but does not account for any dependencies among intermediate tokens. The encoder-decoder-based NASR, like CTC alignment-based single-step non-autoregressive transformer (CASS-NAT), can mitigate the dependency problem but is not able to efficiently integrate SFM. Inspired by the success of recent work of speech-text joint pre-training with a shared transformer encoder, we propose a new encoder-based NASR, UniEnc-CASSNAT, to combine the advantages of CTC and CASS-NAT. UniEnc-CASSNAT consists of only an encoder as the major module, which can be the SFM. The encoder plays the role of both the CASS-NAT encoder and decoder by two forward passes. The first pass of the encoder accepts the speech signal as input, while the concatenation of the speech signal and the token-level acoustic embedding is used as the input for the second pass. Examined on the Librispeech 100 h, MyST, and Aishell1 datasets, the proposed UniEnc-CASSNAT achieves state-of-the-art NASR results and is better or comparable to CASS-NAT with only an encoder and hence, fewer model parameters.