Ching-Feng Yeh

Tzu-hsun Feng, Annie Dong, Ching-Feng Yeh et al. · meta-ai, mit

We present the SUPERB challenge at SLT 2022, which aims at learning self-supervised speech representation for better performance, generalization, and efficiency. The challenge builds upon the SUPERB benchmark and implements metrics to measure the computation requirements of self-supervised learning (SSL) representation and to evaluate its generalizability and performance across the diverse SUPERB tasks. The SUPERB benchmark provides comprehensive coverage of popular speech processing tasks, from speech and speaker recognition to audio generation and semantic understanding. As SSL has gained interest in the speech community and showed promising outcomes, we envision the challenge to uplevel the impact of SSL techniques by motivating more practical designs of techniques beyond task performance. We summarize the results of 14 submitted models in this paper. We also discuss the main findings from those submissions and the future directions of SSL research.

Ching-Feng Yeh, Po-Yao Huang, Vasu Sharma et al. · meta-ai, mit

We propose Fast Language-Audio Pre-training (FLAP), a self-supervised approach that efficiently and effectively learns aligned audio and language representations through masking, contrastive learning and reconstruction. For efficiency, FLAP randomly drops audio spectrogram tokens, focusing solely on the remaining ones for self-supervision. Through inter-modal contrastive learning, FLAP learns to align paired audio and text representations in a shared latent space. Notably, FLAP leverages multiple augmented views via masking for inter-modal contrast and learns to reconstruct the masked portion of audio tokens. Moreover, FLAP leverages large language models (LLMs) to augment the text inputs, contributing to improved performance. These approaches lead to more robust and informative audio-text representations, enabling FLAP to achieve state-of-the-art (SoTA) performance on audio-text retrieval tasks on AudioCaps (achieving 53.0% R@1) and Clotho (achieving 25.5% R@1).

Anuj Diwan, Ching-Feng Yeh, Wei-Ning Hsu et al.

Automatic speech recognition research focuses on training and evaluating on static datasets. Yet, as speech models are increasingly deployed on personal devices, such models encounter user-specific distributional shifts. To simulate this real-world scenario, we introduce LibriContinual, a continual learning benchmark for speaker-specific domain adaptation derived from LibriVox audiobooks, with data corresponding to 118 individual speakers and 6 train splits per speaker of different sizes. Additionally, current speech recognition models and continual learning algorithms are not optimized to be compute-efficient. We adapt a general-purpose training algorithm NetAug for ASR and create a novel Conformer variant called the DisConformer (Disentangled Conformer). This algorithm produces ASR models consisting of a frozen 'core' network for general-purpose use and several tunable 'augment' networks for speaker-specific tuning. Using such models, we propose a novel compute-efficient continual learning algorithm called DisentangledCL. Our experiments show that the DisConformer models significantly outperform baselines on general ASR i.e. LibriSpeech (15.58% rel. WER on test-other). On speaker-specific LibriContinual they significantly outperform trainable-parameter-matched baselines (by 20.65% rel. WER on test) and even match fully finetuned baselines in some settings.

Ching-Feng Yeh, Wei-Ning Hsu, Paden Tomasello et al.

With the development of hardware for machine learning, newer models often come at the cost of both increased sizes and computational complexity. In effort to improve the efficiency for these models, we apply and investigate recent quantization techniques on speech representation learning models. The quantization techniques were evaluated on the SUPERB benchmark. On the ASR task, with aggressive quantization to 1 bit, we achieved 86.32% storage reduction (184.42 -> 25.23), 88% estimated runtime reduction (1.00 -> 0.12) with increased word error rate (7.06 -> 15.96). In comparison with DistillHuBERT which also aims for model compression, the 2-bit configuration yielded slightly smaller storage (35.84 vs. 46.98), better word error rate (12.68 vs. 13.37) and more efficient estimated runtime (0.15 vs. 0.73).

Sungho Jeon, Ching-Feng Yeh, Hakan Inan et al.

In this paper, we show that a simple self-supervised pre-trained audio model can achieve comparable inference efficiency to more complicated pre-trained models with speech transformer encoders. These speech transformers rely on mixing convolutional modules with self-attention modules. They achieve state-of-the-art performance on ASR with top efficiency. We first show that employing these speech transformers as an encoder significantly improves the efficiency of pre-trained audio models as well. However, our study shows that we can achieve comparable efficiency with advanced self-attention solely. We demonstrate that this simpler approach is particularly beneficial with a low-bit weight quantization technique of a neural network to improve efficiency. We hypothesize that it prevents propagating the errors between different quantized modules compared to recent speech transformers mixing quantized convolution and the quantized self-attention modules.

Dong Wang, Yang Li, Ansong Ni et al.

Synthetic data has become increasingly important for training large language models, especially when real data is scarce, expensive, or privacy-sensitive. Many such generation tasks require coordinated multi-agent workflows, where specialized agents collaborate to produce data that is higher quality, more diverse, and structurally richer. However, existing frameworks for multi-agent synthesis often depend on a centralized orchestrator, creating scalability bottlenecks, or are hardcoded for specific domains, limiting flexibility. We present \textbf{Matrix}, a decentralized framework that represents both control and data flow as serialized messages passed through distributed queues. This peer-to-peer design eliminates the central orchestrator. Each task progresses independently through lightweight agents, while compute-intensive operations, such as LLM inference or containerized environments, are handled by distributed services. Built on Ray, Matrix scales to tens of thousands of concurrent agentic workflows and provides a modular, configurable design that enables easy adaptation to a wide range of data generation workflows. We evaluate Matrix across diverse synthesis scenarios, such as multi-agent collaborative dialogue, web-based reasoning data extraction, and tool-use trajectory generation in customer service environments. In all cases, Matrix achieves $2$--$15\times$ higher data generation throughput under identical hardware resources, without compromising output quality.

Zhipeng Cai, Ching-Feng Yeh, Hu Xu et al. · meta-ai, mit

Vision language models (VLMs) can flexibly address various vision tasks through text interactions. Although successful in semantic understanding, state-of-the-art VLMs including GPT-5 still struggle in understanding 3D from 2D inputs. On the other hand, expert pure vision models achieve super-human accuracy in metric depth estimation, a key 3D understanding task. However, they require task-specific architectures and losses. Such difference motivates us to ask: Can VLMs reach expert-level accuracy without architecture or loss change? We take per-pixel metric depth estimation as the representative task and show that the answer is yes! Surprisingly, comprehensive analysis shows that text-based supervised-finetuning with sparse labels is sufficient for VLMs to unlock strong 3D understanding, no dense prediction head or complex regression/regularization loss is needed. The bottleneck for VLMs lies actually in pixel reference and cross-dataset camera ambiguity, which we address through visual prompting and intrinsic-conditioned augmentation. With much smaller models, our method DepthLM surpasses the accuracy of most advanced VLMs by over 2x, making VLMs for the first time comparable with pure vision models. Interestingly, without explicit enforcement during training, VLMs trained with DepthLM naturally avoids over-smoothing, having much fewer flying points at boundary regions than pure vision models. The simplicity of DepthLM also enables a single VLM to cover various 3D tasks beyond metric depth. Our code and model will be released at the link below.

Yao-Yuan Yang, Moto Hira, Zhaoheng Ni et al.

This document describes version 0.10 of TorchAudio: building blocks for machine learning applications in the audio and speech processing domain. The objective of TorchAudio is to accelerate the development and deployment of machine learning applications for researchers and engineers by providing off-the-shelf building blocks. The building blocks are designed to be GPU-compatible, automatically differentiable, and production-ready. TorchAudio can be easily installed from Python Package Index repository and the source code is publicly available under a BSD-2-Clause License (as of September 2021) at https://github.com/pytorch/audio. In this document, we provide an overview of the design principles, functionalities, and benchmarks of TorchAudio. We also benchmark our implementation of several audio and speech operations and models. We verify through the benchmarks that our implementations of various operations and models are valid and perform similarly to other publicly available implementations.

Hu Xu, Po-Yao Huang, Xiaoqing Ellen Tan et al. · meta-ai, mit

This paper focuses on creating synthetic data to improve the quality of image captions. Existing works typically have two shortcomings. First, they caption images from scratch, ignoring existing alt-text metadata, and second, lack transparency if the captioners' training data (e.g. GPT) is unknown. In this paper, we study a principled approach Altogether based on the key idea to edit and re-align existing alt-texts associated with the images. To generate training data, we perform human annotation where annotators start with the existing alt-text and re-align it to the image content in multiple rounds, consequently constructing captions with rich visual concepts. This differs from prior work that carries out human annotation as a one-time description task solely based on images and annotator knowledge. We train a captioner on this data that generalizes the process of re-aligning alt-texts at scale. Our results show our Altogether approach leads to richer image captions that also improve text-to-image generation and zero-shot image classification tasks.

Yung-Sung Chuang, Yang Li, Dong Wang et al. · meta-ai, mit

Contrastive Language-Image Pretraining (CLIP) is a popular foundation model, supporting from zero-shot classification, retrieval to encoders for multimodal large language models (MLLMs). Although CLIP is successfully trained on billion-scale image-text pairs from the English world, scaling CLIP's training further to learning from the worldwide web data is still challenging: (1) no curation method is available to handle data points from non-English world; (2) the English performance from existing multilingual CLIP is worse than its English-only counterpart, i.e., "curse of multilinguality" that is common in LLMs. Here, we present Meta CLIP 2, the first recipe training CLIP from scratch on worldwide web-scale image-text pairs. To generalize our findings, we conduct rigorous ablations with minimal changes that are necessary to address the above challenges and present a recipe enabling mutual benefits from English and non-English world data. In zero-shot ImageNet classification, Meta CLIP 2 ViT-H/14 surpasses its English-only counterpart by 0.8% and mSigLIP by 0.7%, and surprisingly sets new state-of-the-art without system-level confounding factors (e.g., translation, bespoke architecture changes) on multilingual benchmarks, such as CVQA with 57.4%, Babel-ImageNet with 50.2% and XM3600 with 64.3% on image-to-text retrieval.

Yangyang Shi, Varun Nagaraja, Chunyang Wu et al.

We propose a dynamic encoder transducer (DET) for on-device speech recognition. One DET model scales to multiple devices with different computation capacities without retraining or finetuning. To trading off accuracy and latency, DET assigns different encoders to decode different parts of an utterance. We apply and compare the layer dropout and the collaborative learning for DET training. The layer dropout method that randomly drops out encoder layers in the training phase, can do on-demand layer dropout in decoding. Collaborative learning jointly trains multiple encoders with different depths in one single model. Experiment results on Librispeech and in-house data show that DET provides a flexible accuracy and latency trade-off. Results on Librispeech show that the full-size encoder in DET relatively reduces the word error rate of the same size baseline by over 8%. The lightweight encoder in DET trained with collaborative learning reduces the model size by 25% but still gets similar WER as the full-size baseline. DET gets similar accuracy as a baseline model with better latency on a large in-house data set by assigning a lightweight encoder for the beginning part of one utterance and a full-size encoder for the rest.

Suyoun Kim, Abhinav Arora, Duc Le et al.

Word Error Rate (WER) has been the predominant metric used to evaluate the performance of automatic speech recognition (ASR) systems. However, WER is sometimes not a good indicator for downstream Natural Language Understanding (NLU) tasks, such as intent recognition, slot filling, and semantic parsing in task-oriented dialog systems. This is because WER takes into consideration only literal correctness instead of semantic correctness, the latter of which is typically more important for these downstream tasks. In this study, we propose a novel Semantic Distance (SemDist) measure as an alternative evaluation metric for ASR systems to address this issue. We define SemDist as the distance between a reference and hypothesis pair in a sentence-level embedding space. To represent the reference and hypothesis as a sentence embedding, we exploit RoBERTa, a state-of-the-art pre-trained deep contextualized language model based on the transformer architecture. We demonstrate the effectiveness of our proposed metric on various downstream tasks, including intent recognition, semantic parsing, and named entity recognition.

Xiaohui Zhang, Frank Zhang, Chunxi Liu et al.

In this work, to measure the accuracy and efficiency for a latency-controlled streaming automatic speech recognition (ASR) application, we perform comprehensive evaluations on three popular training criteria: LF-MMI, CTC and RNN-T. In transcribing social media videos of 7 languages with training data 3K-14K hours, we conduct large-scale controlled experimentation across each criterion using identical datasets and encoder model architecture. We find that RNN-T has consistent wins in ASR accuracy, while CTC models excel at inference efficiency. Moreover, we selectively examine various modeling strategies for different training criteria, including modeling units, encoder architectures, pre-training, etc. Given such large-scale real-world streaming ASR application, to our best knowledge, we present the first comprehensive benchmark on these three widely used training criteria across a great many languages.

Jay Mahadeokar, Yuan Shangguan, Duc Le et al.

There is a growing interest in the speech community in developing Recurrent Neural Network Transducer (RNN-T) models for automatic speech recognition (ASR) applications. RNN-T is trained with a loss function that does not enforce temporal alignment of the training transcripts and audio. As a result, RNN-T models built with uni-directional long short term memory (LSTM) encoders tend to wait for longer spans of input audio, before streaming already decoded ASR tokens. In this work, we propose a modification to the RNN-T loss function and develop Alignment Restricted RNN-T (Ar-RNN-T) models, which utilize audio-text alignment information to guide the loss computation. We compare the proposed method with existing works, such as monotonic RNN-T, on LibriSpeech and in-house datasets. We show that the Ar-RNN-T loss provides a refined control to navigate the trade-offs between the token emission delays and the Word Error Rate (WER). The Ar-RNN-T models also improve downstream applications such as the ASR End-pointing by guaranteeing token emissions within any given range of latency. Moreover, the Ar-RNN-T loss allows for bigger batch sizes and 4 times higher throughput for our LSTM model architecture, enabling faster training and convergence on GPUs.

Ching-Feng Yeh, Yongqiang Wang, Yangyang Shi et al.

Attention-based models have been gaining popularity recently for their strong performance demonstrated in fields such as machine translation and automatic speech recognition. One major challenge of attention-based models is the need of access to the full sequence and the quadratically growing computational cost concerning the sequence length. These characteristics pose challenges, especially for low-latency scenarios, where the system is often required to be streaming. In this paper, we build a compact and streaming speech recognition system on top of the end-to-end neural transducer architecture with attention-based modules augmented with convolution. The proposed system equips the end-to-end models with the streaming capability and reduces the large footprint from the streaming attention-based model using augmented memory. On the LibriSpeech dataset, our proposed system achieves word error rates 2.7% on test-clean and 5.8% on test-other, to our best knowledge the lowest among streaming approaches reported so far.

Yongqiang Wang, Yangyang Shi, Frank Zhang et al.

In this paper, we summarize the application of transformer and its streamable variant, Emformer based acoustic model for large scale speech recognition applications. We compare the transformer based acoustic models with their LSTM counterparts on industrial scale tasks. Specifically, we compare Emformer with latency-controlled BLSTM (LCBLSTM) on medium latency tasks and LSTM on low latency tasks. On a low latency voice assistant task, Emformer gets 24% to 26% relative word error rate reductions (WERRs). For medium latency scenarios, comparing with LCBLSTM with similar model size and latency, Emformer gets significant WERR across four languages in video captioning datasets with 2-3 times inference real-time factors reduction.

Yangyang Shi, Yongqiang Wang, Chunyang Wu et al.

This paper proposes an efficient memory transformer Emformer for low latency streaming speech recognition. In Emformer, the long-range history context is distilled into an augmented memory bank to reduce self-attention's computation complexity. A cache mechanism saves the computation for the key and value in self-attention for the left context. Emformer applies a parallelized block processing in training to support low latency models. We carry out experiments on benchmark LibriSpeech data. Under average latency of 960 ms, Emformer gets WER $2.50\%$ on test-clean and $5.62\%$ on test-other. Comparing with a strong baseline augmented memory transformer (AM-TRF), Emformer gets $4.6$ folds training speedup and $18\%$ relative real-time factor (RTF) reduction in decoding with relative WER reduction $17\%$ on test-clean and $9\%$ on test-other. For a low latency scenario with an average latency of 80 ms, Emformer achieves WER $3.01\%$ on test-clean and $7.09\%$ on test-other. Comparing with the LSTM baseline with the same latency and model size, Emformer gets relative WER reduction $9\%$ and $16\%$ on test-clean and test-other, respectively.

Yangyang Shi, Yongqiang Wang, Chunyang Wu et al.

Transformers, originally proposed for natural language processing (NLP) tasks, have recently achieved great success in automatic speech recognition (ASR). However, adjacent acoustic units (i.e., frames) are highly correlated, and long-distance dependencies between them are weak, unlike text units. It suggests that ASR will likely benefit from sparse and localized attention. In this paper, we propose Weak-Attention Suppression (WAS), a method that dynamically induces sparsity in attention probabilities. We demonstrate that WAS leads to consistent Word Error Rate (WER) improvement over strong transformer baselines. On the widely used LibriSpeech benchmark, our proposed method reduced WER by 10%$ on test-clean and 5% on test-other for streamable transformers, resulting in a new state-of-the-art among streaming models. Further analysis shows that WAS learns to suppress attention of non-critical and redundant continuous acoustic frames, and is more likely to suppress past frames rather than future ones. It indicates the importance of lookahead in attention-based ASR models.

Chunyang Wu, Yongqiang Wang, Yangyang Shi et al.

Transformer-based acoustic modeling has achieved great suc-cess for both hybrid and sequence-to-sequence speech recogni-tion. However, it requires access to the full sequence, and thecomputational cost grows quadratically with respect to the in-put sequence length. These factors limit its adoption for stream-ing applications. In this work, we proposed a novel augmentedmemory self-attention, which attends on a short segment of theinput sequence and a bank of memories. The memory bankstores the embedding information for all the processed seg-ments. On the librispeech benchmark, our proposed methodoutperforms all the existing streamable transformer methods bya large margin and achieved over 15% relative error reduction,compared with the widely used LC-BLSTM baseline. Our find-ings are also confirmed on some large internal datasets.

Yi-Chen Chen, Zhaojun Yang, Ching-Feng Yeh et al.

As one of the major sources in speech variability, accents have posed a grand challenge to the robustness of speech recognition systems. In this paper, our goal is to build a unified end-to-end speech recognition system that generalizes well across accents. For this purpose, we propose a novel pre-training framework AIPNet based on generative adversarial nets (GAN) for accent-invariant representation learning: Accent Invariant Pre-training Networks. We pre-train AIPNet to disentangle accent-invariant and accent-specific characteristics from acoustic features through adversarial training on accented data for which transcriptions are not necessarily available. We further fine-tune AIPNet by connecting the accent-invariant module with an attention-based encoder-decoder model for multi-accent speech recognition. In the experiments, our approach is compared against four baselines including both accent-dependent and accent-independent models. Experimental results on 9 English accents show that the proposed approach outperforms all the baselines by 2.3 \sim 4.5% relative reduction on average WER when transcriptions are available in all accents and by 1.6 \sim 6.1% relative reduction when transcriptions are only available in US accent.

Mahaveer Jain, Kjell Schubert, Jay Mahadeokar et al.

Neural transducer-based systems such as RNN Transducers (RNN-T) for automatic speech recognition (ASR) blend the individual components of a traditional hybrid ASR systems (acoustic model, language model, punctuation model, inverse text normalization) into one single model. This greatly simplifies training and inference and hence makes RNN-T a desirable choice for ASR systems. In this work, we investigate use of RNN-T in applications that require a tune-able latency budget during inference time. We also improved the decoding speed of the originally proposed RNN-T beam search algorithm. We evaluated our proposed system on English videos ASR dataset and show that neural RNN-T models can achieve comparable WER and better computational efficiency compared to a well tuned hybrid ASR baseline.

Ching-Feng Yeh, Jay Mahadeokar, Kaustubh Kalgaonkar et al.

We explore options to use Transformer networks in neural transducer for end-to-end speech recognition. Transformer networks use self-attention for sequence modeling and comes with advantages in parallel computation and capturing contexts. We propose 1) using VGGNet with causal convolution to incorporate positional information and reduce frame rate for efficient inference 2) using truncated self-attention to enable streaming for Transformer and reduce computational complexity. All experiments are conducted on the public LibriSpeech corpus. The proposed Transformer-Transducer outperforms neural transducer with LSTM/BLSTM networks and achieved word error rates of 6.37 % on the test-clean set and 15.30 % on the test-other set, while remaining streamable, compact with 45.7M parameters for the entire system, and computationally efficient with complexity of O(T), where T is input sequence length.

Sining Sun, Ching-Feng Yeh, Mei-Yuh Hwang et al.

In this paper, we propose a domain adversarial training (DAT) algorithm to alleviate the accented speech recognition problem. In order to reduce the mismatch between labeled source domain data ("standard" accent) and unlabeled target domain data (with heavy accents), we augment the learning objective for a Kaldi TDNN network with a domain adversarial training (DAT) objective to encourage the model to learn accent-invariant features. In experiments with three Mandarin accents, we show that DAT yields up to 7.45% relative character error rate reduction when we do not have transcriptions of the accented speech, compared with the baseline trained on standard accent data only. We also find a benefit from DAT when used in combination with training from automatic transcriptions on the accented data. Furthermore, we find that DAT is superior to multi-task learning for accented speech recognition.

Sining Sun, Ching-Feng Yeh, Mari Ostendorf et al.

This paper explores the use of adversarial examples in training speech recognition systems to increase robustness of deep neural network acoustic models. During training, the fast gradient sign method is used to generate adversarial examples augmenting the original training data. Different from conventional data augmentation based on data transformations, the examples are dynamically generated based on current acoustic model parameters. We assess the impact of adversarial data augmentation in experiments on the Aurora-4 and CHiME-4 single-channel tasks, showing improved robustness against noise and channel variation. Further improvement is obtained when combining adversarial examples with teacher/student training, leading to a 23% relative word error rate reduction on Aurora-4.