Tatiana Likhomanenko

Shuangfei Zhai, Tatiana Likhomanenko, Etai Littwin et al. · apple-ml, meta-ai

Training stability is of great importance to Transformers. In this work, we investigate the training dynamics of Transformers by examining the evolution of the attention layers. In particular, we track the attention entropy for each attention head during the course of training, which is a proxy for model sharpness. We identify a common pattern across different architectures and tasks, where low attention entropy is accompanied by high training instability, which can take the form of oscillating loss or divergence. We denote the pathologically low attention entropy, corresponding to highly concentrated attention scores, as $\textit{entropy collapse}$. As a remedy, we propose $σ$Reparam, a simple and efficient solution where we reparametrize all linear layers with spectral normalization and an additional learned scalar. We demonstrate that $σ$Reparam successfully prevents entropy collapse in the attention layers, promoting more stable training. Additionally, we prove a tight lower bound of the attention entropy, which decreases exponentially fast with the spectral norm of the attention logits, providing additional motivation for our approach. We conduct experiments with $σ$Reparam on image classification, image self-supervised learning, machine translation, speech recognition, and language modeling tasks. We show that $σ$Reparam provides stability and robustness with respect to the choice of hyperparameters, going so far as enabling training (a) a Vision Transformer {to competitive performance} without warmup, weight decay, layer normalization or adaptive optimizers; (b) deep architectures in machine translation and (c) speech recognition to competitive performance without warmup and adaptive optimizers. Code is available at \url{https://github.com/apple/ml-sigma-reparam}.

Martin Pelikan, Sheikh Shams Azam, Vitaly Feldman et al. · apple-ml

While federated learning (FL) and differential privacy (DP) have been extensively studied, their application to automatic speech recognition (ASR) remains largely unexplored due to the challenges in training large transformer models. Specifically, large models further exacerbate issues in FL as they are particularly susceptible to gradient heterogeneity across layers, unlike the relatively uniform gradient behavior observed in shallow models. As a result, prior works struggle to converge with standard optimization techniques, even in the absence of DP mechanisms. To the best of our knowledge, no existing work establishes a competitive, practical recipe for FL with DP in the context of ASR. To address this gap, we establish \textbf{the first benchmark for FL with DP in end-to-end ASR}. Our approach centers on per-layer clipping and layer-wise gradient normalization: theoretical analysis reveals that these techniques together mitigate clipping bias and gradient heterogeneity across layers in deeper models. Consistent with these theoretical insights, our empirical results show that FL with DP is viable under strong privacy guarantees, provided a population of at least several million users. Specifically, we achieve user-level (7.2, $10^{-9}$)-DP (resp. (4.5, $10^{-9}$)-DP) with only a 1.3% (resp. 4.6%) absolute drop in word error rate when extrapolating to high (resp. low) population scales for FL with DP in ASR. Although our experiments focus on ASR, the underlying principles we uncover - particularly those concerning gradient heterogeneity and layer-wise gradient normalization - offer broader guidance for designing scalable, privacy-preserving FL algorithms for large models across domains. Code of all experiments and benchmarks is available at https://github.com/apple/ml-pfl4asr.

Sarthak Garg, Mozhdeh Gheini, Clara Emmanuel et al. · apple-ml

Machine translation (MT) systems often translate terms with ambiguous gender (e.g., English term "the nurse") into the gendered form that is most prevalent in the systems' training data (e.g., "enfermera", the Spanish term for a female nurse). This often reflects and perpetuates harmful stereotypes present in society. With MT user interfaces in mind that allow for resolving gender ambiguity in a frictionless manner, we study the problem of generating all grammatically correct gendered translation alternatives. We open source train and test datasets for five language pairs and establish benchmarks for this task. Our key technical contribution is a novel semi-supervised solution for generating alternatives that integrates seamlessly with standard MT models and maintains high performance without requiring additional components or increasing inference overhead.

Jason Ramapuram, Federico Danieli, Eeshan Dhekane et al. · apple-ml, berkeley

Attention is a key part of the transformer architecture. It is a sequence-to-sequence mapping that transforms each sequence element into a weighted sum of values. The weights are typically obtained as the softmax of dot products between keys and queries. Recent work has explored alternatives to softmax attention in transformers, such as ReLU and sigmoid activations. In this work, we revisit sigmoid attention and conduct an in-depth theoretical and empirical analysis. Theoretically, we prove that transformers with sigmoid attention are universal function approximators and benefit from improved regularity compared to softmax attention. Through detailed empirical analysis, we identify stabilization of large initial attention norms during the early stages of training as a crucial factor for the successful training of models with sigmoid attention, outperforming prior attempts. We also introduce FLASHSIGMOID, a hardware-aware and memory-efficient implementation of sigmoid attention yielding a 17% inference kernel speed-up over FLASHATTENTION2 on H100 GPUs. Experiments across language, vision, and speech show that properly normalized sigmoid attention matches the strong performance of softmax attention on a wide range of domains and scales, which previous attempts at sigmoid attention were unable to fully achieve. Our work unifies prior art and establishes best practices for sigmoid attention as a drop-in softmax replacement in transformers.

Dan Busbridge, Jason Ramapuram, Pierre Ablin et al. · apple-ml, berkeley

Preserving training dynamics across batch sizes is an important tool for practical machine learning as it enables the trade-off between batch size and wall-clock time. This trade-off is typically enabled by a scaling rule, for example, in stochastic gradient descent, one should scale the learning rate linearly with the batch size. Another important machine learning tool is the model EMA, a functional copy of a target model, whose parameters move towards those of its target model according to an Exponential Moving Average (EMA) at a rate parameterized by a momentum hyperparameter. This model EMA can improve the robustness and generalization of supervised learning, stabilize pseudo-labeling, and provide a learning signal for Self-Supervised Learning (SSL). Prior works have not considered the optimization of the model EMA when performing scaling, leading to different training dynamics across batch sizes and lower model performance. In this work, we provide a scaling rule for optimization in the presence of a model EMA and demonstrate the rule's validity across a range of architectures, optimizers, and data modalities. We also show the rule's validity where the model EMA contributes to the optimization of the target model, enabling us to train EMA-based pseudo-labeling and SSL methods at small and large batch sizes. For SSL, we enable training of BYOL up to batch size 24,576 without sacrificing performance, a 6$\times$ wall-clock time reduction under idealized hardware settings.

Haoping Bai, Shancong Mou, Tatiana Likhomanenko et al. · apple-ml

Despite progress in vision-based inspection algorithms, real-world industrial challenges -- specifically in data availability, quality, and complex production requirements -- often remain under-addressed. We introduce the VISION Datasets, a diverse collection of 14 industrial inspection datasets, uniquely poised to meet these challenges. Unlike previous datasets, VISION brings versatility to defect detection, offering annotation masks across all splits and catering to various detection methodologies. Our datasets also feature instance-segmentation annotation, enabling precise defect identification. With a total of 18k images encompassing 44 defect types, VISION strives to mirror a wide range of real-world production scenarios. By supporting two ongoing challenge competitions on the VISION Datasets, we hope to foster further advancements in vision-based industrial inspection.

Shuangfei Zhai, Navdeep Jaitly, Jason Ramapuram et al. · apple-ml

Transformers have gained increasing popularity in a wide range of applications, including Natural Language Processing (NLP), Computer Vision and Speech Recognition, because of their powerful representational capacity. However, harnessing this representational capacity effectively requires a large amount of data, strong regularization, or both, to mitigate overfitting. Recently, the power of the Transformer has been unlocked by self-supervised pretraining strategies based on masked autoencoders which rely on reconstructing masked inputs, directly, or contrastively from unmasked content. This pretraining strategy which has been used in BERT models in NLP, Wav2Vec models in Speech and, recently, in MAE models in Vision, forces the model to learn about relationships between the content in different parts of the input using autoencoding related objectives. In this paper, we propose a novel, but surprisingly simple alternative to content reconstruction~-- that of predicting locations from content, without providing positional information for it. Doing so requires the Transformer to understand the positional relationships between different parts of the input, from their content alone. This amounts to an efficient implementation where the pretext task is a classification problem among all possible positions for each input token. We experiment on both Vision and Speech benchmarks, where our approach brings improvements over strong supervised training baselines and is comparable to modern unsupervised/self-supervised pretraining methods. Our method also enables Transformers trained without position embeddings to outperform ones trained with full position information.

Dan Berrebbi, Ronan Collobert, Samy Bengio et al. · apple-ml

Self-training (ST), or pseudo-labeling has sparked significant interest in the automatic speech recognition (ASR) community recently because of its success in harnessing unlabeled data. Unlike prior semi-supervised learning approaches that relied on iteratively regenerating pseudo-labels (PLs) from a trained model and using them to train a new model, recent state-of-the-art methods perform `continuous training' where PLs are generated using a very recent version of the model being trained. Nevertheless, these approaches still rely on bootstrapping the ST using an initial supervised learning phase where the model is trained on labeled data alone. We believe this has the potential for over-fitting to the labeled dataset in low resource settings and that ST from the start of training should reduce over-fitting. In this paper we show how we can do this by dynamically controlling the evolution of PLs during the training process in ASR. To the best of our knowledge, this is the first study that shows the feasibility of generating PLs from the very start of the training. We are able to achieve this using two techniques that avoid instabilities which lead to degenerate models that do not generalize. Firstly, we control the evolution of PLs through a curriculum that uses the online changes in PLs to control the membership of the cache of PLs and improve generalization. Secondly, we find that by sampling transcriptions from the predictive distribution, rather than only using the best transcription, we can stabilize training further. With these techniques, our ST models match prior works without an external language model.

Tatiana Likhomanenko, Ronan Collobert, Navdeep Jaitly et al. · apple-ml

Continuous pseudo-labeling (PL) algorithms such as slimIPL have recently emerged as a powerful strategy for semi-supervised learning in speech recognition. In contrast with earlier strategies that alternated between training a model and generating pseudo-labels (PLs) with it, here PLs are generated in end-to-end manner as training proceeds, improving training speed and the accuracy of the final model. PL shares a common theme with teacher-student models such as distillation in that a teacher model generates targets that need to be mimicked by the student model being trained. However, interestingly, PL strategies in general use hard-labels, whereas distillation uses the distribution over labels as the target to mimic. Inspired by distillation we expect that specifying the whole distribution (aka soft-labels) over sequences as the target for unlabeled data, instead of a single best pass pseudo-labeled transcript (hard-labels) should improve PL performance and convergence. Surprisingly and unexpectedly, we find that soft-labels targets can lead to training divergence, with the model collapsing to a degenerate token distribution per frame. We hypothesize that the reason this does not happen with hard-labels is that training loss on hard-labels imposes sequence-level consistency that keeps the model from collapsing to the degenerate solution. In this paper, we show several experiments that support this hypothesis, and experiment with several regularization approaches that can ameliorate the degenerate collapse when using soft-labels. These approaches can bring the accuracy of soft-labels closer to that of hard-labels, and while they are unable to outperform them yet, they serve as a useful framework for further improvements.

Sheikh Shams Azam, Tatiana Likhomanenko, Martin Pelikan et al. · apple-ml

In this paper, we start by training End-to-End Automatic Speech Recognition (ASR) models using Federated Learning (FL) and examining the fundamental considerations that can be pivotal in minimizing the performance gap in terms of word error rate between models trained using FL versus their centralized counterpart. Specifically, we study the effect of (i) adaptive optimizers, (ii) loss characteristics via altering Connectionist Temporal Classification (CTC) weight, (iii) model initialization through seed start, (iv) carrying over modeling setup from experiences in centralized training to FL, e.g., pre-layer or post-layer normalization, and (v) FL-specific hyperparameters, such as number of local epochs, client sampling size, and learning rate scheduler, specifically for ASR under heterogeneous data distribution. We shed light on how some optimizers work better than others via inducing smoothness. We also summarize the applicability of algorithms, trends, and propose best practices from prior works in FL (in general) toward End-to-End ASR models.

Richard He Bai, Tatiana Likhomanenko, Ruixiang Zhang et al. · apple-ml

Large language models have revolutionized natural language processing by leveraging self-supervised pretraining on vast textual data. Inspired by this success, researchers have investigated various compression-based speech tokenization methods to discretize continuous speech signals, enabling the application of language modeling techniques to discrete tokens. However, audio compressor introduces additional complexity and computational cost, and often fail on out-of-domain audio signals. In this work, we introduce a novel speech representation (dmel) that discretizes mel-filterbank channels into intensity bins, creating a simpler yet more effective representation compared to existing speech tokenization methods. Our approach demonstrates superior performance in preserving audio content, robustness to out-of-domain data, and offers a training-free, natural, and streamable representation. To address the high-dimensional nature of log-mel spectrograms, we propose an efficient parallel encoding and decoding method for high-dimensional tokens using an LM-style transformer architecture. This innovation enables us to develop RichTTS and RichASR, two models sharing the same architecture while achieving comparable or better results than specialized existing methods. Our results demonstrate the effectiveness of dmel in achieving high performance on both speech synthesis and recognition tasks within a unified framework, paving the way for efficient and effective joint modeling of speech and text.

Zijin Gu, Tatiana Likhomanenko, He Bai et al. · apple-ml

Language models play a central role in automatic speech recognition (ASR), yet most methods rely on text-only models unaware of ASR error patterns. Recently, large language models (LLMs) have been applied to ASR correction, but introduce latency and hallucination concerns. We revisit ASR error correction with compact seq2seq models, trained on ASR errors from real and synthetic audio. To scale training, we construct synthetic corpora via cascaded TTS and ASR, finding that matching the diversity of realistic error distributions is key. We propose correction-first decoding, where the correction model generates candidates rescored using ASR acoustic scores. With 15x fewer parameters than LLMs, our model achieves 1.5/3.3% WER on LibriSpeech test-clean/other, outperforms LLMs, generalizes across ASR architectures (CTC, Seq2seq, Transducer) and diverse domains, and provides precise corrections in the low-error regime where LLMs struggle.

Zijin Gu, Tatiana Likhomanenko, Vimal Thilak et al. · apple-ml

Sparse Mixture-of-Experts (MoE) architectures enable efficient scaling by activating only a subset of parameters for each input. However, conventional MoE routing selects each layer's experts independently, creating N^L possible expert paths -- for N experts across L layers. This far exceeds typical training set sizes, leading to statistical inefficiency as the model may not learn meaningful structure over such a vast path space. To constrain it, we propose \pathmoe, which shares router parameters across consecutive layers. Experiments on 0.9B and 16B parameter models demonstrate consistent improvements on perplexity and downstream tasks over independent routing, while eliminating the need for auxiliary load balancing losses. Analysis reveals that tokens following the same path naturally cluster by linguistic function, with \pathmoe{} producing more concentrated groups, better cross-layer consistency, and greater robustness to routing perturbations. These results offer a new perspective for understanding MoE architectures through the lens of expert paths.

Dan Berrebbi, Ronan Collobert, Navdeep Jaitly et al.

Self-training (ST) and self-supervised learning (SSL) methods have demonstrated strong improvements in automatic speech recognition (ASR). In spite of these advances, to the best of our knowledge, there is no analysis of how the composition of the labelled and unlabelled datasets used in these methods affects the results. In this work we aim to analyse the effect of number of speakers in the training data on a recent SSL algorithm (wav2vec 2.0), and a recent ST algorithm (slimIPL). We perform a systematic analysis on both labeled and unlabeled data by varying the number of speakers while keeping the number of hours fixed and vice versa. Our findings suggest that SSL requires a large amount of unlabeled data to produce high accuracy results, while ST requires a sufficient number of speakers in the labelled data, especially in the low-regime setting. In this manner these two approaches improve supervised learning in different regimes of data composition.

Andrew Rouditchenko, Ronan Collobert, Tatiana Likhomanenko

Audio-visual speech contains synchronized audio and visual information that provides cross-modal supervision to learn representations for both automatic speech recognition (ASR) and visual speech recognition (VSR). We introduce continuous pseudo-labeling for audio-visual speech recognition (AV-CPL), a semi-supervised method to train an audio-visual speech recognition (AVSR) model on a combination of labeled and unlabeled videos with continuously regenerated pseudo-labels. Our models are trained for speech recognition from audio-visual inputs and can perform speech recognition using both audio and visual modalities, or only one modality. Our method uses the same audio-visual model for both supervised training and pseudo-label generation, mitigating the need for external speech recognition models to generate pseudo-labels. AV-CPL obtains significant improvements in VSR performance on the LRS3 dataset while maintaining practical ASR and AVSR performance. Finally, using visual-only speech data, our method is able to leverage unlabeled visual speech to improve VSR.

Mozhdeh Gheini, Tatiana Likhomanenko, Matthias Sperber et al.

Self-training has been shown to be helpful in addressing data scarcity for many domains, including vision, speech, and language. Specifically, self-training, or pseudo-labeling, labels unsupervised data and adds that to the training pool. In this work, we investigate and use pseudo-labeling for a recently proposed novel setup: joint transcription and translation of speech, which suffers from an absence of sufficient data resources. We show that under such data-deficient circumstances, the unlabeled data can significantly vary in domain from the supervised data, which results in pseudo-label quality degradation. We investigate two categories of remedies that require no additional supervision and target the domain mismatch: pseudo-label filtering and data augmentation. We show that pseudo-label analysis and processing as such results in additional gains on top of the vanilla pseudo-labeling setup resulting in total improvements of up to 0.6% absolute WER and 2.2 BLEU points.

Zakaria Aldeneh, Skyler Seto, Maureen de Seyssel et al.

Modern ASR systems are typically trained on large-scale pseudo-labeled, in-the-wild data spanning multiple domains. While such heterogeneous data benefit generalist models designed for broad deployment, they pose challenges for specialist models targeting specific domains: specialist models lack the capacity to learn from all available data, and one must pay closer attention to addressing the mismatch between training and test conditions. In this work, we study targeted data selection as a strategy to address these challenges, selecting relevant subsets from 100k hours of in-the-wild training data to optimize performance on target domains. We represent speech samples using embeddings that capture complementary characteristic--speaker attributes, phonetic content, and semantic meaning--and analyze how relevance and diversity along these axes when performing data selection affect downstream ASR performance. Our experiments with CTC-based Conformer models show that training on a strategically selected 5% subset can exceed the performance of models trained on the full dataset by up to 36.8% relative WER reduction on target domains.

Jacob Kahn, Vineel Pratap, Tatiana Likhomanenko et al.

As the computational requirements for machine learning systems and the size and complexity of machine learning frameworks increases, essential framework innovation has become challenging. While computational needs have driven recent compiler, networking, and hardware advancements, utilization of those advancements by machine learning tools is occurring at a slower pace. This is in part due to the difficulties involved in prototyping new computational paradigms with existing frameworks. Large frameworks prioritize machine learning researchers and practitioners as end users and pay comparatively little attention to systems researchers who can push frameworks forward -- we argue that both are equally important stakeholders. We introduce Flashlight, an open-source library built to spur innovation in machine learning tools and systems by prioritizing open, modular, customizable internals and state-of-the-art, research-ready models and training setups across a variety of domains. Flashlight allows systems researchers to rapidly prototype and experiment with novel ideas in machine learning computation and has low overhead, competing with and often outperforming other popular machine learning frameworks. We see Flashlight as a tool enabling research that can benefit widely used libraries downstream and bring machine learning and systems researchers closer together. Flashlight is available at https://github.com/flashlight/flashlight .

Wei-Ning Hsu, Anuroop Sriram, Alexei Baevski et al.

Self-supervised learning of speech representations has been a very active research area but most work is focused on a single domain such as read audio books for which there exist large quantities of labeled and unlabeled data. In this paper, we explore more general setups where the domain of the unlabeled data for pre-training data differs from the domain of the labeled data for fine-tuning, which in turn may differ from the test data domain. Our experiments show that using target domain data during pre-training leads to large performance improvements across a variety of setups. On a large-scale competitive setup, we show that pre-training on unlabeled in-domain data reduces the gap between models trained on in-domain and out-of-domain labeled data by 66%-73%. This has obvious practical implications since it is much easier to obtain unlabeled target domain data than labeled data. Moreover, we find that pre-training on multiple domains improves generalization performance on domains not seen during training. Code and models will be made available at https://github.com/pytorch/fairseq.

Jacob Kahn, Morgane Rivière, Weiyi Zheng et al.

We introduce a new collection of spoken English audio suitable for training speech recognition systems under limited or no supervision. It is derived from open-source audio books from the LibriVox project. It contains over 60K hours of audio, which is, to our knowledge, the largest freely-available corpus of speech. The audio has been segmented using voice activity detection and is tagged with SNR, speaker ID and genre descriptions. Additionally, we provide baseline systems and evaluation metrics working under three settings: (1) the zero resource/unsupervised setting (ABX), (2) the semi-supervised setting (PER, CER) and (3) the distant supervision setting (WER). Settings (2) and (3) use limited textual resources (10 minutes to 10 hours) aligned with the speech. Setting (3) uses large amounts of unaligned text. They are evaluated on the standard LibriSpeech dev and test sets for comparison with the supervised state-of-the-art.

Jason Ramapuram, Eeshan Gunesh Dhekane, Amitis Shidani et al.

Speech-only spoken language models (SLMs) lag behind text and text-speech models in performance, with recent discrete autoregressive (AR) SLMs indicating significant computational and data demands to match text models. Since discretizing continuous speech for AR creates bottlenecks, we explore whether continuous diffusion (CD) SLM is more viable. To quantify the SLMs linguistic quality, we introduce the phoneme Jensen-Shannon divergence (pJSD) metric. Our analysis reveals CD SLMs, mirroring AR behavior, exhibit scaling laws for validation loss and pJSD, and show optimal token-to-parameter ratios decreasing as compute scales. However, for the latter, loss becomes insensitive to choice of data and model sizes, showing potential for fast inference. Scaling CD SLMs to 16B parameters with tens of millions of hours of conversational data enables generation of emotive, prosodic, multi-speaker, multilingual speech, though achieving long-form coherence remains a significant challenge.

Akshita Gupta, Tatiana Likhomanenko, Karren Dai Yang et al.

The rapid progress of foundation models and large language models (LLMs) has fueled significantly improvement in the capabilities of machine learning systems that benefit from mutlimodal input data. However, existing multimodal models are predominantly built on top of pre-trained LLMs, which can limit accurate modeling of temporal dependencies across other modalities and thus limit the model's ability to jointly process and leverage multimodal inputs. To specifically investigate the alignment of text, video, and speech modalities in LLM-style (decoder-only) models, we consider a simplified multimodal generation task, Video-Text to Speech (VTTS): speech generation conditioned on both its corresponding text and video of talking people. The ultimate goal is to generate speech that not only follows the text but also aligns temporally with the video and is consistent with the facial expressions. In this paper, we first introduce Visatronic, a unified multimodal decoder-only transformer model that adopts an LLM-style architecture to embed visual, textual, and speech inputs into a shared subspace, treating all modalities as temporally aligned token streams. Next, we carefully explore different token mixing strategies to understand the best way to propagate information from the steps where video and text conditioning is input to the steps where the audio is generated. We extensively evaluate Visatronic on the challenging VoxCeleb2 dataset and demonstrate zero-shot generalization to LRS3, where Visatronic, trained on VoxCeleb2, achieves a 4.5% WER, outperforming prior SOTA methods trained only on LRS3, which report a 21.4% WER. Additionally, we propose a new objective metric, TimeSync, specifically designed to measure phoneme-level temporal alignment between generated and reference speech, further ensuring synchronization quality. Demo: https://apple.github.io/visatronic-demo/

Richard He Bai, Zijin Gu, Tatiana Likhomanenko et al.

The latency bottleneck of traditional text-to-speech (TTS) systems fundamentally hinders the potential of streaming large language models (LLMs) in conversational AI. These TTS systems, typically trained and inferenced on complete utterances, introduce unacceptable delays, even with optimized inference speeds, when coupled with streaming LLM outputs. This is particularly problematic for creating responsive conversational agents where low first-token latency is critical. In this paper, we present SpeakStream, a streaming TTS system that generates audio incrementally from streaming text using a decoder-only architecture. SpeakStream is trained using a next-step prediction loss on interleaved text-speech data. During inference, it generates speech incrementally while absorbing streaming input text, making it particularly suitable for cascaded conversational AI agents where an LLM streams text to a TTS system. Our experiments demonstrate that SpeakStream achieves state-of-the-art latency results in terms of first-token latency while maintaining the quality of non-streaming TTS systems.

Tatiana Likhomanenko, Luke Carlson, Richard He Bai et al. · apple-ml

The emergence of large language models (LLMs) has transformed spoken dialog systems, yet the optimal architecture for real-time on-device voice agents remains an open question. While end-to-end approaches promise theoretical advantages, cascaded systems (CSs) continue to outperform them in language understanding tasks, despite being constrained by sequential processing latency. In this work, we introduce ChipChat, a novel low-latency CS that overcomes traditional bottlenecks through architectural innovations and streaming optimizations. Our system integrates streaming (a) conversational speech recognition with mixture-of-experts, (b) state-action augmented LLM, (c) text-to-speech synthesis, (d) neural vocoder, and (e) speaker modeling. Implemented using MLX, ChipChat achieves sub-second response latency on a Mac Studio without dedicated GPUs, while preserving user privacy through complete on-device processing. Our work shows that strategically redesigned CSs can overcome their historical latency limitations, offering a promising path forward for practical voice-based AI agents.

Zijin Gu, Tatiana Likhomanenko, Navdeep Jaitly

Mixture-of-experts (MoE) architectures have expanded from language modeling to automatic speech recognition (ASR). Traditional MoE methods, such as the Switch Transformer, route experts independently within each layer. Our analysis reveals that routers in most layers make expert choices that are not strongly correlated with the choices of the routers in other layers. To increase the cooperation between experts in different layers and encourage greater specialization, we use a shared router across different MoE layers. We call this model Omni-router Transformer. Extensive experiments on a large-scale pseudo-labeled dataset and evaluations across 10 diverse, out-of-domain ASR benchmarks demonstrate that the Omni-router Transformer is able to achieve lower training loss and consistently outperform dense and Switch Transformer models, reducing average word error rates by 11.2% and 8.2%, respectively, while providing structured expert usage and improved robustness to diverse data.

Santiago Cuervo, Skyler Seto, Maureen de Seyssel et al.

Large Language Models (LLMs) can be adapted to extend their text capabilities to speech inputs. However, these speech-adapted LLMs consistently underperform their text-based counterparts--and even cascaded pipelines--on language understanding tasks. We term this shortfall the text-speech understanding gap: the performance drop observed when a speech-adapted LLM processes spoken inputs relative to when the original text-based LLM processes the equivalent text. Recent approaches to narrowing this gap either rely on large-scale speech synthesis of text corpora, which is costly and heavily dependent on synthetic data, or on large-scale proprietary speech datasets, which are not reproducible. As a result, there remains a need for more data-efficient alternatives for closing the text-speech understanding gap. In this work, we analyze the gap as driven by two factors: (i) forgetting of text capabilities during adaptation, and (ii) cross-modal misalignment between speech and text. Based on this analysis, we introduce SALAD--Sample-efficient Alignment with Learning through Active selection and cross-modal Distillation--which combines cross-modal distillation with targeted synthetic data to improve alignment while mitigating forgetting. Applied to 3B and 7B LLMs, SALAD achieves competitive performance with a strong open-weight model across broad-domain benchmarks in knowledge, language understanding, and reasoning, while training on over an order of magnitude less speech data from public corpora.

Hyung Gun Chi, Zakaria Aldeneh, Tatiana Likhomanenko et al. · apple-ml

We introduce DiceHuBERT, a knowledge distillation framework for compressing HuBERT, a widely used self-supervised learning (SSL)-based speech foundation model. Unlike existing distillation methods that rely on layer-wise and feature-wise mapping between teacher and student models, DiceHuBERT leverages HuBERT's iterative self-distillation mechanism by directly replacing the original model with a student model. This replacement allows the student to be trained using the same SSL objective used when pre-training HuBERT, eliminating the need for additional modules or architectural constraints. Experimental results on SUPERB show that DiceHuBERT consistently outperforms existing distillation methods, improving phoneme recognition performance by over 21% and ASR performance by more than 14%. Furthermore, DiceHuBERT demonstrates competitive performance across multiple tasks, highlighting its clear advantage.

Tatiana Likhomanenko, Loren Lugosch, Ronan Collobert

Recent work has shown that it is possible to train an $\textit{unsupervised}$ automatic speech recognition (ASR) system using only unpaired audio and text. Existing unsupervised ASR methods assume that no labeled data can be used for training. We argue that even if one does not have any labeled audio for a given language, there is $\textit{always}$ labeled data available for other languages. We show that it is possible to use character-level acoustic models (AMs) from other languages to bootstrap an $\textit{unsupervised}$ AM in a new language. Here, "unsupervised" means no labeled audio is available for the $\textit{target}$ language. Our approach is based on two key ingredients: (i) generating pseudo-labels (PLs) of the $\textit{target}$ language using some $\textit{other}$ language AM and (ii) constraining these PLs with a $\textit{target language model}$. Our approach is effective on Common Voice: e.g. transfer of English AM to Swahili achieves 18% WER. It also outperforms character-based wav2vec-U 2.0 by 15% absolute WER on LJSpeech with 800h of labeled German data instead of 60k hours of unlabeled English data.

Loren Lugosch, Tatiana Likhomanenko, Gabriel Synnaeve et al.

Semi-supervised learning through pseudo-labeling has become a staple of state-of-the-art monolingual speech recognition systems. In this work, we extend pseudo-labeling to massively multilingual speech recognition with 60 languages. We propose a simple pseudo-labeling recipe that works well even with low-resource languages: train a supervised multilingual model, fine-tune it with semi-supervised learning on a target language, generate pseudo-labels for that language, and train a final model using pseudo-labels for all languages, either from scratch or by fine-tuning. Experiments on the labeled Common Voice and unlabeled VoxPopuli datasets show that our recipe can yield a model with better performance for many languages that also transfers well to LibriSpeech.

Vineel Pratap, Qiantong Xu, Tatiana Likhomanenko et al.

In this paper, we study training of automatic speech recognition system in a weakly supervised setting where the order of words in transcript labels of the audio training data is not known. We train a word-level acoustic model which aggregates the distribution of all output frames using LogSumExp operation and uses a cross-entropy loss to match with the ground-truth words distribution. Using the pseudo-labels generated from this model on the training set, we then train a letter-based acoustic model using Connectionist Temporal Classification loss. Our system achieves 2.3%/4.6% on test-clean/test-other subsets of LibriSpeech, which closely matches with the supervised baseline's performance.

Vimal Manohar, Tatiana Likhomanenko, Qiantong Xu et al.

In this paper, we introduce the Kaizen framework that uses a continuously improving teacher to generate pseudo-labels for semi-supervised speech recognition (ASR). The proposed approach uses a teacher model which is updated as the exponential moving average (EMA) of the student model parameters. We demonstrate that it is critical for EMA to be accumulated with full-precision floating point. The Kaizen framework can be seen as a continuous version of the iterative pseudo-labeling approach for semi-supervised training. It is applicable for different training criteria, and in this paper we demonstrate its effectiveness for frame-level hybrid hidden Markov model-deep neural network (HMM-DNN) systems as well as sequence-level Connectionist Temporal Classification (CTC) based models. For large scale real-world unsupervised public videos in UK English and Italian languages the proposed approach i) shows more than 10% relative word error rate (WER) reduction over standard teacher-student training; ii) using just 10 hours of supervised data and a large amount of unsupervised data closes the gap to the upper-bound supervised ASR system that uses 650h or 2700h respectively.

Tatiana Likhomanenko, Qiantong Xu, Gabriel Synnaeve et al.

Without positional information, attention-based Transformer neural networks are permutation-invariant. Absolute or relative positional embeddings are the most popular ways to feed Transformer models with positional information. Absolute positional embeddings are simple to implement, but suffer from generalization issues when evaluating on sequences longer than seen at training time. Relative positions are more robust to input length change, but are more complex to implement and yield inferior model throughput due to extra computational and memory costs. In this paper, we propose an augmentation-based approach (CAPE) for absolute positional embeddings, which keeps the advantages of both absolute (simplicity and speed) and relative positional embeddings (better generalization). In addition, our empirical evaluation on state-of-the-art models in machine translation, image and speech recognition demonstrates that CAPE leads to better generalization performance as well as increased stability with respect to training hyper-parameters.

Chaitanya Talnikar, Tatiana Likhomanenko, Ronan Collobert et al.

Self-supervised learning (SSL) has shown promise in learning representations of audio that are useful for automatic speech recognition (ASR). But, training SSL models like wav2vec~2.0 requires a two-stage pipeline. In this paper we demonstrate a single-stage training of ASR models that can utilize both unlabeled and labeled data. During training, we alternately minimize two losses: an unsupervised masked Contrastive Predictive Coding (CPC) loss and the supervised audio-to-text alignment loss Connectionist Temporal Classification (CTC). We show that this joint training method directly optimizes performance for the downstream ASR task using unsupervised data while achieving similar word error rates to wav2vec~2.0 on the Librispeech 100-hour dataset. Finally, we postulate that solving the contrastive task is a regularization for the supervised CTC loss.

Tatiana Likhomanenko, Qiantong Xu, Vineel Pratap et al.

Is pushing numbers on a single benchmark valuable in automatic speech recognition? Research results in acoustic modeling are typically evaluated based on performance on a single dataset. While the research community has coalesced around various benchmarks, we set out to understand generalization performance in acoustic modeling across datasets - in particular, if models trained on a single dataset transfer to other (possibly out-of-domain) datasets. We show that, in general, reverberative and additive noise augmentation improves generalization performance across domains. Further, we demonstrate that when a large enough set of benchmarks is used, average word error rate (WER) performance over them provides a good proxy for performance on real-world noisy data. Finally, we show that training a single acoustic model on the most widely-used datasets - combined - reaches competitive performance on both research and real-world benchmarks.

Tatiana Likhomanenko, Qiantong Xu, Jacob Kahn et al.

Recent results in end-to-end automatic speech recognition have demonstrated the efficacy of pseudo-labeling for semi-supervised models trained both with Connectionist Temporal Classification (CTC) and Sequence-to-Sequence (seq2seq) losses. Iterative Pseudo-Labeling (IPL), which continuously trains a single model using pseudo-labels iteratively re-generated as the model learns, has been shown to further improve performance in ASR. We improve upon the IPL algorithm: as the model learns, we propose to iteratively re-generate transcriptions with hard labels (the most probable tokens), that is, without a language model. We call this approach Language-Model-Free IPL (slimIPL) and give a resultant training setup for low-resource settings with CTC-based models. slimIPL features a dynamic cache for pseudo-labels which reduces sensitivity to changes in relabeling hyperparameters and results in improves training stability. slimIPL is also highly-efficient and requires 3.5-4x fewer computational resources to converge than other state-of-the-art semi/self-supervised approaches. With only 10 hours of labeled audio, slimIPL is competitive with self-supervised approaches, and is state-of-the-art with 100 hours of labeled audio without the use of a language model both at test time and during pseudo-label generation.

Qiantong Xu, Alexei Baevski, Tatiana Likhomanenko et al.

Self-training and unsupervised pre-training have emerged as effective approaches to improve speech recognition systems using unlabeled data. However, it is not clear whether they learn similar patterns or if they can be effectively combined. In this paper, we show that pseudo-labeling and pre-training with wav2vec 2.0 are complementary in a variety of labeled data setups. Using just 10 minutes of labeled data from Libri-light as well as 53k hours of unlabeled data from LibriVox achieves WERs of 3.0%/5.2% on the clean and other test sets of Librispeech - rivaling the best published systems trained on 960 hours of labeled data only a year ago. Training on all labeled data of Librispeech achieves WERs of 1.5%/3.1%.

Qiantong Xu, Tatiana Likhomanenko, Jacob Kahn et al.

Pseudo-labeling has recently shown promise in end-to-end automatic speech recognition (ASR). We study Iterative Pseudo-Labeling (IPL), a semi-supervised algorithm which efficiently performs multiple iterations of pseudo-labeling on unlabeled data as the acoustic model evolves. In particular, IPL fine-tunes an existing model at each iteration using both labeled data and a subset of unlabeled data. We study the main components of IPL: decoding with a language model and data augmentation. We then demonstrate the effectiveness of IPL by achieving state-of-the-art word-error rate on the Librispeech test sets in both standard and low-resource setting. We also study the effect of language models trained on different corpora to show IPL can effectively utilize additional text. Finally, we release a new large in-domain text corpus which does not overlap with the Librispeech training transcriptions to foster research in low-resource, semi-supervised ASR

Vineel Pratap, Qiantong Xu, Jacob Kahn et al.

We design an online end-to-end speech recognition system based on Time-Depth Separable (TDS) convolutions and Connectionist Temporal Classification (CTC). We improve the core TDS architecture in order to limit the future context and hence reduce latency while maintaining accuracy. The system has almost three times the throughput of a well tuned hybrid ASR baseline while also having lower latency and a better word error rate. Also important to the efficiency of the recognizer is our highly optimized beam search decoder. To show the impact of our design choices, we analyze throughput, latency, accuracy, and discuss how these metrics can be tuned based on the user requirements.

Gabriel Synnaeve, Qiantong Xu, Jacob Kahn et al.

We study pseudo-labeling for the semi-supervised training of ResNet, Time-Depth Separable ConvNets, and Transformers for speech recognition, with either CTC or Seq2Seq loss functions. We perform experiments on the standard LibriSpeech dataset, and leverage additional unlabeled data from LibriVox through pseudo-labeling. We show that while Transformer-based acoustic models have superior performance with the supervised dataset alone, semi-supervision improves all models across architectures and loss functions and bridges much of the performance gaps between them. In doing so, we reach a new state-of-the-art for end-to-end acoustic models decoded with an external language model in the standard supervised learning setting, and a new absolute state-of-the-art with semi-supervised training. Finally, we study the effect of leveraging different amounts of unlabeled audio, propose several ways of evaluating the characteristics of unlabeled audio which improve acoustic modeling, and show that acoustic models trained with more audio rely less on external language models.

Tatiana Likhomanenko, Gabriel Synnaeve, Ronan Collobert

Lexicon-free speech recognition naturally deals with the problem of out-of-vocabulary (OOV) words. In this paper, we show that character-based language models (LM) can perform as well as word-based LMs for speech recognition, in word error rates (WER), even without restricting the decoding to a lexicon. We study character-based LMs and show that convolutional LMs can effectively leverage large (character) contexts, which is key for good speech recognition performance downstream. We specifically show that the lexicon-free decoding performance (WER) on utterances with OOV words using character-based LMs is better than lexicon-based decoding, both with character or word-based LMs.

Alex Rogozhnikov, Tatiana Likhomanenko

In machine learning ensemble methods have demonstrated high accuracy for the variety of problems in different areas. Two notable ensemble methods widely used in practice are gradient boosting and random forests. In this paper we present InfiniteBoost - a novel algorithm, which combines important properties of these two approaches. The algorithm constructs the ensemble of trees for which two properties hold: trees of the ensemble incorporate the mistakes done by others; at the same time the ensemble could contain the infinite number of trees without the over-fitting effect. The proposed algorithm is evaluated on the regression, classification, and ranking tasks using large scale, publicly available datasets.

Tatiana Likhomanenko, Denis Derkach, Alex Rogozhnikov

Identifying the flavour of neutral $B$ mesons production is one of the most important components needed in the study of time-dependent $CP$ violation. The harsh environment of the Large Hadron Collider makes it particularly hard to succeed in this task. We present an inclusive flavour-tagging algorithm as an upgrade of the algorithms currently used by the LHCb experiment. Specifically, a probabilistic model which efficiently combines information from reconstructed vertices and tracks using machine learning is proposed. The algorithm does not use information about underlying physics process. It reduces the dependence on the performance of lower level identification capacities and thus increases the overall performance. The proposed inclusive flavour-tagging algorithm is applicable to tag the flavour of $B$ mesons in any proton-proton experiment.