Jon Barker

Aaron Blakeman, Aarti Basant, Abhinav Khattar et al. · nvidia

As inference-time scaling becomes critical for enhanced reasoning capabilities, it is increasingly becoming important to build models that are efficient to infer. We introduce Nemotron-H, a family of 8B and 56B/47B hybrid Mamba-Transformer models designed to reduce inference cost for a given accuracy level. To achieve this goal, we replace the majority of self-attention layers in the common Transformer model architecture with Mamba layers that perform constant computation and require constant memory per generated token. We show that Nemotron-H models offer either better or on-par accuracy compared to other similarly-sized state-of-the-art open-sourced Transformer models (e.g., Qwen-2.5-7B/72B and Llama-3.1-8B/70B), while being up to 3$\times$ faster at inference. To further increase inference speed and reduce the memory required at inference time, we created Nemotron-H-47B-Base from the 56B model using a new compression via pruning and distillation technique called MiniPuzzle. Nemotron-H-47B-Base achieves similar accuracy to the 56B model, but is 20% faster to infer. In addition, we introduce an FP8-based training recipe and show that it can achieve on par results with BF16-based training. This recipe is used to train the 56B model. We are releasing Nemotron-H base model checkpoints with support in Hugging Face and NeMo.

Wenliang Dai, Nayeon Lee, Boxin Wang et al. · nvidia

We introduce NVLM 1.0, a family of frontier-class multimodal large language models (LLMs) that achieve state-of-the-art results on vision-language tasks, rivaling the leading proprietary models (e.g., GPT-4o) and open-access models (e.g., Llama 3-V 405B and InternVL 2). Remarkably, NVLM 1.0 shows improved text-only performance over its LLM backbone after multimodal training. In terms of model design, we perform a comprehensive comparison between decoder-only multimodal LLMs (e.g., LLaVA) and cross-attention-based models (e.g., Flamingo). Based on the strengths and weaknesses of both approaches, we propose a novel architecture that enhances both training efficiency and multimodal reasoning capabilities. Furthermore, we introduce a 1-D tile-tagging design for tile-based dynamic high-resolution images, which significantly boosts performance on multimodal reasoning and OCR-related tasks. Regarding training data, we meticulously curate and provide detailed information on our multimodal pretraining and supervised fine-tuning datasets. Our findings indicate that dataset quality and task diversity are more important than scale, even during the pretraining phase, across all architectures. Notably, we develop production-grade multimodality for the NVLM-1.0 models, enabling them to excel in vision-language tasks while maintaining and even improving text-only performance compared to their LLM backbones. To achieve this, we craft and integrate a high-quality text-only dataset into multimodal training, alongside a substantial amount of multimodal math and reasoning data, leading to enhanced math and coding capabilities across modalities. To advance research in the field, we release the model weights at https://huggingface.co/nvidia/NVLM-D-72B and will open-source the training code for the community soon.

Gerardo Roa Dabike, Michael A. Akeroyd, Scott Bannister et al.

This paper reports on the design and results of the 2024 ICASSP SP Cadenza Challenge: Music Demixing/Remixing for Hearing Aids. The Cadenza project is working to enhance the audio quality of music for those with a hearing loss. The scenario for the challenge was listening to stereo reproduction over loudspeakers via hearing aids. The task was to: decompose pop/rock music into vocal, drums, bass and other (VDBO); rebalance the different tracks with specified gains and then remixing back to stereo. End-to-end approaches were also accepted. 17 systems were submitted by 11 teams. Causal systems performed poorer than non-causal approaches. 9 systems beat the baseline. A common approach was to fine-tuning pretrained demixing models. The best approach used an ensemble of models.

Gerardo Roa Dabike, Scott Bannister, Jennifer Firth et al.

The Cadenza project aims to improve the audio quality of music for those who have a hearing loss. This is being done through a series of signal processing challenges, to foster better and more inclusive technologies. In the first round, two common listening scenarios are considered: listening to music over headphones, and with a hearing aid in a car. The first scenario is cast as a demixing-remixing problem, where the music is decomposed into vocals, bass, drums and other components. These can then be intelligently remixed in a personalized way, to increase the audio quality for a person who has a hearing loss. In the second scenario, music is coming from car loudspeakers, and the music has to be enhanced to overcome the masking effect of the car noise. This is done by taking into account the music, the hearing ability of the listener, the hearing aid and the speed of the car. The audio quality of the submissions will be evaluated using the Hearing Aid Audio Quality Index (HAAQI) for objective assessment and by a panel of people with hearing loss for subjective evaluation.

Jisi Zhang, Catalin Zorila, Rama Doddipatla et al.

In this paper, we explore an improved framework to train a monoaural neural enhancement model for robust speech recognition. The designed training framework extends the existing mixture invariant training criterion to exploit both unpaired clean speech and real noisy data. It is found that the unpaired clean speech is crucial to improve quality of separated speech from real noisy speech. The proposed method also performs remixing of processed and unprocessed signals to alleviate the processing artifacts. Experiments on the single-channel CHiME-3 real test sets show that the proposed method improves significantly in terms of speech recognition performance over the enhancement system trained either on the mismatched simulated data in a supervised fashion or on the matched real data in an unsupervised fashion. Between 16% and 39% relative WER reduction has been achieved by the proposed system compared to the unprocessed signal using end-to-end and hybrid acoustic models without retraining on distorted data.

Shinji Watanabe, Michael Mandel, Jon Barker et al.

Following the success of the 1st, 2nd, 3rd, 4th and 5th CHiME challenges we organize the 6th CHiME Speech Separation and Recognition Challenge (CHiME-6). The new challenge revisits the previous CHiME-5 challenge and further considers the problem of distant multi-microphone conversational speech diarization and recognition in everyday home environments. Speech material is the same as the previous CHiME-5 recordings except for accurate array synchronization. The material was elicited using a dinner party scenario with efforts taken to capture data that is representative of natural conversational speech. This paper provides a baseline description of the CHiME-6 challenge for both segmented multispeaker speech recognition (Track 1) and unsegmented multispeaker speech recognition (Track 2). Of note, Track 2 is the first challenge activity in the community to tackle an unsegmented multispeaker speech recognition scenario with a complete set of reproducible open source baselines providing speech enhancement, speaker diarization, and speech recognition modules.

Rhiannon Mogridge, George Close, Robert Sutherland et al.

Neural networks have been successfully used for non-intrusive speech intelligibility prediction. Recently, the use of feature representations sourced from intermediate layers of pre-trained self-supervised and weakly-supervised models has been found to be particularly useful for this task. This work combines the use of Whisper ASR decoder layer representations as neural network input features with an exemplar-based, psychologically motivated model of human memory to predict human intelligibility ratings for hearing-aid users. Substantial performance improvement over an established intrusive HASPI baseline system is found, including on enhancement systems and listeners unseen in the training data, with a root mean squared error of 25.3 compared with the baseline of 28.7.

Max Ehrlich, Jon Barker, Namitha Padmanabhan et al.

Video compression is a central feature of the modern internet powering technologies from social media to video conferencing. While video compression continues to mature, for many compression settings, quality loss is still noticeable. These settings nevertheless have important applications to the efficient transmission of videos over bandwidth constrained or otherwise unstable connections. In this work, we develop a deep learning architecture capable of restoring detail to compressed videos which leverages the underlying structure and motion information embedded in the video bitstream. We show that this improves restoration accuracy compared to prior compression correction methods and is competitive when compared with recent deep-learning-based video compression methods on rate-distortion while achieving higher throughput. Furthermore, we condition our model on quantization data which is readily available in the bitstream. This allows our single model to handle a variety of different compression quality settings which required an ensemble of models in prior work.

Jisi Zhang, Catalin Zorila, Rama Doddipatla et al.

In this paper, we introduce a novel semi-supervised learning framework for end-to-end speech separation. The proposed method first uses mixtures of unseparated sources and the mixture invariant training (MixIT) criterion to train a teacher model. The teacher model then estimates separated sources that are used to train a student model with standard permutation invariant training (PIT). The student model can be fine-tuned with supervised data, i.e., paired artificial mixtures and clean speech sources, and further improved via model distillation. Experiments with single and multi channel mixtures show that the teacher-student training resolves the over-separation problem observed in the original MixIT method. Further, the semisupervised performance is comparable to a fully-supervised separation system trained using ten times the amount of supervised data.

Zehai Tu, Ning Ma, Jon Barker

Current hearing aids normally provide amplification based on a general prescriptive fitting, and the benefits provided by the hearing aids vary among different listening environments despite the inclusion of noise suppression feature. Motivated by this fact, this paper proposes a data-driven machine learning technique to develop hearing aid fittings that are customised to speech in different noisy environments. A differentiable hearing loss model is proposed and used to optimise fittings with back-propagation. The customisation is reflected on the data of speech in different noise with also the consideration of noise suppression. The objective evaluation shows the advantages of optimised custom fittings over general prescriptive fittings.

Zehai Tu, Ning Ma, Jon Barker

Hearing aids are expected to improve speech intelligibility for listeners with hearing impairment. An appropriate amplification fitting tuned for the listener's hearing disability is critical for good performance. The developments of most prescriptive fittings are based on data collected in subjective listening experiments, which are usually expensive and time-consuming. In this paper, we explore an alternative approach to finding the optimal fitting by introducing a hearing aid speech processing framework, in which the fitting is optimised in an automated way using an intelligibility objective function based on the HASPI physiological auditory model. The framework is fully differentiable, thus can employ the back-propagation algorithm for efficient, data-driven optimisation. Our initial objective experiments show promising results for noise-free speech amplification, where the automatically optimised processors outperform one of the well recognised hearing aid prescriptions.

Gerardo Roa Dabike, Jon Barker

In this paper, we ask whether vocal source features (pitch, shimmer, jitter, etc) can improve the performance of automatic sung speech recognition, arguing that conclusions previously drawn from spoken speech studies may not be valid in the sung speech domain. We first use a parallel singing/speaking corpus (NUS-48E) to illustrate differences in sung vs spoken voicing characteristics including pitch range, syllables duration, vibrato, jitter and shimmer. We then use this analysis to inform speech recognition experiments on the sung speech DSing corpus, using a state of the art acoustic model and augmenting conventional features with various voice source parameters. Experiments are run with three standard (increasingly large) training sets, DSing1 (15.1 hours), DSing3 (44.7 hours) and DSing30 (149.1 hours). Pitch combined with degree of voicing produces a significant decrease in WER from 38.1% to 36.7% when training with DSing1 however smaller decreases in WER observed when training with the larger more varied DSing3 and DSing30 sets were not seen to be statistically significant. Voicing quality characteristics did not improve recognition performance although analysis suggests that they do contribute to an improved discrimination between voiced/unvoiced phoneme pairs.

Jisi Zhang, Catalin Zorila, Rama Doddipatla et al.

In this paper, we present a novel multi-channel speech extraction system to simultaneously extract multiple clean individual sources from a mixture in noisy and reverberant environments. The proposed method is built on an improved multi-channel time-domain speech separation network which employs speaker embeddings to identify and extract multiple targets without label permutation ambiguity. To efficiently inform the speaker information to the extraction model, we propose a new speaker conditioning mechanism by designing an additional speaker branch for receiving external speaker embeddings. Experiments on 2-channel WHAMR! data show that the proposed system improves by 9% relative the source separation performance over a strong multi-channel baseline, and it increases the speech recognition accuracy by more than 16% relative over the same baseline.

Jisi Zhang, Catalin Zorila, Rama Doddipatla et al.

This paper introduces a new method for multi-channel time domain speech separation in reverberant environments. A fully-convolutional neural network structure has been used to directly separate speech from multiple microphone recordings, with no need of conventional spatial feature extraction. To reduce the influence of reverberation on spatial feature extraction, a dereverberation pre-processing method has been applied to further improve the separation performance. A spatialized version of wsj0-2mix dataset has been simulated to evaluate the proposed system. Both source separation and speech recognition performance of the separated signals have been evaluated objectively. Experiments show that the proposed fully-convolutional network improves the source separation metric and the word error rate (WER) by more than 13% and 50% relative, respectively, over a reference system with conventional features. Applying dereverberation as pre-processing to the proposed system can further reduce the WER by 29% relative using an acoustic model trained on clean and reverberated data.

Fitsum A. Reda, Guilin Liu, Kevin J. Shih et al.

We present an approach for high-resolution video frame prediction by conditioning on both past frames and past optical flows. Previous approaches rely on resampling past frames, guided by a learned future optical flow, or on direct generation of pixels. Resampling based on flow is insufficient because it cannot deal with disocclusions. Generative models currently lead to blurry results. Recent approaches synthesis a pixel by convolving input patches with a predicted kernel. However, their memory requirement increases with kernel size. Here, we spatially-displaced convolution (SDC) module for video frame prediction. We learn a motion vector and a kernel for each pixel and synthesize a pixel by applying the kernel at a displaced location in the source image, defined by the predicted motion vector. Our approach inherits the merits of both vector-based and kernel-based approaches, while ameliorating their respective disadvantages. We train our model on 428K unlabelled 1080p video game frames. Our approach produces state-of-the-art results, achieving an SSIM score of 0.904 on high-definition YouTube-8M videos, 0.918 on Caltech Pedestrian videos. Our model handles large motion effectively and synthesizes crisp frames with consistent motion.

Mandar Gogate, Ahsan Adeel, Ricard Marxer et al.

Human auditory cortex excels at selectively suppressing background noise to focus on a target speaker. The process of selective attention in the brain is known to contextually exploit the available audio and visual cues to better focus on target speaker while filtering out other noises. In this study, we propose a novel deep neural network (DNN) based audiovisual (AV) mask estimation model. The proposed AV mask estimation model contextually integrates the temporal dynamics of both audio and noise-immune visual features for improved mask estimation and speech separation. For optimal AV features extraction and ideal binary mask (IBM) estimation, a hybrid DNN architecture is exploited to leverages the complementary strengths of a stacked long short term memory (LSTM) and convolution LSTM network. The comparative simulation results in terms of speech quality and intelligibility demonstrate significant performance improvement of our proposed AV mask estimation model as compared to audio-only and visual-only mask estimation approaches for both speaker dependent and independent scenarios.

Jon Barker, Shinji Watanabe, Emmanuel Vincent et al.

The CHiME challenge series aims to advance robust automatic speech recognition (ASR) technology by promoting research at the interface of speech and language processing, signal processing , and machine learning. This paper introduces the 5th CHiME Challenge, which considers the task of distant multi-microphone conversational ASR in real home environments. Speech material was elicited using a dinner party scenario with efforts taken to capture data that is representative of natural conversational speech and recorded by 6 Kinect microphone arrays and 4 binaural microphone pairs. The challenge features a single-array track and a multiple-array track and, for each track, distinct rankings will be produced for systems focusing on robustness with respect to distant-microphone capture vs. systems attempting to address all aspects of the task including conversational language modeling. We discuss the rationale for the challenge and provide a detailed description of the data collection procedure, the task, and the baseline systems for array synchronization, speech enhancement, and conventional and end-to-end ASR.

Edward Raff, Jon Barker, Jared Sylvester et al.

In this work we introduce malware detection from raw byte sequences as a fruitful research area to the larger machine learning community. Building a neural network for such a problem presents a number of interesting challenges that have not occurred in tasks such as image processing or NLP. In particular, we note that detection from raw bytes presents a sequence problem with over two million time steps and a problem where batch normalization appear to hinder the learning process. We present our initial work in building a solution to tackle this problem, which has linear complexity dependence on the sequence length, and allows for interpretable sub-regions of the binary to be identified. In doing so we will discuss the many challenges in building a neural network to process data at this scale, and the methods we used to work around them.