Author name cluster

Nima Mesgarani

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

10 papers

2 author rows

AAAI Conference 2026 Conference Paper

DMOSpeech 2: Reinforcement Learning for Duration Prediction in Metric-Optimized Speech Synthesis

Yinghao Aaron Li
Xilin Jiang
Fei Tao
Cheng Niu
Kaifeng Xu
Juntong Song
Nima Mesgarani

Diffusion-based text-to-speech (TTS) systems have made remarkable progress in zero-shot speech synthesis, yet optimizing all components for perceptual metrics remains challenging. Prior work with DMOSpeech demonstrated direct metric optimization for speech generation components, but duration prediction remained unoptimized. This paper presents DMOSpeech 2, which extends metric optimization to the duration predictor through a reinforcement learning approach. The proposed system implements a novel duration policy framework using group relative preference optimization (GRPO) with speaker similarity and word error rate as reward signals. By optimizing this previously unoptimized component, DMOSpeech 2 creates a more complete metric-optimized synthesis pipeline. Additionally, this paper introduces teacher-guided sampling, a hybrid approach leveraging a teacher model for initial denoising steps before transitioning to the student model, significantly improving output diversity while maintaining efficiency. Comprehensive evaluations demonstrate superior performance across all metrics compared to previous systems, while reducing sampling steps by half without quality degradation. These advances represent a significant step toward speech synthesis systems with metric optimization across multiple components.

PDF Details DOI

NeurIPS Conference 2025 Conference Paper

Far from the Shallow: Brain-Predictive Reasoning Embedding through Residual Disentanglement

Linyang He
Tianjun Zhong
Richard Antonello
Gavin Mischler
Micah Goldblum
Nima Mesgarani

Understanding how the human brain progresses from processing simple linguistic inputs to performing high-level reasoning is a fundamental challenge in neuroscience. While modern large language models (LLMs) are increasingly used to model neural responses to language, their internal representations are highly "entangled, " mixing information about lexicon, syntax, meaning, and reasoning. This entanglement biases conventional brain encoding analyses toward linguistically shallow features (e. g. , lexicon and syntax), making it difficult to isolate the neural substrates of cognitively deeper processes. Here, we introduce a residual disentanglement method that computationally isolates these components. By first probing an LM to identify feature-specific layers, our method iteratively regresses out lower-level representations to produce four nearly orthogonal embeddings for lexicon, syntax, meaning, and, critically, reasoning. We used these disentangled embeddings to model intracranial (ECoG) brain recordings from neurosurgical patients listening to natural speech. We show that: 1) This isolated reasoning embedding exhibits unique predictive power, accounting for variance in neural activity not explained by other linguistic features and even extending to the recruitment of visual regions beyond classical language areas. 2) The neural signature for reasoning is temporally distinct, peaking later (~350-400ms) than signals related to lexicon, syntax, and meaning, consistent with its position atop a processing hierarchy. 3) Standard, non-disentangled LLM embeddings can be misleading, as their predictive success is primarily attributable to linguistically shallow features, masking the more subtle contributions of deeper cognitive processing. Our work provides compelling neural evidence for an abstract reasoning computation during language comprehension and offers a robust framework for mapping distinct cognitive functions from artificial models to the human brain.

PDF Details

NeurIPS Conference 2025 Conference Paper

ZeroSep: Separate Anything in Audio with Zero Training

Chao Huang
Yuesheng Ma
Junxuan Huang
Susan Liang
Yunlong Tang
Jing Bi
Wenqiang Liu
Nima Mesgarani

Audio source separation is fundamental for machines to understand complex acoustic environments and underpins numerous audio applications. Current supervised deep learning approaches, while powerful, are limited by the need for extensive, task-specific labeled data and struggle to generalize to the immense variability and open-set nature of real-world acoustic scenes. Inspired by the success of generative foundation models, we investigate whether pre-trained text-guided audio diffusion models can overcome these limitations. We make a surprising discovery: zero-shot source separation can be achieved purely through a pre-trained text-guided audio diffusion model under the right configuration. Our method, named ZeroSep, works by inverting the mixed audio into the diffusion model's latent space and then using text conditioning to guide the denoising process to recover individual sources. Without any task-specific training or fine-tuning, ZeroSep repurposes the generative diffusion model for a discriminative separation task and inherently supports open-set scenarios through its rich textual priors. ZeroSep is compatible with a variety of pre-trained text-guided audio diffusion backbones and delivers strong separation performance on multiple separation benchmarks, surpassing even supervised methods.

PDF Details

YNIMG Journal 2023 Journal Article

Deep neural networks effectively model neural adaptation to changing background noise and suggest nonlinear noise filtering methods in auditory cortex

Gavin Mischler
Menoua Keshishian
Stephan Bickel
Ashesh D. Mehta
Nima Mesgarani

The human auditory system displays a robust capacity to adapt to sudden changes in background noise, allowing for continuous speech comprehension despite changes in background environments. However, despite comprehensive studies characterizing this ability, the computations that underly this process are not well understood. The first step towards understanding a complex system is to propose a suitable model, but the classical and easily interpreted model for the auditory system, the spectro-temporal receptive field (STRF), cannot match the nonlinear neural dynamics involved in noise adaptation. Here, we utilize a deep neural network (DNN) to model neural adaptation to noise, illustrating its effectiveness at reproducing the complex dynamics at the levels of both individual electrodes and the cortical population. By closely inspecting the model's STRF-like computations over time, we find that the model alters both the gain and shape of its receptive field when adapting to a sudden noise change. We show that the DNN model's gain changes allow it to perform adaptive gain control, while the spectro-temporal change creates noise filtering by altering the inhibitory region of the model's receptive field. Further, we find that models of electrodes in nonprimary auditory cortex also exhibit noise filtering changes in their excitatory regions, suggesting differences in noise filtering mechanisms along the cortical hierarchy. These findings demonstrate the capability of deep neural networks to model complex neural adaptation and offer new hypotheses about the computations the auditory cortex performs to enable noise-robust speech perception in real-world, dynamic environments.

Details DOI

NeurIPS Conference 2023 Conference Paper

StyleTTS 2: Towards Human-Level Text-to-Speech through Style Diffusion and Adversarial Training with Large Speech Language Models

Yinghao Aaron Li
Cong Han
Vinay Raghavan
Gavin Mischler
Nima Mesgarani

In this paper, we present StyleTTS 2, a text-to-speech (TTS) model that leverages style diffusion and adversarial training with large speech language models (SLMs) to achieve human-level TTS synthesis. StyleTTS 2 differs from its predecessor by modeling styles as a latent random variable through diffusion models to generate the most suitable style for the text without requiring reference speech, achieving efficient latent diffusion while benefiting from the diverse speech synthesis offered by diffusion models. Furthermore, we employ large pre-trained SLMs, such as WavLM, as discriminators with our novel differentiable duration modeling for end-to-end training, resulting in improved speech naturalness. StyleTTS 2 surpasses human recordings on the single-speaker LJSpeech dataset and matches it on the multispeaker VCTK dataset as judged by native English speakers. Moreover, when trained on the LibriTTS dataset, our model outperforms previous publicly available models for zero-shot speaker adaptation. This work achieves the first human-level TTS on both single and multispeaker datasets, showcasing the potential of style diffusion and adversarial training with large SLMs. The audio demos and source code are available at https: //styletts2. github. io/.

PDF Details

YNIMG Journal 2021 Journal Article

Functional characterization of human Heschl's gyrus in response to natural speech

Bahar Khalighinejad
Prachi Patel
Jose L. Herrero
Stephan Bickel
Ashesh D. Mehta
Nima Mesgarani

Heschl's gyrus (HG) is a brain area that includes the primary auditory cortex in humans. Due to the limitations in obtaining direct neural measurements from this region during naturalistic speech listening, the functional organization and the role of HG in speech perception remain uncertain. Here, we used intracranial EEG to directly record neural activity in HG in eight neurosurgical patients as they listened to continuous speech stories. We studied the spatial distribution of acoustic tuning and the organization of linguistic feature encoding. We found a main gradient of change from posteromedial to anterolateral parts of HG. We also observed a decrease in frequency and temporal modulation tuning and an increase in phonemic representation, speaker normalization, speech sensitivity, and response latency. We did not observe a difference between the two brain hemispheres. These findings reveal a functional role for HG in processing and transforming simple to complex acoustic features and inform neurophysiological models of speech processing in the human auditory cortex.

Details DOI

YNIMG Journal 2021 Journal Article

Neural representation of linguistic feature hierarchy reflects second-language proficiency

Giovanni M. Di Liberto
Jingping Nie
Jeremy Yeaton
Bahar Khalighinejad
Shihab A. Shamma
Nima Mesgarani

Acquiring a new language requires individuals to simultaneously and gradually learn linguistic attributes on multiple levels. Here, we investigated how this learning process changes the neural encoding of natural speech by assessing the encoding of the linguistic feature hierarchy in second-language listeners. Electroencephalography (EEG) signals were recorded from native Mandarin speakers with varied English proficiency and from native English speakers while they listened to audio-stories in English. We measured the temporal response functions (TRFs) for acoustic, phonemic, phonotactic, and semantic features in individual participants and found a main effect of proficiency on linguistic encoding. This effect of second-language proficiency was particularly prominent on the neural encoding of phonemes, showing stronger encoding of "new" phonemic contrasts (i.e., English contrasts that do not exist in Mandarin) with increasing proficiency. Overall, we found that the nonnative listeners with higher proficiency levels had a linguistic feature representation more similar to that of native listeners, which enabled the accurate decoding of language proficiency. This result advances our understanding of the cortical processing of linguistic information in second-language learners and provides an objective measure of language proficiency.

Details DOI

NeurIPS Conference 2021 Conference Paper

Understanding Adaptive, Multiscale Temporal Integration In Deep Speech Recognition Systems

Menoua Keshishian
Samuel Norman-Haignere
Nima Mesgarani

Natural signals such as speech are hierarchically structured across many different timescales, spanning tens (e. g. , phonemes) to hundreds (e. g. , words) of milliseconds, each of which is highly variable and context-dependent. While deep neural networks (DNNs) excel at recognizing complex patterns from natural signals, relatively little is known about how DNNs flexibly integrate across multiple timescales. Here, we show how a recently developed method for studying temporal integration in biological neural systems – the temporal context invariance (TCI) paradigm – can be used to understand temporal integration in DNNs. The method is simple: we measure responses to a large number of stimulus segments presented in two different contexts and estimate the smallest segment duration needed to achieve a context invariant response. We applied our method to understand how the popular DeepSpeech2 model learns to integrate across time in speech. We find that nearly all of the model units, even in recurrent layers, have a compact integration window within which stimuli substantially alter the response and outside of which stimuli have little effect. We show that training causes these integration windows to shrink at early layers and expand at higher layers, creating a hierarchy of integration windows across the network. Moreover, by measuring integration windows for time-stretched/compressed speech, we reveal a transition point, midway through the trained network, where integration windows become yoked to the duration of stimulus structures (e. g. , phonemes or words) rather than absolute time. Similar phenomena were observed in a purely recurrent and purely convolutional network although structure-yoked integration was more prominent in the recurrent network. These findings suggest that deep speech recognition systems use a common motif to encode the hierarchical structure of speech: integrating across short, time-yoked windows at early layers and long, structure-yoked windows at later layers. Our method provides a straightforward and general-purpose toolkit for understanding temporal integration in black-box machine learning models.

PDF Details

YNIMG Journal 2020 Journal Article

Brain-informed speech separation (BISS) for enhancement of target speaker in multitalker speech perception

Enea Ceolini
Jens Hjortkjær
Daniel D.E. Wong
James O’Sullivan
Vinay S. Raghavan
Jose Herrero
Ashesh D. Mehta
Shih-Chii Liu

Hearing-impaired people often struggle to follow the speech stream of an individual talker in noisy environments. Recent studies show that the brain tracks attended speech and that the attended talker can be decoded from neural data on a single-trial level. This raises the possibility of “neuro-steered” hearing devices in which the brain-decoded intention of a hearing-impaired listener is used to enhance the voice of the attended speaker from a speech separation front-end. So far, methods that use this paradigm have focused on optimizing the brain decoding and the acoustic speech separation independently. In this work, we propose a novel framework called brain-informed speech separation (BISS) 1 1 BISS: brain-informed speech separation. in which the information about the attended speech, as decoded from the subject’s brain, is directly used to perform speech separation in the front-end. We present a deep learning model that uses neural data to extract the clean audio signal that a listener is attending to from a multi-talker speech mixture. We show that the framework can be applied successfully to the decoded output from either invasive intracranial electroencephalography (iEEG) or non-invasive electroencephalography (EEG) recordings from hearing-impaired subjects. It also results in improved speech separation, even in scenes with background noise. The generalization capability of the system renders it a perfect candidate for neuro-steered hearing-assistive devices.

Details DOI

ICML Conference 2017 Conference Paper

Understanding the Representation and Computation of Multilayer Perceptrons: A Case Study in Speech Recognition

Tasha Nagamine
Nima Mesgarani

Despite the recent success of deep learning, the nature of the transformations they apply to the input features remains poorly understood. This study provides an empirical framework to study the encoding properties of node activations in various layers of the network, and to construct the exact function applied to each data point in the form of a linear transform. These methods are used to discern and quantify properties of feed-forward neural networks trained to map acoustic features to phoneme labels. We show a selective and nonlinear warping of the feature space, achieved by forming prototypical functions to account for the possible variation of each class. This study provides a joint framework where the properties of node activations and the functions implemented by the network can be linked together.

Details