arXiv subjects

Brendan Fahy

Publications and source records attributed to Brendan Fahy.

16 recordsLinked to original sources

arXiv2026

Earnings25: A Comprehensive 500-Hour Speech Benchmark for Finance

We introduce Earnings25, a finance-domain benchmark for evaluating automatic speech recognition (ASR) on English-language earnings calls under realistic conditions. Earnings25 comprises two complementary test sets: (i) testset-full, 498 hours of full English-language S&P 500 earnings calls from Q4 2025, and (ii) testset-segmented, a 46-hour industry-balanced set of 290 segments sampled from English-language U.S. earnings calls in 2025. The benchmark provides aligned transcripts and structured metadata, including speaker roles, industry labels, and call structure, enabling speaker- and industry-aware evaluation beyond aggregate word error rate (WER). We report reproducible baselines for Whisper and Parakeet-TDT using standardized scoring.

cs.CL↗

arXiv2025

Adapting Whisper for Streaming Speech Recognition via Two-Pass Decoding

OpenAI Whisper is a family of robust Automatic Speech Recognition (ASR) models trained on 680,000 hours of audio. However, its encoder-decoder architecture, trained with a sequence-to-sequence objective, lacks native support for streaming ASR. In this paper, we fine-tune Whisper for streaming ASR using the WeNet toolkit by adopting a Unified Two-pass (U2) structure. We introduce an additional Connectionist Temporal Classification (CTC) decoder trained with causal attention masks to generate streaming partial transcripts, while the original Whisper decoder reranks these partial outputs. Our experiments on LibriSpeech and an earnings call dataset demonstrate that, with adequate fine-tuning data, Whisper can be adapted into a capable streaming ASR model. We also introduce a hybrid tokenizer approach, which uses a smaller token space for the CTC decoder while retaining Whisper's original token space for the attention decoder, resulting in improved data efficiency and generalization.

cs.SD↗

arXiv2019

Dialogue Act Classification in Group Chats with DAG-LSTMs

Dialogue act (DA) classification has been studied for the past two decades and has several key applications such as workflow automation and conversation analytics. Researchers have used, to address this problem, various traditional machine learning models, and more recently deep neural network models such as hierarchical convolutional neural networks (CNNs) and long short-term memory (LSTM) networks. In this paper, we introduce a new model architecture, directed-acyclic-graph LSTM (DAG-LSTM) for DA classification. A DAG-LSTM exploits the turn-taking structure naturally present in a multi-party conversation, and encodes this relation in its model structure. Using the STAC corpus, we show that the proposed method performs roughly 0.8% better in accuracy and 1.2% better in macro-F1 score when compared to existing methods. The proposed method is generic and not limited to conversation applications.

cs.CL↗

arXiv2017

Approaching the Bottom Using Fine Lattices With Domain-Wall Fermions

We explore the heavy-quark mass region above the charm mass using Möbius domain-wall fermions on fine lattices at $a = 0.080$, $0.055$, and $0.044$ fm. We examine masses and decay constants using a series of heavy quark masses up to 3 times the charm quark. We analyze the cutoff effects for heavy quarks above the charm and account for the leading order discretization effects using ideas from HQET. We extrapolate to the bottom quark mass and report preliminary results for $f_{B}$ and $f_{B_s}$

hep-lat↗

arXiv2017

Determination of charm quark mass from temporal moments of charmonium correlator with Mobius domain-wall fermion

We extract the charm quark mass and the strong coupling constant from the charmonium current correlators calculated with $n_f = 2 + 1$ Mobius domain wall fermions. We match our lattice calculation for the temporal moments of the correlator with perturbative result known up to four-loop order, and extract the charm quark mass with uncertainty less than 1%. Using the temporal moments, we also confirm the correlators in the vector channel to be consistent with the experimental data for the R-ratio. We used the ensembles generated by the JLQCD collaboration at lattice spacings $a$ = 0.080 fm, 0.055 fm and 0.044 fm.

hep-lat↗

arXiv2016

Short-distance charmonium correlator on the lattice with Möbius domain-wall fermion and a determination of charm quark mass

We calculate charmonium correlators on the lattice with 2+1-flavors of sea quarks and charm valence quark both described by the Möbius domain-wall fermion. Temporal moments of the correlators are calculated and matched to perturbative QCD formulae to extract the charm quark mass $m_c(μ)$ and strong coupling constant $α_s(μ)$. Lattice data at three lattice spacings, 0.044, 0.055, and 0.080~fm, are extrapolated to the continuum limit. The correlators in the vector channel are confirmed to be consistent with the experimental data for $e^+e^-\to c\bar{c}$, while the pseudo-scalar channel is used to extract $m_c(μ)$ and $α_s(μ)$. We obtain $m_\mathrm{c}(3 \mathrm{\ GeV})$ = 1.003(10)~GeV and $α_s^{\bar{\mathrm{MS}}(4)}(3\mathrm{\ GeV})$ = 0.253(13). Dominant source of the error is the truncation of perturbative expansion at $α_s^3$.

hep-lat↗

arXiv2016

$I=1$ and $I=2$ $π-π$ scattering phase shifts from $N_{\mathrm{f}} = 2+1$ lattice QCD

The $I=1$ $p$-wave and $I=2$ $s$-wave elastic $π$-$π$ scattering amplitudes are calculated from a first-principles lattice QCD simulation using a single ensemble of gauge field configurations with $N_{\mathrm{f}} = 2+1$ dynamical flavors of anisotropic clover-improved Wilson fermions. This ensemble has a large spatial volume $V=(3.7\mathrm{fm})^3$, pion mass $m_π = 230\mathrm{MeV}$, and spatial lattice spacing $a_s = 0.11\mathrm{fm}$. Calculation of the necessary temporal correlation matrices is efficiently performed using the stochastic LapH method, while the large volume enables an improved energy resolution compared to previous work. For this single ensemble we obtain $m_ρ/m_π = 3.350(24)$, $g_{ρππ} = 5.99(26)$, and a clear signal for the $I=2$ $s$-wave. The success of the stochastic LapH method in this proof-of-principle large-volume calculation paves the way for quantitative study of the lattice spacing effects and quark mass dependence of scattering amplitudes using state-of-the-art ensembles.

hep-lat↗

arXiv2015

Charmonium current-current correlators with Mobius domain-wall fermion

We calculate the charmonium correlators on the lattice with $n_f = 2+ 1$ Moebius domain wall fermion, and extract the charm quark mass and the strong coupling constant. Time moments are defined by current-current correlators, which have been calculated in the continuum theory by perturbation theory. We extract the charm quark mass by matching the lattice results with the corresponding perturbative QCD calculations, using the recently generated ensembles by the JLQCD collaboration at lattice spacings $a = 0.083, 0.055$, and $0.044$ fm.

hep-lat↗

arXiv2015

Pion-pion scattering and the timelike pion form factor from $N_{\mathrm{f}} = 2+1$ lattice QCD simulations using the stochastic LapH method

We report on progress applying the stochastic LapH method to estimate all-to-all propagators required in correlation functions of multi-hadron operators relevant for pion-pion scattering. Large-volume results for $I=2$ and $I=1$ pion-pion scattering phase shifts with good statistical precision are obtained from an $N_{\rm f} = 2+1$ anisotropic Wilson clover ensemble with $m_π = 240\mathrm{MeV}$. We also present a preliminary determination of the $I=1$ pion-pion scattering phase shift and timelike pion form factor on an isotropic $N_{\rm f}=2+1$ flavour ensemble generated by the Coordinated Lattice Simulation (CLS) community effort.

hep-lat↗

arXiv2015

Excited-state energies and scattering phase shifts from lattice QCD with the stochastic LapH method

Recent results in computing excited-state energies and meson-meson scattering phase shifts in lattice QCD are presented. A stochastic method of treating the low-lying modes of quark propagation that exploits Laplacian Heaviside quark-field smearing makes such studies possible now on large 32^3 x 256 and 48^3 x 128 lattices at near physical pion masses. Levels are identified using a variety of probe interpolating operators, which include both single-hadron and a large number of two-hadron operators.

hep-lat↗

arXiv2014

Excited isovector mesons using the stochastic LapH method

The spectrum of excited isovector mesons is studied using a 32^3 x 256 anisotropic lattice with u,d quark masses set to give a pion mass near 240 MeV. Results in the bosonic isovector nonstrange symmetry channels of zero total momentum are presented using correlation matrices of unprecedented size. In addition to spatially-extended single-meson operators, large numbers of two-meson operators are used, involving a wide variety of light isovector, isoscalar, and strange meson operators of varying relative momenta. All needed Wick contractions are efficiently evaluated using a stochastic method of treating the low-lying modes of quark propagation that exploits Laplacian Heaviside quark-field smearing. Level identification is discussed.

hep-lat↗

arXiv2014

Pion-pion scattering phase shifts with the stochastic LapH method

Progress in calculating scattering phase shifts on $N_f=2+1$ anisotropic clover Wilson lattices is described. The stochastic LapH method facilitates computations in large volumes and for light pion masses. Results for pion masses down to 240 MeV, keeping $m_πL > 4$, are presented.

hep-lat↗

arXiv2014

Excited states in lattice QCD with the stochastic LapH method

Progress in computing the spectrum of excited baryons and mesons in lattice QCD is described. Results in the zero-momentum bosonic I=1/2, S=1, T1u symmetry sector of QCD using a correlation matrix of 58 operators are presented. All needed Wick contractions are efficiently evaluated using a stochastic method of treating the low-lying modes of quark propagation that exploits Laplacian Heaviside quark-field smearing. Level identification using probe operators is discussed.

hep-lat↗

arXiv2014

The finite volume spectrum of excited states from lattice QCD simulations

We present results for the spectrum of excited mesons obtained from temporal correlations of spatially-extended single-hadron and multi-hadron operators computed in lattice QCD. The stochastic LapH algorithm is implemented on anisotropic, dynamical lattices for isovectors for pions of mass $390$ MeV. A large correlation matrix with single-particle and two-particle probe operators is diagonalized to identify resonances. The masses of excited states in the $I=1, S=0, T_{1u}^+$ channel as well as the mixing of single and multi-particle probe operators are presented.

hep-lat↗

arXiv2013

Spectrum of excited states using the stochastic LapH method

Progress in computing the spectrum of excited baryons and mesons in lattice QCD is described. Our first results in the zero-momentum bosonic I=1, S=0, T1u+ symmetry sector of QCD using a correlation matrix of 56 operators are presented. In addition to a dozen spatially-extended meson operators, 44 two-meson operators are used, involving a wide variety of light isovector, isoscalar, and strange meson operators of varying relative momenta. All needed Wick contractions are efficiently evaluated using a stochastic method of treating the low-lying modes of quark propagation that exploits Laplacian Heaviside quark-field smearing. Level identification is discussed.

hep-lat↗

arXiv2013

Applications of the Stochastic LapH Method

Progress in computing the hadron spectrum in lattice QCD using stochastic LapH quark propaga- tors is described. The stochastic LapH algorithm is a particular quark smearing algorithm that also allows the computation of all-to-all quark propagators. All-to-all quark propagators are required in our approach of using a large set of spatially extended hadron operators and explicit multi- particle operators to access excited states. We report on the progress made in the various isospin channels on 2+1 dynamical, anisotropic lattices generated by the Hadron Spectrum Collaboration.

hep-lat↗