InvestorNerd: An Investment and Financial Insights System Based on User Profiles
该研究通过构建一个基于用户资料的AI驱动平台InvestorNerd，解决了普通个体特别是无金融教育背景者理解投资和个人财务决策难题。
0 citationsRead paper
Selected work
Representative Papers
Seeds Before Objectives: Rethinking Evaluation for Low-Resource Garhwali ASR
This study addresses the high sensitivity of single-run evaluations to random seeds in low-resource Garhwali speech recognition, which often obscures genuine performance gains from stochastic noise. To remedy this, the authors establish the first reproducible ASR benchmark on the official VAANI dataset using multiple random seeds and propose a new evaluation paradigm centered on multi-seed assessment and statistical significance testing. Systematic re-evaluation of various optimization objectives and transfer strategies reveals that standard CTC combined with w2v-BERT 2.0 achieves a 47.0% WER across five seeds, outperforming larger models such as MMS-1B. While speed perturbation yields consistent minor improvements, more complex approaches like Focal CTC and matra weighting fail to demonstrate statistically significant gains. The findings underscore the fragility of common enhancements in low-resource settings and highlight the superiority of thoughtful pretraining design over mere model scale expansion.
0 citationsRead paper
Class-Specific Branch Attention for Mitigating Gradient Interference under Class Imbalance
This work addresses the suppression of minority-class learning in deep neural networks under severe class imbalance, which arises from shared feature representations. To mitigate this issue, the authors propose a Class-Specific Branch Attention (CSBA) mechanism. By analyzing inter-layer gradient flows and constructing a gradient conflict matrix based on class-specific gradient cosine similarity, they reveal—through the lens of optimization dynamics—for the first time how majority classes dominate and suppress gradients of minority classes. A lightweight channel reweighting module is integrated into a multi-branch convolutional architecture to implicitly decouple features and gradients in a class-aware manner. Experiments demonstrate that the proposed method significantly improves minority-class performance without compromising overall accuracy: the F1 score for the Physical-Damage class increases from 0.261 to 0.522, and Macro-F1 on CIFAR-10-LT rises from 0.595 to 0.655.
0 citationsRead paper
ProSarc: Prosody-Aware Sarcasm Recognition Framework via Temporal Prosodic Incongruity
This study addresses the challenge of detecting verbal sarcasm using only audio signals by modeling temporal prosodic incongruity between local prosodic dynamics and a global emotional baseline. The proposed approach employs a dual-encoder architecture to separately extract global affective and fine-grained prosodic features, coupled with an attention-driven incongruity analyzer for classification. Notably, it localizes sarcasm onset without requiring frame-level annotations and incorporates Monte Carlo Dropout to estimate uncertainty, thereby capturing perceptual ambiguity. Evaluated on multiple benchmarks, the method achieves 75.3% F1 on MUStARD++, and demonstrates robust performance on spontaneous speech in PodSarc (62.9% F1) and the cross-lingual MuSaG dataset (65.6% F1). Human evaluations further confirm a strong alignment between model-predicted uncertainty and subjective ambiguity in sarcasm perception.
0 citationsRead paper
Bridging the Gap: Converting Read Text to Conversational Dialogue
This work addresses the limitations of traditional read-style speech, which lacks the natural prosody required for conversational interaction and struggles to balance naturalness, intelligibility, and real-time performance in applications such as virtual assistants, customer service, and language learning. The authors propose PACC (Prosodic Adjustment with Conversational Context), a novel approach that integrates high-fidelity generative adversarial networks (HiFi-GAN) with context-aware prosody modeling. By leveraging deep neural networks to dynamically adjust intonation, stress, and rhythm, PACC enables high-quality conversion from read speech to natural conversational speech. Experimental results demonstrate that PACC significantly enhances both naturalness and intelligibility across multiple datasets, achieving state-of-the-art mean opinion scores (MOS) and establishing a new benchmark for voice conversion tasks.
0 citationsRead paper
Recent publications
Latest Papers
InvestorNerd: An Investment and Financial Insights System Based on User Profiles
该研究通过构建一个基于用户资料的AI驱动平台InvestorNerd，解决了普通个体特别是无金融教育背景者理解投资和个人财务决策难题。
0 citationsRead paper
Seeds Before Objectives: Rethinking Evaluation for Low-Resource Garhwali ASR
This study addresses the high sensitivity of single-run evaluations to random seeds in low-resource Garhwali speech recognition, which often obscures genuine performance gains from stochastic noise. To remedy this, the authors establish the first reproducible ASR benchmark on the official VAANI dataset using multiple random seeds and propose a new evaluation paradigm centered on multi-seed assessment and statistical significance testing. Systematic re-evaluation of various optimization objectives and transfer strategies reveals that standard CTC combined with w2v-BERT 2.0 achieves a 47.0% WER across five seeds, outperforming larger models such as MMS-1B. While speed perturbation yields consistent minor improvements, more complex approaches like Focal CTC and matra weighting fail to demonstrate statistically significant gains. The findings underscore the fragility of common enhancements in low-resource settings and highlight the superiority of thoughtful pretraining design over mere model scale expansion.
0 citationsRead paper
Class-Specific Branch Attention for Mitigating Gradient Interference under Class Imbalance
This work addresses the suppression of minority-class learning in deep neural networks under severe class imbalance, which arises from shared feature representations. To mitigate this issue, the authors propose a Class-Specific Branch Attention (CSBA) mechanism. By analyzing inter-layer gradient flows and constructing a gradient conflict matrix based on class-specific gradient cosine similarity, they reveal—through the lens of optimization dynamics—for the first time how majority classes dominate and suppress gradients of minority classes. A lightweight channel reweighting module is integrated into a multi-branch convolutional architecture to implicitly decouple features and gradients in a class-aware manner. Experiments demonstrate that the proposed method significantly improves minority-class performance without compromising overall accuracy: the F1 score for the Physical-Damage class increases from 0.261 to 0.522, and Macro-F1 on CIFAR-10-LT rises from 0.595 to 0.655.
0 citationsRead paper
ProSarc: Prosody-Aware Sarcasm Recognition Framework via Temporal Prosodic Incongruity
This study addresses the challenge of detecting verbal sarcasm using only audio signals by modeling temporal prosodic incongruity between local prosodic dynamics and a global emotional baseline. The proposed approach employs a dual-encoder architecture to separately extract global affective and fine-grained prosodic features, coupled with an attention-driven incongruity analyzer for classification. Notably, it localizes sarcasm onset without requiring frame-level annotations and incorporates Monte Carlo Dropout to estimate uncertainty, thereby capturing perceptual ambiguity. Evaluated on multiple benchmarks, the method achieves 75.3% F1 on MUStARD++, and demonstrates robust performance on spontaneous speech in PodSarc (62.9% F1) and the cross-lingual MuSaG dataset (65.6% F1). Human evaluations further confirm a strong alignment between model-predicted uncertainty and subjective ambiguity in sarcasm perception.
0 citationsRead paper
Bridging the Gap: Converting Read Text to Conversational Dialogue
This work addresses the limitations of traditional read-style speech, which lacks the natural prosody required for conversational interaction and struggles to balance naturalness, intelligibility, and real-time performance in applications such as virtual assistants, customer service, and language learning. The authors propose PACC (Prosodic Adjustment with Conversational Context), a novel approach that integrates high-fidelity generative adversarial networks (HiFi-GAN) with context-aware prosody modeling. By leveraging deep neural networks to dynamically adjust intonation, stress, and rhythm, PACC enables high-quality conversion from read speech to natural conversational speech. Experimental results demonstrate that PACC significantly enhances both naturalness and intelligibility across multiple datasets, achieving state-of-the-art mean opinion scores (MOS) and establishing a new benchmark for voice conversion tasks.
0 citationsRead paper