Are Mamba-based Audio Foundation Models the Best Fit for Non-Verbal Emotion Recognition?

Mohd Mujtaba Akhtar; Orchid Chetia Phukan; Girish; Swarup Ranjan Behera; Ananda Chandra Nayak; Sanjib Kumar Nayak; Arun Balaji Buduru; Rajesh Sharma

arXiv · 2506.02258

Are Mamba-based Audio Foundation Models the Best Fit for Non-Verbal Emotion Recognition?

Mohd Mujtaba Akhtar·Orchid Chetia Phukan·Girish·Swarup Ranjan Behera·Ananda Chandra Nayak·Sanjib Kumar Nayak·Arun Balaji Buduru·Rajesh Sharma

View original source ↗Read source document ↗Export metadata ↓

Abstract

In this work, we focus on non-verbal vocal sounds emotion recognition (NVER). We investigate mamba-based audio foundation models (MAFMs) for the first time for NVER and hypothesize that MAFMs will outperform attention-based audio foundation models (AAFMs) for NVER by leveraging its state-space modeling to capture intrinsic emotional structures more effectively. Unlike AAFMs, which may amplify irrelevant patterns due to their attention mechanisms, MAFMs will extract more stable and context-aware representations, enabling better differentiation of subtle non-verbal emotional cues. Our experiments with state-of-the-art (SOTA) AAFMs and MAFMs validates our hypothesis. Further, motivated from related research such as speech emotion recognition, synthetic speech detection, where fusion of foundation models (FMs) have showed improved performance, we also explore fusion of FMs for NVER. To this end, we propose, RENO, that uses renyi-divergence as a novel loss function for effective alignment of the FMs. It also makes use of self-attention for better intra-representation interaction of the FMs. With RENO, through the heterogeneous fusion of MAFMs and AAFMs, we show the topmost performance in comparison to individual FMs, its fusion and also setting SOTA in comparison to previous SOTA work.

Explore related subjects

eess.AS cs.SD

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Mohd Mujtaba Akhtar, Orchid Chetia Phukan, Girish, Swarup Ranjan Behera, Ananda Chandra Nayak, Sanjib Kumar Nayak, Arun Balaji Buduru, Rajesh Sharma. 2025-06-02. Are Mamba-based Audio Foundation Models the Best Fit for Non-Verbal Emotion Recognition?. https://arxiv.org/abs/2506.02258

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Recovering the Zipfian Distribution in Unsupervised Term Discovery

Unsupervised term discovery involves segmenting unlabelled speech into word- or syllable-like units and clustering these into a lexicon of candidate types. True lexicons follow a Zipfian distribution, yet the dominant centre-based clustering approach -- K-means -- produces a more uniform distribution due to an inductive bias toward spherical clusters. In this paper we revisit graph-based clustering as a bottom-up alternative, where segment embeddings are connected by pairwise similarity and partitioned using the Leiden algorithm. We show that graph clustering substantially outperforms centre-based approaches (K-means, GMM, BIRCH) in both word- and syllable-level lexicon discovery across three languages, producing more Zipf-like distributions. Another bottom-up approach, agglomerative clustering with average linkage, also performs well, although it is computationally less efficient and allows for less control over the resulting distribution. Our work calls into question the dominance of centre-based clustering for term discovery, and promotes graph clustering as an attractive alternative.

eess.AS↗

arXiv2026

Preserving Speech-to-Text LLM Capabilities in Speech-to-Speech Generation

Strong speech-to-text (S2T) LLMs already provide robust speech perception and text reasoning, but adding speech-to-speech (S2S) output is challenging: fine-tuning the backbone can degrade the original S2T performance, while attaching a downstream talker reintroduces a serial text-to-speech bottleneck. We present PRIME-Speech, a frozen-backbone S2S conversion framework that trains only speech-generation modules. PRIME-Speech synchronizes a causal audio post-decoder with intermediate hidden states of the frozen backbone, so codec tokens are generated from the model's evolving reasoning trajectory rather than from completed text chunks. The post-decoder uses mixed hidden-state, text, and audio-history conditioning, and a training-time packing strategy with turn-level audio KV-cache and position reset stabilizes multi-turn spoken interaction without additional multi-turn S2S training data. Multi-token prediction further reduces the effective codec prediction rate and improves first-audio latency without modifying the reasoning path. Across speech translation, spoken QA, speech understanding, and multi-turn dialogue, PRIME-Speech preserves the S2T behavior of the frozen backbone while producing accurate, low-WER spoken responses.

eess.AS↗

arXiv2026

Brain2Speech-Net: Fast and Intelligible Brain-to-Speech Synthesis Without Text Decoding

The loss of speech limits communication for individuals with paralysis. Direct neural-to-speech synthesis is challenging due to the limited availability of neural data for training speech brain-computer interfaces. Most existing systems rely on cascaded neural-to-text-to-speech pipelines, which increase inference latency and propagate errors across stages. We present Brain2Speech-Net, a single-stage neural-to-speech generation framework without intermediate text decoding. We use a differentiable phoneme bottleneck and a deep-HMM alignment mechanism to map long neural recordings into the latent space of a text-to-speech (TTS) model, enabling high-quality speech synthesis. Brain2Speech-Net is the only system in our comparison that produces intelligible speech while generating faster than real time.

eess.AS↗