Is Text All You Need? Text as a Universal Information Bottleneck for Speech LLMs

Large language models (LLMs) provide a powerful reasoning backbone for speech understanding, but integrating continuous acoustic signals into a frozen LLM remains challenging. Existing speech-to-LLM interfaces typically operate at two extremes: either enforcing near-discrete token alignment, which benefits transcription but loses paralinguistic information, or learning unconstrained continuous representations, which can drift away from the LLM's input space and degrade autoregressive decoding. In this work, we propose Convex Gate (C-Gate), a speech-to-LLM bridge that constrains all speech representations to lie within the LLM's input embedding manifold with an architectural convex-hull constraint. Concretely, each frame is represented as a convex combination of token embeddings, ensuring compatibility with the pretrained LLM while preserving continuous expressivity. Across automatic speech recognition (ASR) and emotion recognition, C-Gate achieves strong joint performance, improving LibriSpeech WER by up to 48.7% relative while matching or exceeding single-task emotion accuracy. Beyond performance, our analysis reveals a key insight: information is not carried by discrete token identities, but by time-resolved trajectories in the embedding space. Causal interventions confirm that both the trajectory structure and alignment to the pretrained embedding manifold are critical for performance. These results suggest that geometry, rather than token discreteness, is the fundamental design factor in speech-to-LLM interfaces, and provide a controlled regime for studying multimodal integration in frozen LLMs. We release the checkpoint, per-sample outputs, mechanism dumps, and intervention suite for replication.

Rethinking Depth: A study of the Recursive-Transformer for Speech Recognition

Transformer-based architectures have led to significant improvements in Automatic Speech Recognition (ASR), often at the cost of substantially increased model sizes. A promising approach to address this issue is layer sharing through depth recursion, commonly referred to as the Recursive-Transformer, which involves repeatedly applying the same layers within the model. Despite its potential shown in other fields, this technique remains relatively unexplored in ASR. In this paper, we present an experimental study of the Recursive-Transformer applied to ASR encoder architectures. We systematically investigate the impact of recursion depth and layer allocation within the Recursive-based Transformer. Our results demonstrate that the Recursive-Transformer is a viable alternative, especially when recurrence is applied in the latent space with a restricted number of loops, obtaining comparable performance while reducing the parameter count by 66%.

A study on the impact of region specific data on the performance of Indic ASR

Automatic Speech Recognition (ASR) systems are widely deployed across linguistically diverse regions, yet their ability to generalize across fine-grained geographic variation remains underexplored. We present a systematic study of cross-district ASR generalization for Indian languages, analyzing the impact of regional variation on performance. Using finetuning as a controlled probe, we train models on speech from a single district and evaluate them on other districts within the same language. We examine trends across multiple train test district pairs and quantify performance differences. To assess geographic effects, we analyze the correlation between WER and inter district distance using two distance measures. Our results show consistent correlations between geographic distance and WER, highlighting the challenges of regional generalization and the need for geographically diverse speech data in ASR development and evaluation in India.

Factors affecting ASR performance: A study using state of the art ASR models in Indic Languages

ASR performance varies across languages, speakers, and recording conditions, yet systematic analysis for Indic languages remain limited. We present a large-scale study of decoded outputs from multiple open-source ASR models evaluated on diverse Indian speech datasets in zero-shot settings. We analyze linguistic, speaker-level, and acoustic factors across Hindi, Bengali, Kannada, Telugu, and Marathi. We examine correlations between WER and speaker traits such as average word length, speaking rate, and utterance duration across multiple model dataset pairs. For Hindi, we further analyze audio factors including telephone codecs, bit depth, resampling, and background noise. Results reveal both cross lingual patterns and language-specific sensitivities, showing how speaker behavior and signal processing choices affect ASR robustness in real world Indic scenarios.

A Comparative Study of Pre-trained Speech Encoders and Training Objectives for Large-Scale Indic Spoken Language Identification

Spoken language identification (LID) for Indian languages is a challenging problem due to the large number of languages, significant phonetic overlap among related varieties, and the scarcity of labeled data for many low-resource languages. In this work, we present a systematic comparative study of two pre-trained speech encoders -- Whisper and FastConformer -- combined with a linear classifier for large-scale Indic LID spanning 42 languages across four linguistic families. We evaluate both encoders in frozen (linear probing) and fine-tuned settings, and compare three training objectives: cross-entropy (CE), supervised contrastive loss with cross entropy (CE + supCon), and hierarchical softmax (HSM). Models are trained on the Vaani dataset and evaluated in a cross-corpus setting on Vaani-Test (held-out), FLEURS, and Kathbath, providing insights into domain generalization. The frozen FastConformer encoder achieves over 90\% macro accuracy on FLEURS and Kathbath without any task-specific adaptation, substantially outperforming Whisper on out-of-domain benchmarks, while fine-tuned Whisper yields stronger in-domain performance. HSM consistently outperforms CE and CE+SupCon for both encoders across all benchmarks, with the largest gains on out-of-domain test sets. CE+SupCon degrades FastConformer's cross-corpus generalization, suggesting that the contrastive objective over-specializes representations to in-domain conditions. Per-family analysis shows that Central Indo-Aryan varieties are the hardest to discriminate, with Hindi--Urdu and the Sadri--Chhattisgarhi--Surgujia cluster being the dominant confusion pairs.

NüshuVoice: Reviving the Voice of Endangered Nüshu with Pitch-Aware Text-to-Speech

Nüshu is an endangered phonetic script historically used by women in Jiangyong County, southern Hunan, China. While existing computational studies of Nüshu mainly focus on textual digitization and visual recognition, the acoustic reconstruction of its authentic pronunciation remains largely unexplored. Building a Nüshu text-to-speech (TTS) system is particularly challenging because available recordings are extremely limited and mostly consist of isolated syllable-level pronunciations rather than natural sentence-level utterances. In this work, we introduce NüshuVoice, the first TTS benchmark for Nüshu. We construct a sentence-level Nüshu text-to-audio dataset that aligns standardized Unicode Nüshu text, phonetic transcriptions, standard Chinese translations, and archival recordings. To synthesize speech under this extreme low-resource setting, we propose Nüshu-PitchVITS, an F0-conditioned VITS framework that leverages Nüshu's five-level pitch notation as an explicit prosodic inductive bias. Experimental results show that Nüshu-PitchVITS outperforms strong TTS baselines in spectral fidelity, pitch reconstruction, and human-rated intelligibility. We publicly release the dataset and code at: https://anonymous.4open.science/r/Nvshu-TTS-2EB6.

Multi-View Speech Representation Learning for Parkinson's Disease Detection Using Context-guided Cross-modal Attention

Parkinson's disease (PD) is a progressive neurodegenerative disorder that frequently causes speech impairments associated with hypokinetic dysarthria. As speech production relies on the precise coordination of complex neuromuscular mechanisms, speech analysis has emerged as a promising non-invasive and cost-effective biomarker for early PD detection. Recent deep learning approaches have shown encouraging results; however, most existing methods rely on a single speech representation, potentially overlooking complementary pathological information encoded across different feature spaces. In this work, we propose a multi-branch deep learning framework for automatic PD detection from speech. Each recording is segmented into 5-second chunks and represented using three complementary modalities: Log-Mel spectrograms, MFCCs, and HuBERT embeddings extracted from raw waveforms. The spectrograms are processed using a pre-trained ResNet-18 encoder, MFCC sequences are modeled through a BiLSTM network, and raw speech is encoded using a pre-trained HuBERT model. To effectively integrate these heterogeneous representations, we introduce a context-guided cross-modal attention mechanism that dynamically weights temporal HuBERT embeddings according to the global acoustic context derived from the spectrogram and MFCC branches. Experiments conducted on the publicly available Spanish PC-GITA corpus under strict speaker-independent 5-fold cross-validation demonstrate the effectiveness of the proposed approach. The proposed architecture achieves an accuracy of 91.51%, an F1-score of 91.24%, and an AUC of 95.97%. Furthermore, ablation studies confirm the contribution of both the proposed context-guided cross-modal attention mechanism and the integration of complementary speech representations. These findings highlight the potential of heterogeneous speech modeling for robust and clinically reliable PD detection.

End-to-End Training for Discrete Token LLM based TTS System

Recent state-of-the-art (SOTA) text-to-speech (TTS) systems typically adopt a cascaded pipeline consisting of a speech tokenizer, an autoregressive large language model (LLM), and a diffusion based flow-matching (FM) model, with these components trained independently. In this paper, we propose a fully end-to-end (E2E) optimization framework that unifies the training of the speech tokenizer, LLM, FM model, and an additional reward model (RM). Specifically, we first jointly optimize the tokenizer using multi-task objectives derived from reconstruction for FM, next-token prediction for LLM, and multi recognition task for RM. This joint training encourages the discrete speech token space to capture acoustically and semantically salient information that is better tailored to TTS. We then further optimize the LLM using downstream reconstruction and recognition by FM and RM, which reduces inference-time mismatch and steers the LLM toward more preferred generations. Experimental results show that our E2E framework consistently outperforms cascaded baselines. On the Seed-TTS-Eval benchmark, our system achieves a word error rate (WER) of 0.78% and 1.56%, a new SOTA result with a 0.6B-parameter LLM and 0.5B-parameter FM model. These results validate that holistic E2E optimization is critical for improving discrete-token-based TTS systems with a much simpler training pipeline.

HoliDubber: Holistic Video Dubbing for Complex Acoustic Scenes via Text-Guided Audio Synthesis

Video dubbing is a cornerstone of multimedia content creation, aiming to synthesize synchronized acoustic sequences for visual streams. While Text-to-Speech (TTS) and Text-to-Audio (TTA) generation have each achieved remarkable progress, existing dubbing systems remain confined to isolated speech synthesis without incorporating sound effects and ambient audio, forcing practitioners to rely on fragmented workflows and laborious manual post-mixing. To address this limitation, we present HoliDubber, a holistic video dubbing framework that moves beyond speech-only generation by enabling the joint synthesis of speech and sound effects from a single text prompt. Specifically, HoliDubber adopts a patch-based autoregressive diffusion transformer architecture, where a causal language model autoregressively models aggregated patch embeddings to capture global temporal structure, and a Diffusion Transformer decoder generates high-fidelity continuous tokens within each patch, following a divide-and-conquer strategy. To achieve cross-modal alignment, visual features are encoded into patch-level representations and fused with audio patches via cross-attention, enabling the model to ground speech generation in the speaker's visual articulation dynamics. In addition, we introduce HoliDub-Bench, a benchmark curated from established datasets with synchronized video-text-audio triplets designed for holistic dubbing evaluation. Extensive experiments demonstrate that HoliDubber significantly outperforms existing methods across multiple benchmarks in speech quality, synchronization, and speaker similarity. Furthermore, results on HoliDub-Bench validate the effectiveness of joint speech-and-sound generation, establishing a new paradigm for holistic video dubbing in complex acoustic scenes. \footnote{The demo page of the project is https://holidubber.github.io}

BareWave: Waveform-Native Flow-Matching Text-to-Speech

Removing intermediate representations and separately trained decoding stages has become an important direction in generative modeling. In text-to-speech, however, high-quality systems are still commonly built through an intermediate acoustic representation before waveform synthesis. In this work, we present BareWave, a fully waveform-native framework for direct text-to-wave generation in flow-matching TTS. We consider this setting to raise three training challenges: raw-waveform modeling lacks a strong pretrained representational scaffold, different stages of training benefit from different noise schedules, and data-space perceptual objectives do not automatically share the temporal structure of the velocity-space flow objective. As a result, direct waveform training is hard to optimize efficiently, hard to push toward a strong final operating point with a fixed recipe, and hard to integrate effective perceptual refinement. Guided by this view, we develop a direct text-to-wave training framework that combines training-time representation alignment, staged noise scheduling, and velocity-aware perceptual alignment (VAPA), while preserving a single waveform-native inference path without pretrained components at test time. Experiments on zero-shot voice cloning show that strong intelligibility, speaker similarity, and naturalness can be achieved under a fully waveform-native inference path, supporting waveform-native flow-matching TTS as a practical direction. Project page with audio demos is available at https://barewave.github.io/.