The paper introduces TextPro‑SLM, a speech large language model that reduces the modality gap by making spoken input resemble a prosody‑aware text LLM. It combines WhisperPro, a unified speech encoder that outputs synchronized text tokens and prosody embeddings, with an LLM backbone that retains the semantic power of the original text LLM while learning paralinguistic cues. Experiments demonstrate that TextPro‑SLM achieves the lowest modality gap among leading SLMs at both 3B and 7B scales, and performs strongly on paralinguistic understanding tasks using only about 1,000 hours of audio for training.
arXiv:2609.15313v1 Announce Type: cross Abstract: Autoregressive generation of interleaved text and acoustic tokens is a common approach to spoken-response generation in speech large language models....
SPAR-K is a scheduled periodic alternating early‑exit framework for interleaved spoken language models that reduces decoding depth for speech tokens while maintaining quality. It lets most speech positions exit at a fixed intermediate layer and inserts periodic full‑depth refresh steps to counter distribution shift. Experiments on Step‑Audio‑2‑mini and GLM‑4‑Voice show up to 11 % depth reduction with less than 0.82 % drop in question‑answering accuracy and negligible impact on MOS and WER.
The paper investigates how Spoken Language Models (SLMs) process speech compared to text, noting that current SLMs show weak alignment between speech and text representations despite strong downstream performance. The authors propose a framework that separates length mismatch from semantic alignment to better match speech and text representations. Experiments on multiple benchmarks demonstrate that this approach yields competitive results against strong baselines, highlighting the need to explicitly address structural differences between speech and text in SLM training.
The paper introduces Convex Gate (C‑Gate), a bridge that maps each speech frame to a convex combination of frozen token embeddings, enabling a single LLM decoder to process both speech and text. Using a Whisper‑large‑v3 encoder and a partially adapted Qwen2.5‑7B‑Instruct decoder, the authors jointly train on speech recognition, emotion recognition, and spoken reasoning, achieving accurate transcription and emotion detection while maintaining strong performance on text‑only benchmarks. Analyses reveal that the fixed text embeddings can support both semantic and acoustic readouts through continuous, speech‑conditioned mixtures.
GEPARD is a streaming text‑to‑speech model that uses a standard large language model backbone to generate speech autoregressively, decoding audio with an FSQ‑based neural codec. It streams audio chunk‑by‑chunk as text arrives, achieving a real‑time factor of about 0.067 and an aggregate speedup of roughly 204× on a single GPU with 256 concurrent streams. The design keeps all complex auxiliary mechanisms outside the decode loop, enabling deployment with a standard LLM engine (vLLM) without kernel modifications.
arXiv:2609.09719v1 Announce Type: new Abstract: Text-aligned speech tokenization methods have emerged to better align speech tokens with LLM token spaces, enabling more effective utilization of pretr...
arXiv:2604.06487v2 Announce Type: replace Abstract: Conventional end-to-end automatic speech recognition (ASR) systems rely on paired speech-text data for domain adaptation. Recent LLM-based ASR arch...
GSLM2 is a multi‑stream generative spoken language model that learns complementary speech representations at multiple abstraction levels using the SpidR 2.0 encoder and a shared‑backbone SLM with stream‑specific predictors. It improves linguistic competence and scales more efficiently than prior speech models, narrowing the performance gap with text‑based language models by about two orders of magnitude on several benchmarks. The authors release the training code and checkpoints publicly.
arXiv:2610.08085v1 Announce Type: new Abstract: Speech-LLMs often exhibit prompt overfitting, where models solely trained on automatic speech recognition (ASR) instruction fail to generalize to new i...
The paper introduces AZeroS, a speech‑adapted large language model that uses Self‑Generated Instruction‑Free Tuning (SIFT) to avoid task‑specific instruction data. By training only two lightweight projection modules on 25,000 hours of speech‑text pairs and 3,000 hours of paralinguistic labels, AZeroS achieves state‑of‑the‑art results on semantic and paralinguistic benchmarks while keeping the core LLM and audio encoder frozen.
Uni-Stream is a unified streaming speech language model that supports both speech recognition and synthesis within a single architecture. Built on compact pretrained speech models, it processes interleaved text and speech in real time and achieves performance comparable to task‑specific models through a two‑stage training approach. The authors also introduce a data annotation pipeline for streaming training and plan to release the generated transcriptions and word‑level timestamps, while analyzing how speech and text representations occupy distinct latent spaces in the model.
arXiv:2511. 20973v2 Announce Type: replace-cross Abstract: Large Audio Language Models (LALMs) deliver strong performance across speech and audio tasks, but their audio encoders generate high-rate token sequences (e.