Committee

Neuphonic has released NeuTTS Air, an open-source text-to-speech (TTS) speech language model designed to run locally in real time on CPUs. The Hugging Face model card lists 748M parameters (Qwen2 architecture) and ships in GGUF quantizations (Q4/Q8), enabling inference through llama.cpp/llama-cpp-python without cloud dependencies. It is licensed under Apache-2.0 and includes a runnable demo and examples. So, what is new? NeuTTS Air couples a 0.5B-class Qwen backbone with Neuphonic’s NeuCodec audio codec. Neuphonic positions the system as a “super-realistic, on-device” TTS LM that clones a voice from ~3 seconds of reference audio and synthesizes speech in that style, targeting voice agents and privacy-sensitive applications. The model card and repository explicitly emphasize real-time CPU generation and small-footprint deployment. Key Features Realism at sub-1B scale: Human-like prosody and timbre preservation for a ~0.7B (Qwen2-class) text-to-speech LM. On-device deployment: Distributed in GGUF (Q4/Q8) with CPU-first paths; suitable for laptops, phones, and Raspberry Pi-class boards. Instant speaker cloning: Style transfer from ~3 seconds of reference audio (reference WAV + transcript). Compact LM+codec stack: Qwen 0.5B backbone paired with NeuCodec (0.8 kbps / 24 kHz) to balance latency, footprint, and output quality. Explain the model architecture and runtime path? Backbone: Qwen 0.5B used as a lightweight LM to condition speech generation; the hosted artifact is reported as 748M params under the qwen2 architecture on Hugging Face. Codec: NeuCodec provides low-bitrate acoustic tokenization/decoding; it targets 0.8 kbps with 24 kHz output, enabling compact representations for efficient on-device use. Quantization & format: Prebuilt GGUF backbones (Q4/Q8) are available; the repo includes instructions for llama-cpp-python and an optional ONNX decoder path. Dependencies: Uses espeak for phonemization; examples and a Jupyter notebook are provided for end-to-end synthesis. On-device performance focus NeuTTS Air showcases ‘real-time generation on mid-range devices‘ and offers CPU-first defaults; GGUF quantization is intended for laptops and single-board computers. While no fps/RTF numbers are published on the card, the distribution targets local inference without a GPU and demonstrates a working flow through the provided examples and Space. Voice cloning workflow NeuTTS Air requires (1) a reference WAV and (2) the transcript text for that reference. It encodes the reference to style tokens and then synthesizes arbitrary text in the reference speaker’s timbre. The Neuphonic team recommends 3–15 s clean, mono audio and provides pre-encoded samples. Privacy, responsibility, and watermarking Neuphonic frames the model for on-device privacy (no audio/text leaves the machine without user’s approval) and notes that all generated audio includes a Perth (Perceptual Threshold) watermarker to support responsible use and provenance. How it compares? Open, local TTS systems exist (e.g., GGUF-based pipelines), but NeuTTS Air is notable for packaging a small LM + neural codec with instant cloning, CPU-first quantizations, and watermarking under a permissive license. The “world’s first super-realistic, on-device speech LM” phrasing is the vendor’s claim; the verifiable facts are the size, formats, cloning procedure, license, and provided runtimes. Our Comments The focus is on system trade-offs: a ~0.7B Qwen-class backbone with GGUF quantization paired with NeuCodec at 0.8 kbps/24 kHz is a pragmatic recipe for real-time, CPU-only TTS that preserves timbre using ~3–15 s style references while keeping latency and memory predictable. The Apache-2.0 licensing and built-in watermarking are deployment-friendly, but publishing RTF/latency on commodity CPUs and cloning-quality vs. reference-length curves would enable rigorous benchmarking against existing local pipelines. Operationally, an offline path with minimal dependencies (eSpeak, llama.cpp/ONNX) lowers privacy/compliance risk for edge agents without sacrificing intelligibility. Check out the Model Card on Hugging Face and GitHub Page. Feel free to check out our GitHub Page for Tutorials, Codes and Notebooks. Also, feel free to follow us on Twitter and don’t forget to join our 100k+ ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well. The post Neuphonic Open-Sources NeuTTS Air: A 748M-Parameter On-Device Speech Language Model with Instant Voice Cloning appeared first on MarkTechPost.

arXiv:2405.20485v3 Announce Type: replace-cross Abstract: Retrieval Augmented Generation (RAG) expands the capabilities of modern large language models (LLMs), by anchoring, adapting, and personalizing their responses to the most relevant knowledge sources. It is particularly useful in chatbot applications, allowing developers to customize LLM output without expensive retraining. Despite their significant utility in various applications, RAG systems present new security risks. In this work, we propose a novel attack that allows an adversary to inject a single malicious document into a RAG system’s knowledge base, and mount a backdoor poisoning attack. We design Phantom, a general two-stage optimization framework against RAG systems, that crafts a malicious poisoned document leading to an integrity violation in the model’s output. First, the document is constructed to be retrieved only when a specific naturally occurring trigger sequence of tokens appears in the victim’s queries. Second, the document is further optimized with crafted adversarial text that induces various adversarial objectives on the LLM output, including refusal to answer, reputation damage, privacy violations, and harmful behaviors.We demonstrate our attacks on multiple open-source LLM architectures, including Gemma, Vicuna, and Llama, and show that they transfer to closed-source models such as GPT-3.5 Turbo and GPT-4. Finally, we successfully demonstrate our attack on an end-to-end black-box production RAG system: NVIDIA’s “Chat with RTX”.

arXiv:2510.01076v1 Announce Type: new Abstract: This article addresses embodied intelligence and reinforcement learning integration in the field of text processing, aiming to enhance text handling with more intelligence on the basis of embodied intelligence’s perception and action superiority and reinforcement learning’s decision optimization capability. Through detailed theoretical explanation and experimental exploration, a novel integration model is introduced. This model has been demonstrated to be very effective in a wide range oftext processing tasks, validating its applicative potential

arXiv:2510.00694v1 Announce Type: new Abstract: We introduce ALARB, a dataset and suite of tasks designed to evaluate the reasoning capabilities of large language models (LLMs) within the Arabic legal domain. While existing Arabic benchmarks cover some knowledge-intensive tasks such as retrieval and understanding, substantial datasets focusing specifically on multistep reasoning for Arabic LLMs, especially in open-ended contexts, are lacking. The dataset comprises over 13K commercial court cases from Saudi Arabia, with each case including the facts presented, the reasoning of the court, the verdict, as well as the cited clauses extracted from the regulatory documents. We define a set of challenging tasks leveraging this dataset and reflecting the complexity of real-world legal reasoning, including verdict prediction, completion of reasoning chains in multistep legal arguments, and identification of relevant regulations based on case facts. We benchmark a representative selection of current open and closed Arabic LLMs on these tasks and demonstrate the dataset’s utility for instruction tuning. Notably, we show that instruction-tuning a modest 12B parameter model using ALARB significantly enhances its performance in verdict prediction and Arabic verdict generation, reaching a level comparable to that of GPT-4o.

arXiv:2510.00977v1 Announce Type: cross Abstract: Group Relative Policy Optimization (GRPO) is a prominent reinforcement learning algorithm for post-training Large Language Models (LLMs). It is commonly believed that GRPO necessitates a large group size to ensure stable training via precise statistical estimation, which incurs substantial computational overhead. In this work, we challenge this assumption by reframing GRPO as a form of contrastive learning, which reveals a fundamental connection to Direct Preference Optimization (DPO). Motivated by DPO’s empirical success, we investigate the minimal two-rollout case (2-GRPO), a configuration previously deemed infeasible. We provide a rigorous theoretical analysis to validate 2-GRPO and demonstrate empirically that it achieves performance on par with 16-GRPO, despite using only 1/8 of the rollouts and reducing training time by over 70%.

arXiv:2509.23946v2 Announce Type: replace-cross Abstract: Chain-of-Thought (CoT) and its variants have markedly advanced the reasoning abilities of Large Language Models (LLMs), yet their monolithic and auto-regressive architecture inherently conflates high-level strategic planning with low-level step-by-step execution, leading to computational inefficiency, limited exploration of reasoning paths, and reduced interpretability. To overcome these issues, we propose the Explore-Execute Chain ($E^2C$), a structured reasoning framework that decouples reasoning into two distinct phases: an exploratory phase that stochastically generates succinct high-level plans, followed by an execution phase that deterministically carries out the chosen plan. Our approach incorporates a two-stage training methodology, which combines Supervised Fine-Tuning (SFT) – augmented by a novel data generation algorithm enforcing strict plan adherence – with a subsequent Reinforcement Learning (RL) stage that capitalizes on the informativeness of exploration and reinforces the determinism of execution. This decomposition enables an efficient test-time scaling strategy: on AIME’2024, $E^2C$ Test Time Scaling reaches 58.1% accuracy using

arXiv:2509.26124v1 Announce Type: new Abstract: When using an LLM to process text outside the training domain(s), an often overlooked factor is vocabulary mismatch, where the general-domain tokenizer fails to capture frequent domain-specific terms, leading to higher token fertility and thus a decrease in processing speed due to suboptimal sub-word splits. We address this limitation by augmenting the pretrained vocabulary with a set of domain-specific tokens. To this end, we design an algorithm that extends an existing tokenizer while guaranteeing it never decreases tokenization efficiency: every input sequence is segmented into at most the same number of tokens as before. Evaluated on real-world e-Commerce use-cases, the augmented tokenizer significantly shortens input sequences by up to 20% and reduces inference latency on downstream tasks while preserving predictive quality. We further analyze secondary effects, such as the impact on forward pass speed and the rate at which the model adopts the newly introduced tokens, to illustrate the broader benefits of vocabulary adaptation.

arXiv:2509.26313v1 Announce Type: new Abstract: Supervised fine-tuning (SFT) is the predominant method for adapting large language models (LLMs), yet it often struggles with generalization compared to reinforcement learning (RL). In this work, we posit that this performance disparity stems not just from the loss function, but from a more fundamental difference: SFT learns from a fixed, pre-collected dataset, whereas RL utilizes on-policy data sampled from the current policy. Building on this hypothesis, we introduce one-token rollout (OTR), a novel fine-tuning algorithm that guides SFT with the policy gradient method. OTR reframes the autoregressive learning process by treating each token generation as a single-step reinforcement learning trajectory. At each step, it performs a Monte Carlo “rollout” by sampling multiple candidate tokens from the current policy’s distribution. The ground-truth token from the supervised data is then used to provide a reward signal to these samples. Guided by policy gradient, our algorithm repurposes static, off-policy supervised data into a dynamic, on-policy signal at the token level, capturing the generalization benefits of on-policy learning while bypassing the costly overhead of full sentence generation. Through extensive experiments on a diverse suite of challenging benchmarks spanning mathematical reasoning, code generation, and general domain reasoning, we demonstrate that OTR consistently outperforms standard SFT. Our findings establish OTR as a powerful and practical alternative for fine-tuning LLMs and provide compelling evidence that the on-policy nature of data is a critical driver of generalization, offering a promising new direction for fine-tuning LLMs.