VoxMind: An End-to-End Agentic Spoken Dialogue System
Summary
VoxMind introduces an end-to-end spoken dialogue system with agentic capabilities, using a "Think-before-Speak" mechanism and dynamic tool management to boost task completion from 34.88% to 74.57% while preserving conversational quality.
View Cached Full Text
Cached at: 04/21/26, 03:38 PM
Paper page - VoxMind: An End-to-End Agentic Spoken Dialogue System
Source: https://huggingface.co/papers/2604.15710
Abstract
VoxMind enhances spoken dialogue models with agentic capabilities through a “Think-before-Speak” mechanism and dynamic tool management to improve task completion rates while maintaining conversational quality.
Recentend-to-end spoken dialogue modelsenable natural interaction. However, as user demands become increasingly complex, models that rely solely on conversational abilities often struggle to cope. Incorporatingagentic capabilitiesis therefore essential: by enablingtool use, these models can extend their knowledge boundaries and better solve real-world tasks. Yet, existing research has largely concentrated on core perception and generation, with comparatively limited exploration of such tool-augmented extensions. To bridge this gap, we present VoxMind, an integrated framework designed to equipend-to-end spoken dialogue modelswith comprehensive agentic abilities. Leveraging our curated 470-hourAgentChat dataset, we incorporate a “Think-before-Speak” mechanism, enabling the model to internalize structured reasoning as a critical prerequisite for planning and response generation. Furthermore, to mitigate latency bottlenecks caused by large-scale tool integration, we propose aMulti-Agent Dynamic Tool Managementarchitecture. By asynchronously delegatingretrieval tasksto an auxiliary agent aligned with the main model’s reasoning trajectory, this system effectively decouplesinference latencyfrom toolset size. Experimental results confirm that VoxMind achieves significant improvements in agent performance: compared with strong baselines, thetask completion rateincreases from 34.88% to 74.57%, outperforming Gemini-2.5-Pro on spoken agent tasks while preserving generalconversational quality. The source code and associated data are publicly available at https://github.com/MM-Speech/VoxMind.
View arXiv pageView PDFGitHub21Add to collection
Get this paper in your agent:
hf papers read 2604\.15710
Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash
Models citing this paper1
#### leungtianle/VoxMind 8B• Updatedabout 7 hours ago • 17 • 1
Datasets citing this paper0
No dataset linking this paper
Cite arxiv.org/abs/2604.15710 in a dataset README.md to link it from this page.
Spaces citing this paper0
No Space linking this paper
Cite arxiv.org/abs/2604.15710 in a Space README.md to link it from this page.
Collections including this paper0
No Collection including this paper
Add this paper to acollectionto link it from this page.
Similar Articles
Thinking While Speaking: Inference-Time Knowledge Transfer for Responsive and Intelligent Conversational Voice Agents
This paper introduces a conversational voice agent system that uses a lightweight on-device 'Talker' model to start responding immediately, then incorporates knowledge from a frontier LLM 'Reasoner' as it becomes available, achieving 7-19x faster time-to-first-response while approaching frontier-level performance on a laptop.
Beyond Dyadic Memory: Interaction-Aware Multimodal Memory with Adaptive Agentic Retrieval for Multi-Party Spoken Conversations
Proposes VoxPolyMem, an interaction-aware multimodal memory framework with adaptive agentic retrieval for multi-party spoken conversations, achieving state-of-the-art performance on new benchmarks.
Vox
Vox is a product that enables users to talk to GitHub Copilot out loud, providing a voice interface for coding assistance.
Almost Human, Except When It Matters: VoxParity and the Decisions a Voice Should Change
VoxParity introduces a benchmark testing whether voice agents act on paralinguistic audio cues (background sounds, second voices, emotion) beyond transcripts, finding that most of 28 systems fail the words-only null test and routinely under-respond when audio signals the need for protection.
@svpino: Why do so many AI-powered phone agents sound smart until you interrupt them? Even when these agents give you reasonable…
Deepgram released Flux TTS, a streaming conversation-native text-to-speech model that retains tone, pacing, and context across turns, handles interruptions, and runs with latency as low as 80ms to make voice AI feel more natural.