Datasets:

beatsprom
/

audio-speech-realtime-voice-agents-2026

paper_id string	title string	published_date string	arxiv_categories list	legal_license string	license_url string	commercial_use_allowed bool	commercial_ip_safety_score int64	commercial_ip_verdict string	commercial_ip_justification string	field_weighted_citation_impact_fwci float64	tldr_neural_summary string	primary_author string	total_authors_count int64	authors list	affiliations list	arxiv_url string	code_audit_status string	primary_repo_url string	repo_urls list	github_stars int64	github_forks int64	github_last_commit string	github_license string	importance_score float64	abstract string	title_vector_384d list	abstract_vector_384d list	target_industry string	audio_speech_task_paradigm string	speech_latency_profile string	acoustic_sampling_rates list	speech_foundation_backbone string	tested_benchmarks list	extracted_benchmark_scores list	benchmark_leaderboard_json list	semantic_nearest_neighbors_top3 list	reproduction_recipe string	core_problem_addressed string	key_technical_innovation string	key_quantitative_result string	cluster_id int64	cluster_topic_name string	trend_velocity_tier string	papers_last_30_days int64	audited_at string
2603.06193v2	Whisper-CD: Accurate Long-Form Speech Recognition using Multi-Negative Contrastive Decoding	2026-03-06T12:04:12Z	[ "cs.SD", "cs.AI", "eess.AS" ]	ArXiv Standard	http://arxiv.org/licenses/nonexclusive-distrib/1.0/	true	60	Internal R&D Only (Copyleft or Academic Terms)	ArXiv Standard Distribution; Repository License Unspecified	1	Proposes Whisper-CD, a training-free contrastive decoding framework that contrasts clean-audio logi... to enhance real-time voice and audio generation, achieving Across five English long-form benchmarks, Whisper-CD reduces WER by up to 24.3pp....	Hoseong Ahn	4	[ "Hoseong Ahn", "Jeongyun Chae", "Yoonji Park", "Kyuhong Shim" ]	[ "Speech & Audio AI Intelligence Lab" ]	http://arxiv.org/abs/2603.06193v2	VERIFIED_LIVE	https://github.com/openai/whisper	[ "https://github.com/openai/whisper" ]	108,000	0	2026-08-22	Unspecified	133.91	Long-form speech recognition with large encoder-decoder models such as Whisper often exhibit hallucinations, repetition loops, and content omissions. These errors can accumulate and be further amplified when the previous segment's transcription is used as decoding context. We propose Whisper-CD, a training-free contras...	[ -0.02266700007021427, -0.0623599998652935, 0.060152001678943634, -0.07590600103139877, -0.07109499722719193, 0.02861200086772442, -0.016620999202132225, -0.08219499886035919, -0.005398999899625778, -0.05955599993467331, -0.008589000441133976, 0.017587000504136086, -0.04441500082612038, 0.0...	[ -0.014950999990105629, -0.11826200038194656, 0.04373199865221977, -0.015355000272393227, 0.026823999360203743, -0.034756001085042953, -0.004683000035583973, -0.10786200314760208, 0.04429800063371658, -0.08925200253725052, -0.04279400035738945, -0.05453500151634216, -0.03263799846172333, 0....	Audio AI, Speech Foundation Models & Real-Time Voice Agents	End-to-End Multilingual Speech Recognition (ASR)	Streaming Chunk-based ASR (Causal Conformer)	[ "24 kHz High-Quality Speech", "16 kHz Wideband ASR Standard" ]	Neural Continuous & Discrete Acoustic Transformer	[ "LibriSpeech & Seed-TTS Zero-Shot Audio Benchmark" ]	[ "Sub-200ms Latency & High-Fidelity SOTA Acoustic Metric" ]	[ { "benchmark_name": "LibriSpeech & Seed-TTS Zero-Shot Audio Benchmark", "metric": "WER / MOS / RTF / Latency", "score": "Sub-200ms Latency & High-Fidelity SOTA Acoustic Metric" } ]	[ { "paper_id": "2604.15383v1", "title": "Temporal Contrastive Decoding: A Training-Free Method for Large Audio-Language Models", "cosine_sim": 0.7414 }, { "paper_id": "2510.25150v1", "title": "Explainable Disentanglement on Discrete Speech Representations for Noise-Robust ASR", "cosine_si...	git clone https://github.com/openai/whisper && cd whisper && (pip install -e . \|\| pip install -r requirements.txt)	Long-form speech recognition with large encoder-decoder models such as Whisper often exhibit hallucinations, repetition loops, and content omissions.	We propose Whisper-CD, a training-free contrastive decoding framework that contrasts clean-audio logits against negative logits computed from three acoustically motivated perturbations: Gaussian noise injection, silence signal, and audio temporal shift.	Across five English long-form benchmarks, Whisper-CD reduces WER by up to 24.3pp on CORAAL and shows 48% faster token generation throughput than beam search.	2	Low-Latency Zero-Shot TTS & Instant Voice Cloning	Explosive (>50/mo)	126	2026-08-22T14:26:56.211046
2607.01108v1	NPUsper: Eliminating Redundant Computation for Real-Time Whisper on Mobile NPUs	2026-07-01T16:00:16Z	[ "cs.SD" ]	ArXiv Standard	http://arxiv.org/licenses/nonexclusive-distrib/1.0/	true	80	Enterprise Safe (Commercial Training & Deployment Allowed)	ArXiv Standard Distribution; Permissive Open-Source Software (MIT)	1	Proposes For efficient mobile-NPU execution, we propose controlled unrolling, which executes autore... to enhance real-time voice and audio generation, achieving NPUsper achieves up to 4.84x lower per-word latency, up to 33.2x lower time-to-f....	Sihyeon Lee	6	[ "Sihyeon Lee", "Hojeong Lee", "Sungwon Woo", "Chengpo Yan", "Suman Banerjee", "Seyeon Kim" ]	[ "Speech & Audio AI Intelligence Lab" ]	http://arxiv.org/abs/2607.01108v1	VERIFIED_LIVE	https://github.com/ggml-org/whisper.cpp	[ "https://github.com/ggml-org/whisper.cpp", "https://github.com/npusper/NPUsper" ]	53,096	6,087	2026-08-22	MIT	132.42	We present NPUsper, a live transcription system that makes Whisper efficient on mobile NPUs by eliminating redundant computation. To avoid the heavy padding used by prior streaming systems, NPUsper detects hallucinated tokens online from temporal patterns in decoder cross-attention, allowing each inference round to pro...	[ -0.04177499935030937, -0.01759899966418743, 0.04320000112056732, 0.019021999090909958, 0.040686000138521194, -0.1333480030298233, 0.0154600003734231, -0.08744499832391739, 0.015832999721169472, 0.020867999643087387, -0.005843999795615673, -0.04442699998617172, -0.03385400027036667, -0.0731...	[ -0.07623499631881714, -0.048836998641490936, 0.00612299982458353, 0.002452000044286251, 0.05337100103497505, -0.04251199960708618, -0.07883100211620331, -0.09771300107240677, 0.013667000457644463, -0.001104000024497509, -0.06088399887084961, -0.0388530008494854, -0.10634200274944305, -0.02...	Audio AI, Speech Foundation Models & Real-Time Voice Agents	End-to-End Multilingual Speech Recognition (ASR)	Streaming Chunk-based ASR (Causal Conformer)	[ "24 kHz High-Quality Speech", "16 kHz Wideband ASR Standard" ]	Neural Continuous & Discrete Acoustic Transformer	[ "LibriSpeech & Seed-TTS Zero-Shot Audio Benchmark" ]	[ "Sub-200ms Latency & High-Fidelity SOTA Acoustic Metric" ]	[ { "benchmark_name": "LibriSpeech & Seed-TTS Zero-Shot Audio Benchmark", "metric": "WER / MOS / RTF / Latency", "score": "Sub-200ms Latency & High-Fidelity SOTA Acoustic Metric" } ]	[ { "paper_id": "2604.25611v1", "title": "WhisperPipe: A Resource-Efficient Streaming Architecture for Real-Time Automatic Speech Recognition", "cosine_sim": 0.7507 }, { "paper_id": "2509.15969v2", "title": "VoXtream: Full-Stream Text-to-Speech with Extremely Low Latency", "cosine_sim": 0....	git clone https://github.com/ggml-org/whisper.cpp && cd whisper.cpp && (pip install -e . \|\| pip install -r requirements.txt)	We present NPUsper, a live transcription system that makes Whisper efficient on mobile NPUs by eliminating redundant computation.	For efficient mobile-NPU execution, we propose controlled unrolling, which executes autoregressive decoding as K-step chunk graphs, removing unnecessary KV-cache computation and reducing graph-dispatch overhead.	NPUsper achieves up to 4.84x lower per-word latency, up to 33.2x lower time-to-first-token (TTFT), and up to 88.64% lower average power consumption compared with baselines, while maintaining comparable transcription accuracy.	5	Generative Music, Audio Diffusion & Sound Synthesis	Explosive (>50/mo)	126	2026-08-22T14:25:37.543529
2601.18184v2	VIBEVOICE-ASR Technical Report	2026-01-26T06:11:51Z	[ "cs.SD", "cs.AI", "eess.AS" ]	ArXiv Standard	http://arxiv.org/licenses/nonexclusive-distrib/1.0/	true	60	Internal R&D Only (Copyleft or Academic Terms)	ArXiv Standard Distribution; Repository License Unspecified	1	Proposes Unlike traditional pipelined approaches that rely on audio chunking, VibeVoice-ASRsupports... to enhance real-time voice and audio generation, achieving Furthermore, we introduce a prompt-based context injection mechanism that allows....	Zhiliang Peng	24	[ "Zhiliang Peng", "Jianwei Yu", "Yaoyao Chang", "Zilong Wang", "Li Dong", "Yingbo Hao", "Yujie Tu", "Chenyu Yang", "Wenhui Wang", "Songchen Xu", "Yutao Sun", "Hangbo Bao", "Weijiang Xu", "Yi Zhu", "Zehua Wang", "Ting Song", "Yan Xia", "Zewen Chi", "Shaohan Huang", "Liang Wang", ...	[ "Speech & Audio AI Intelligence Lab" ]	http://arxiv.org/abs/2601.18184v2	VERIFIED_LIVE	https://github.com/microsoft/VibeVoice	[ "https://github.com/microsoft/VibeVoice" ]	53,100	0	2026-08-22	Unspecified	126.18	This report presents VibeVoice-ASR, a general-purpose speech understanding framework built upon VibeVoice, designed to address the persistent challenges of context fragmentation and multi-speaker complexity in long-form audio (e.g., meetings, podcasts) that remain despite recent advancements in short-form speech recogn...	[ -0.08646500110626221, 0.026520999148488045, -0.040502000600099564, 0.013597000390291214, -0.06326799839735031, -0.005998999811708927, -0.07846099883317947, -0.01896899938583374, -0.03384900093078613, -0.056327998638153076, 0.020601000636816025, 0.007422000169754028, 0.019658999517560005, -...	[ -0.08629699796438217, -0.009196000173687935, -0.0020639998838305473, -0.06459099799394608, -0.041367001831531525, -0.004050000105053186, 0.009600999765098095, -0.04848200082778931, 0.0037430000957101583, -0.12686200439929962, -0.09053800255060196, -0.06486500054597855, -0.042743999511003494,...	Audio AI, Speech Foundation Models & Real-Time Voice Agents	End-to-End Multilingual Speech Recognition (ASR)	Offline Long-Form Transcription & Analysis	[ "24 kHz High-Quality Speech", "16 kHz Wideband ASR Standard" ]	Neural Continuous & Discrete Acoustic Transformer	[ "LibriSpeech & Seed-TTS Zero-Shot Audio Benchmark" ]	[ "Sub-200ms Latency & High-Fidelity SOTA Acoustic Metric" ]	[ { "benchmark_name": "LibriSpeech & Seed-TTS Zero-Shot Audio Benchmark", "metric": "WER / MOS / RTF / Latency", "score": "Sub-200ms Latency & High-Fidelity SOTA Acoustic Metric" } ]	[ { "paper_id": "2606.01483v1", "title": "MURMUR: An Efficient Inference System for Long-Form ASR", "cosine_sim": 0.7174 }, { "paper_id": "2603.04809v1", "title": "WhisperAlign: Word-Boundary-Aware ASR and WhisperX-Anchored Pyannote Diarization for Long-Form Bengali Speech", "cosine_sim": ...	git clone https://github.com/microsoft/VibeVoice && cd VibeVoice && (pip install -e . \|\| pip install -r requirements.txt)	This report presents VibeVoice-ASR, a general-purpose speech understanding framework built upon VibeVoice, designed to address the persistent challenges of context fragmentation and multi-speaker complexity in long-form audio (e.g., meetings, podcasts) that remain despite recent advancements in short-form speech recogn...	Unlike traditional pipelined approaches that rely on audio chunking, VibeVoice-ASRsupports single-pass processing for up to 60 minutes of audio.	Furthermore, we introduce a prompt-based context injection mechanism that allows users to supply customized conetxt, significantly improving accuracy on domain-specific terminology and polyphonic character disambiguation.	8	Spoken Dialogue Evaluation & Acoustic Benchmarking	Explosive (>50/mo)	126	2026-08-22T14:27:21.041434
2605.30748v2	Chatterbox-Flash: Prior-Calibrated Block Diffusion for Streaming Zero-Shot TTS	2026-05-29T02:25:02Z	[ "cs.SD", "cs.AI", "eess.AS" ]	ArXiv Standard	http://arxiv.org/licenses/nonexclusive-distrib/1.0/	true	60	Internal R&D Only (Copyleft or Academic Terms)	ArXiv Standard Distribution; Repository License Unspecified	1	Proposes We find that naively transferring mainstream block-diffusion decoding to discrete speech t... to enhance real-time voice and audio generation, achieving On standard zero-shot TTS benchmarks, Chatterbox-Flash attains high-fidelity syn....	Deokjin Seo	3	[ "Deokjin Seo", "Gangin Park", "Kihyun Nam" ]	[ "Speech & Audio AI Intelligence Lab" ]	http://arxiv.org/abs/2605.30748v2	VERIFIED_LIVE	https://github.com/resemble-ai/chatterbox	[ "https://github.com/resemble-ai/chatterbox", "https://github.com/resemble-ai/chatterbox-flash" ]	26,100	0	2026-08-22	Unspecified	124.93	We present Chatterbox-Flash, a zero-shot text-to-speech model obtained by fine-tuning a pretrained autoregressive TTS decoder into a block-diffusion decoder, enabling parallel token generation within each block while retaining block-by-block streaming. We find that naively transferring mainstream block-diffusion decodi...	[ 0.052170999348163605, -0.06516599655151367, 0.02769700065255165, -0.057916998863220215, 0.030990000814199448, -0.03824099898338318, -0.017077000811696053, -0.03331400081515312, 0.029371999204158783, -0.010625000111758709, 0.030740000307559967, 0.005175999831408262, -0.11700599640607834, 0....	[ -0.03656600043177605, -0.13157400488853455, 0.020726000890135765, 0.011812999844551086, 0.05767599865794182, 0.005609000101685524, -0.07415500283241272, -0.07460000365972519, 0.09041599929332733, -0.037682000547647476, -0.024839000776410103, -0.0307839997112751, -0.11639899760484695, 0.012...	Audio AI, Speech Foundation Models & Real-Time Voice Agents	Low-Latency Zero-Shot TTS & Voice Cloning	Interactive Low-Latency Voice Synthesis (<300ms)	[ "24 kHz High-Quality Speech", "16 kHz Wideband ASR Standard" ]	Neural Continuous & Discrete Acoustic Transformer	[ "LibriSpeech & Seed-TTS Zero-Shot Audio Benchmark" ]	[ "Sub-200ms Latency & High-Fidelity SOTA Acoustic Metric" ]	[ { "benchmark_name": "LibriSpeech & Seed-TTS Zero-Shot Audio Benchmark", "metric": "WER / MOS / RTF / Latency", "score": "Sub-200ms Latency & High-Fidelity SOTA Acoustic Metric" } ]	[ { "paper_id": "2603.26364v1", "title": "LLaDA-TTS: Unifying Speech Synthesis and Zero-Shot Editing via Masked Diffusion Modeling", "cosine_sim": 0.7562 }, { "paper_id": "2608.00011v1", "title": "DLLM-TTS: Block Discrete Diffusion Language Model for Text-to-Speech Synthesis", "cosine_sim"...	git clone https://github.com/resemble-ai/chatterbox && cd chatterbox && (pip install -e . \|\| pip install -r requirements.txt)	We present Chatterbox-Flash, a zero-shot text-to-speech model obtained by fine-tuning a pretrained autoregressive TTS decoder into a block-diffusion decoder, enabling parallel token generation within each block while retaining block-by-block streaming.	We find that naively transferring mainstream block-diffusion decoding to discrete speech tokens degrades quality, as a long-tail token distribution biases parallel position selection toward a few high-frequency tokens.	On standard zero-shot TTS benchmarks, Chatterbox-Flash attains high-fidelity synthesis comparable to strong autoregressive and non-autoregressive baselines, while supporting streaming inference with time-to-first-packet on par with streaming AR systems and substantially lower real-time factor.	3	End-to-End Multilingual Speech Recognition (ASR & Whisper)	Explosive (>50/mo)	126	2026-08-22T14:26:10.592107
2603.08823v2	Fish Audio S2 Technical Report	2026-03-09T18:34:33Z	[ "cs.SD", "cs.AI", "cs.CL" ]	ArXiv Standard	http://arxiv.org/licenses/nonexclusive-distrib/1.0/	true	60	Internal R&D Only (Copyleft or Academic Terms)	ArXiv Standard Distribution; Repository License Unspecified	1	Proposes To scale training, we develop a multi-stage training recipe together with a staged data pi... to enhance real-time voice and audio generation, achieving We highly encourage readers to visit https://fish.audio to try custom voices..	Shijia Liao	14	[ "Shijia Liao", "Yuxuan Wang", "Songting Liu", "Yifan Cheng", "Ruoyi Zhang", "Tianyu Li", "Shidong Li", "Yisheng Zheng", "Xingwei Liu", "Qingzheng Wang", "Zhizhuo Zhou", "Jiahua Liu", "Xin Chen", "Dawei Han" ]	[ "Speech & Audio AI Intelligence Lab" ]	http://arxiv.org/abs/2603.08823v2	VERIFIED_LIVE	https://github.com/fishaudio/fish-speech	[ "https://github.com/fishaudio/fish-speech" ]	32,299	0	2026-08-22	Unspecified	123.54	We introduce Fish Audio S2, an open-sourced text-to-speech system featuring multi-speaker, multi-turn generation, and, most importantly, instruction-following control via natural-language descriptions. To scale training, we develop a multi-stage training recipe together with a staged data pipeline covering video captio...	[ -0.0470069982111454, -0.02628600038588047, -0.0433490015566349, -0.07602299749851227, -0.003010000102221966, 0.030657999217510223, -0.05734499916434288, 0.04521999880671501, -0.07632900029420853, -0.05581599846482277, -0.006436000112444162, -0.04111799970269203, -0.04141800105571747, -0.02...	[ -0.0814250037074089, -0.11005699634552002, 0.01486899983137846, -0.05286400020122528, 0.0018520000157877803, 0.04314899817109108, -0.037710998207330704, -0.005055000074207783, -0.0153609998524189, -0.09036999940872192, -0.05306499823927879, -0.1069440022110939, -0.03030100092291832, 0.0023...	Audio AI, Speech Foundation Models & Real-Time Voice Agents	Low-Latency Zero-Shot TTS & Voice Cloning	Interactive Low-Latency Voice Synthesis (<300ms)	[ "24 kHz High-Quality Speech", "16 kHz Wideband ASR Standard" ]	Neural Continuous & Discrete Acoustic Transformer	[ "LibriSpeech & Seed-TTS Zero-Shot Audio Benchmark" ]	[ "RTF of 0.195" ]	[ { "benchmark_name": "LibriSpeech & Seed-TTS Zero-Shot Audio Benchmark", "metric": "WER / MOS / RTF / Latency", "score": "RTF of 0.195" } ]	[ { "paper_id": "2605.27258v2", "title": "PilotTTS: A Disciplined Modular Recipe for Competitive Speech Synthesis", "cosine_sim": 0.7478 }, { "paper_id": "2608.15690v1", "title": "Adding Voice Cloning to Text-to-Audio-Video Models with a Single Zero-Initialised Layer", "cosine_sim": 0.7315...	git clone https://github.com/fishaudio/fish-speech && cd fish-speech && (pip install -e . \|\| pip install -r requirements.txt)	We introduce Fish Audio S2, an open-sourced text-to-speech system featuring multi-speaker, multi-turn generation, and, most importantly, instruction-following control via natural-language descriptions.	To scale training, we develop a multi-stage training recipe together with a staged data pipeline covering video captioning and speech captioning, voice-quality assessment, and reward modeling.	We highly encourage readers to visit https://fish.audio to try custom voices.	4	Neural Audio Codecs & Discrete Acoustic Tokenizers	Explosive (>50/mo)	126	2026-08-22T14:26:54.244618
2601.02914v1	Vulnerabilities of Audio-Based Biometric Authentication Systems Against Deepfake Speech Synthesis	2026-01-06T10:55:32Z	[ "cs.SD", "cs.CR" ]	ArXiv Standard	http://arxiv.org/licenses/nonexclusive-distrib/1.0/	true	60	Internal R&D Only (Copyleft or Academic Terms)	ArXiv Standard Distribution; Repository License Unspecified	1	Proposes This paper presents a systematic empirical evaluation of state-of-the-art speaker authenti... to enhance real-time voice and audio generation, achieving These findings call for a reconsideration of security measures and stress the ne....	Mengze Hong	6	[ "Mengze Hong", "Di Jiang", "Zeying Xie", "Weiwei Zhao", "Guan Wang", "Chen Jason Zhang" ]	[ "Speech & Audio AI Intelligence Lab" ]	http://arxiv.org/abs/2601.02914v1	VERIFIED_LIVE	https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI	[ "https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI" ]	37,800	0	2026-08-22	Unspecified	122.43	As audio deepfakes transition from research artifacts to widely available commercial tools, robust biometric authentication faces pressing security threats in high-stakes industries. This paper presents a systematic empirical evaluation of state-of-the-art speaker authentication systems based on a large-scale speech sy...	[ -0.10196900367736816, -0.01413199957460165, -0.004143999889492989, -0.08028800040483475, -0.053697001188993454, 0.027473999187350273, -0.046946000307798386, -0.047839999198913574, -0.05033399909734726, -0.020066000521183014, -0.01634499989449978, -0.04318000003695488, -0.015768999233841896, ...	[ -0.13398900628089905, -0.02685599960386753, -0.01634800061583519, -0.06453599780797958, 0.02367500029504299, 0.021577000617980957, -0.010383999906480312, -0.0654899999499321, 0.007387999910861254, -0.053718000650405884, -0.03883200138807297, -0.023300999775528908, 0.028946999460458755, -0....	Audio AI, Speech Foundation Models & Real-Time Voice Agents	Low-Latency Zero-Shot TTS & Voice Cloning	Interactive Low-Latency Voice Synthesis (<300ms)	[ "24 kHz High-Quality Speech", "16 kHz Wideband ASR Standard" ]	Neural Continuous & Discrete Acoustic Transformer	[ "LibriSpeech & Seed-TTS Zero-Shot Audio Benchmark" ]	[ "Sub-200ms Latency & High-Fidelity SOTA Acoustic Metric" ]	[ { "benchmark_name": "LibriSpeech & Seed-TTS Zero-Shot Audio Benchmark", "metric": "WER / MOS / RTF / Latency", "score": "Sub-200ms Latency & High-Fidelity SOTA Acoustic Metric" } ]	[ { "paper_id": "2607.14753v1", "title": "Large Audio Language Models for Spoofing-Aware Speaker Verification", "cosine_sim": 0.7899 }, { "paper_id": "2603.14767v1", "title": "Investigating the Impact of Speech Enhancement on Audio Deepfake Detection in Noisy Environments", "cosine_sim": 0...	git clone https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI && cd Retrieval-based-Voice-Conversion-WebUI && (pip install -e . \|\| pip install -r requirements.txt)	As audio deepfakes transition from research artifacts to widely available commercial tools, robust biometric authentication faces pressing security threats in high-stakes industries.	This paper presents a systematic empirical evaluation of state-of-the-art speaker authentication systems based on a large-scale speech synthesis dataset, revealing two major security vulnerabilities: 1) modern voice cloning models trained on very small samples can easily bypass commercial speaker verification systems; ...	These findings call for a reconsideration of security measures and stress the need for architectural innovations, adaptive defenses, and the transition towards multi-factor authentication.	2	Low-Latency Zero-Shot TTS & Instant Voice Cloning	Explosive (>50/mo)	126	2026-08-22T14:27:32.854294
2509.24650v1	VoxCPM: Tokenizer-Free TTS for Context-Aware Speech Generation and True-to-Life Voice Cloning	2025-09-29T12:00:24Z	[ "cs.SD" ]	ArXiv Standard	http://arxiv.org/licenses/nonexclusive-distrib/1.0/	true	60	Internal R&D Only (Copyleft or Academic Terms)	ArXiv Standard Distribution; Repository License Unspecified	1	Proposes This challenge has driven the field towards multi-stage pipelines that rely on pre-trained... to enhance real-time voice and audio generation, achieving Trained on a massive 1.8 million hours of bilingual corpus, our VoxCPM-0.5B mode....	Yixuan Zhou	12	[ "Yixuan Zhou", "Guoyang Zeng", "Xin Liu", "Xiang Li", "Renjie Yu", "Ziyang Wang", "Runchuan Ye", "Weiyue Sun", "Jiancheng Gui", "Kehan Li", "Zhiyong Wu", "Zhiyuan Liu" ]	[ "Speech & Audio AI Intelligence Lab" ]	http://arxiv.org/abs/2509.24650v1	VERIFIED_LIVE	https://github.com/OpenBMB/VoxCPM	[ "https://github.com/OpenBMB/VoxCPM" ]	36,000	0	2026-08-22	Unspecified	118.05	Generative models for speech synthesis face a fundamental trade-off: discrete tokens ensure stability but sacrifice expressivity, while continuous signals retain acoustic richness but suffer from error accumulation due to task entanglement. This challenge has driven the field towards multi-stage pipelines that rely on ...	[ -0.04521999880671501, -0.07266999781131744, 0.033601999282836914, -0.02166599966585636, 0.026386000216007233, -0.01432500034570694, 0.020512999966740608, -0.049573998898267746, 0.013837000355124474, -0.06845399737358093, -0.01170399971306324, -0.07495900243520737, -0.01296399999409914, -0....	[ -0.029392000287771225, -0.10114499926567078, 0.04529900103807449, 0.01630299910902977, 0.038086000829935074, -0.015867000445723534, -0.06587900221347809, -0.0947050005197525, 0.001914000022225082, -0.08585599809885025, -0.035705000162124634, -0.11659300327301025, -0.022525999695062637, -0....	Audio AI, Speech Foundation Models & Real-Time Voice Agents	Low-Latency Zero-Shot TTS & Voice Cloning	Interactive Low-Latency Voice Synthesis (<300ms)	[ "24 kHz High-Quality Speech", "16 kHz Wideband ASR Standard" ]	Neural Continuous & Discrete Acoustic Transformer	[ "LibriSpeech & Seed-TTS Zero-Shot Audio Benchmark" ]	[ "Sub-200ms Latency & High-Fidelity SOTA Acoustic Metric" ]	[ { "benchmark_name": "LibriSpeech & Seed-TTS Zero-Shot Audio Benchmark", "metric": "WER / MOS / RTF / Latency", "score": "Sub-200ms Latency & High-Fidelity SOTA Acoustic Metric" } ]	[ { "paper_id": "2510.16841v2", "title": "SAC: Neural Speech Codec with Semantic-Acoustic Dual-Stream Quantization", "cosine_sim": 0.7952 }, { "paper_id": "2608.11737v1", "title": "Phoenix TTS: High-Fidelity Synthesis and Voice Conversion via Flow-Matching-Driven Speech Tokenization", "cos...	git clone https://github.com/OpenBMB/VoxCPM && cd VoxCPM && (pip install -e . \|\| pip install -r requirements.txt)	Generative models for speech synthesis face a fundamental trade-off: discrete tokens ensure stability but sacrifice expressivity, while continuous signals retain acoustic richness but suffer from error accumulation due to task entanglement.	This challenge has driven the field towards multi-stage pipelines that rely on pre-trained speech tokenizers, but these create a semantic-acoustic divide, limiting holistic and expressive speech generation.	Trained on a massive 1.8 million hours of bilingual corpus, our VoxCPM-0.5B model achieves state-of-the-art zero-shot TTS performance among open-source systems, demonstrating that our approach delivers expressive and stable synthesis.	3	End-to-End Multilingual Speech Recognition (ASR & Whisper)	Explosive (>50/mo)	126	2026-08-22T14:28:13.198833
2603.05413v2	Building Enterprise Realtime Voice Agents from Scratch: A Technical Tutorial	2026-03-05T17:35:59Z	[ "cs.SD" ]	ArXiv Standard	http://arxiv.org/licenses/nonexclusive-distrib/1.0/	true	60	Internal R&D Only (Copyleft or Academic Terms)	ArXiv Standard Distribution; Repository License Unspecified	1	Proposes While end-to-end speech-to-speech models may ultimately provide the best latency for voice... to enhance real-time voice and audio generation, achieving We evaluate the closest candidate, Qwen3-Omni, across three configurations: its....	Jielin Qiu	14	[ "Jielin Qiu", "Zixiang Chen", "Liangwei Yang", "Ming Zhu", "Zhiwei Liu", "Juntao Tan", "Wenting Zhao", "Rithesh Murthy", "Roshan Ram", "Akshara Prabhakar", "Shelby Heinecke", "Caiming Xiong", "Silvio Savarese", "Huan Wang" ]	[ "Speech & Audio AI Intelligence Lab" ]	http://arxiv.org/abs/2603.05413v2	VERIFIED_LIVE	https://github.com/livekit/agents	[ "https://github.com/livekit/agents" ]	13,100	0	2026-08-22	Unspecified	115.55	We present a technical tutorial for building enterprise-grade realtime voice agents from first principles. While end-to-end speech-to-speech models may ultimately provide the best latency for voice agents, fully self-hosted end-to-end solutions are not yet available. We evaluate the closest candidate, Qwen3-Omni, acros...	[ -0.10271400213241577, -0.10067600011825562, -0.08707399666309357, -0.06406699866056442, -0.07019899785518646, 0.0012550000101327896, 0.014274000190198421, -0.027041999623179436, -0.024468999356031418, -0.04393300041556358, -0.052535999566316605, -0.06365100294351578, -0.01690099947154522, ...	[ -0.06984300166368484, -0.06199999898672104, -0.04630199819803238, -0.07402399927377701, -0.0446930006146431, -0.06111599877476692, -0.05368499830365181, -0.04140700027346611, 0.039167001843452454, -0.010532000102102757, -0.013073000125586987, -0.07447899878025055, -0.06307899951934814, -0....	Audio AI, Speech Foundation Models & Real-Time Voice Agents	Full-Duplex Speech-to-Speech LLM & Voice Agent	Interactive Low-Latency Voice Synthesis (<300ms)	[ "24 kHz High-Quality Speech", "16 kHz Wideband ASR Standard" ]	Neural Continuous & Discrete Acoustic Transformer	[ "LibriSpeech & Seed-TTS Zero-Shot Audio Benchmark" ]	[ "Sub-200ms Latency & High-Fidelity SOTA Acoustic Metric" ]	[ { "benchmark_name": "LibriSpeech & Seed-TTS Zero-Shot Audio Benchmark", "metric": "WER / MOS / RTF / Latency", "score": "Sub-200ms Latency & High-Fidelity SOTA Acoustic Metric" } ]	[ { "paper_id": "2601.17097v1", "title": "Sink or SWIM: Tackling Real-Time ASR at Scale", "cosine_sim": 0.7298 }, { "paper_id": "2608.07631v1", "title": "PACE: A Playback-Aligned Context Engine for LLM-Based Full-Duplex Voice Dialogue", "cosine_sim": 0.6984 }, { "paper_id": "2604.0...	git clone https://github.com/livekit/agents && cd agents && (pip install -e . \|\| pip install -r requirements.txt)	We present a technical tutorial for building enterprise-grade realtime voice agents from first principles.	While end-to-end speech-to-speech models may ultimately provide the best latency for voice agents, fully self-hosted end-to-end solutions are not yet available.	We evaluate the closest candidate, Qwen3-Omni, across three configurations: its cloud-only DashScope Realtime API achieves $\sim$702ms audio-to-audio latency with streaming, but is not self-hostable; its local vLLM deployment supports only the Thinker (text generation from audio, 516ms), not the Talker (audio synthesis...	5	Generative Music, Audio Diffusion & Sound Synthesis	Explosive (>50/mo)	126	2026-08-22T14:26:57.397327
2601.15621v1	Qwen3-TTS Technical Report	2026-01-22T03:51:43Z	[ "cs.SD", "cs.CL", "eess.AS" ]	ArXiv Standard	http://arxiv.org/licenses/nonexclusive-distrib/1.0/	true	60	Internal R&D Only (Copyleft or Academic Terms)	ArXiv Standard Distribution; Repository License Unspecified	1	Proposes Qwen3-TTS supports state-of-the-art 3-second voice cloning and description-based control,... to enhance real-time voice and audio generation, achieving 2) Qwen-TTS-Tokenizer-12Hz achieves extreme bitrate reduction and ultra-low-late....	Hangrui Hu	16	[ "Hangrui Hu", "Xinfa Zhu", "Ting He", "Dake Guo", "Bin Zhang", "Xiong Wang", "Zhifang Guo", "Ziyue Jiang", "Hongkun Hao", "Zishan Guo", "Xinyu Zhang", "Pei Zhang", "Baosong Yang", "Jin Xu", "Jingren Zhou", "Junyang Lin" ]	[ "Speech & Audio AI Intelligence Lab" ]	http://arxiv.org/abs/2601.15621v1	VERIFIED_LIVE	https://github.com/QwenLM/Qwen3-TTS	[ "https://github.com/QwenLM/Qwen3-TTS" ]	13,100	0	2026-08-22	Unspecified	113.87	In this report, we present the Qwen3-TTS series, a family of advanced multilingual, controllable, robust, and streaming text-to-speech models. Qwen3-TTS supports state-of-the-art 3-second voice cloning and description-based control, allowing both the creation of entirely novel voices and fine-grained manipulation over ...	[ -0.0957610011100769, 0.02101000025868416, 0.00432500010356307, -0.06101600080728531, -0.02369599975645542, 0.0022319999989122152, 0.009046000428497791, 0.06696400046348572, 0.016840999945998192, 0.049001000821590424, 0.059431999921798706, -0.02329999953508377, 0.02261500060558319, 0.046911...	[ -0.12352000176906586, -0.05215800181031227, 0.034439001232385635, -0.08145000040531158, -0.018938999623060226, -0.0044929999858140945, -0.0311489999294281, -0.05580199882388115, 0.05460200086236, -0.03421400114893913, -0.025567999109625816, -0.07406199723482132, 0.00037200000951997936, -0....	Audio AI, Speech Foundation Models & Real-Time Voice Agents	Full-Duplex Speech-to-Speech LLM & Voice Agent	Real-Time Ultra-Low Latency (<200ms Streaming)	[ "24 kHz High-Quality Speech", "16 kHz Wideband ASR Standard" ]	Moshi / PersonaVoice Full-Duplex Backbone	[ "LibriSpeech & Seed-TTS Zero-Shot Audio Benchmark" ]	[ "Sub-200ms Latency & High-Fidelity SOTA Acoustic Metric" ]	[ { "benchmark_name": "LibriSpeech & Seed-TTS Zero-Shot Audio Benchmark", "metric": "WER / MOS / RTF / Latency", "score": "Sub-200ms Latency & High-Fidelity SOTA Acoustic Metric" } ]	[ { "paper_id": "2607.23938v1", "title": "Qwen-Audio-3.0-TTS: Freely Controllable and Highly Robust Speech Synthesis with Multi-Stage Training Paradigm", "cosine_sim": 0.7704 }, { "paper_id": "2511.12347v1", "title": "VoiceCraft-X: Unifying Multilingual, Voice-Cloning Speech Synthesis and Spee...	git clone https://github.com/QwenLM/Qwen3-TTS && cd Qwen3-TTS && (pip install -e . \|\| pip install -r requirements.txt)	In this report, we present the Qwen3-TTS series, a family of advanced multilingual, controllable, robust, and streaming text-to-speech models.	Qwen3-TTS supports state-of-the-art 3-second voice cloning and description-based control, allowing both the creation of entirely novel voices and fine-grained manipulation over the output speech.	2) Qwen-TTS-Tokenizer-12Hz achieves extreme bitrate reduction and ultra-low-latency streaming, enabling immediate first-packet emission ($97\,\mathrm{ms}$) through its 12.5 Hz, 16-layer multi-codebook design and a lightweight causal ConvNet.	5	Generative Music, Audio Diffusion & Sound Synthesis	Explosive (>50/mo)	126	2026-08-22T14:27:24.249122
2604.00688v3	OmniVoice: Towards Omnilingual Zero-Shot Text-to-Speech with Diffusion Language Models	2026-04-01T09:45:51Z	[ "cs.CL", "eess.AS" ]	ArXiv Standard	http://arxiv.org/licenses/nonexclusive-distrib/1.0/	true	60	Internal R&D Only (Copyleft or Academic Terms)	ArXiv Standard Distribution; Repository License Unspecified	1	"Proposes At its core is a novel diffusion language model-style discrete non-autoregressive (NAR) ar(...TRUNCATED)	Han Zhu	10	["Han Zhu","Lingxuan Ye","Wei Kang","Zengwei Yao","Liyong Guo","Fangjun Kuang","Zhifeng Han","Weiji (...TRUNCATED)	[ "Speech & Audio AI Intelligence Lab" ]	http://arxiv.org/abs/2604.00688v3	VERIFIED_LIVE	https://github.com/k2-fsa/OmniVoice	[ "https://github.com/k2-fsa/OmniVoice" ]	9,300	0	2026-08-22	Unspecified	113.65	"We present OmniVoice, a massively multilingual zero-shot text-to-speech (TTS) model that scales to (...TRUNCATED)	[-0.012010999955236912,0.010067000053822994,0.02797500044107437,0.026737000793218613,0.0449660010635(...TRUNCATED)	[-0.05519299954175949,-0.0312579981982708,0.04747699946165085,0.021883999928832054,0.006374999880790(...TRUNCATED)	Audio AI, Speech Foundation Models & Real-Time Voice Agents	Low-Latency Zero-Shot TTS & Voice Cloning	Interactive Low-Latency Voice Synthesis (<300ms)	[ "24 kHz High-Quality Speech", "16 kHz Wideband ASR Standard" ]	Neural Continuous & Discrete Acoustic Transformer	[ "LibriSpeech & Seed-TTS Zero-Shot Audio Benchmark" ]	[ "Sub-200ms Latency & High-Fidelity SOTA Acoustic Metric" ]	[{"benchmark_name":"LibriSpeech & Seed-TTS Zero-Shot Audio Benchmark","metric":"WER / MOS / RTF / La(...TRUNCATED)	[{"paper_id":"2603.26364v1","title":"LLaDA-TTS: Unifying Speech Synthesis and Zero-Shot Editing via (...TRUNCATED)	"git clone https://github.com/k2-fsa/OmniVoice && cd OmniVoice && (pip install -e . \|\| pip install -(...TRUNCATED)	"We present OmniVoice, a massively multilingual zero-shot text-to-speech (TTS) model that scales to (...TRUNCATED)	"At its core is a novel diffusion language model-style discrete non-autoregressive (NAR) architectur(...TRUNCATED)	"By leveraging a 581k-hour multilingual dataset curated entirely from open-source data, OmniVoice ac(...TRUNCATED)	6	Audio-Visual Speech Synthesis, Talking Heads & Lip-Sync	Explosive (>50/mo)	126	2026-08-22T14:26:37.501966

End of preview. Expand in Data Studio

🎙️ Audio, Speech Foundation Models & Real-Time Voice Agents Dataset (2026 Edition)

A structured research dataset featuring 1,722 domain-verified research papers and 298 official code repositories focused on Full-Duplex Speech-to-Speech LLMs, Real-Time Voice Agents (<200ms Latency), Zero-Shot TTS, Voice Cloning, OpenAI Whisper-v3, Neural Audio Codecs (EnCodec/DAC/SNAC), and Generative Music (2023–2026).

Built with Universal Scientific Engine V18.1 Diamond, providing 48 schema attributes with verified repository attribution, 8 AI topological semantic clusters, pre-calculated Top-3 Semantic Nearest Neighbors Graph, structured benchmark leaderboards, and native 384-dimensional dense PyTorch embeddings.

📊 Dataset Schema Highlights (48 Columns)

Field	Type	Description
`paper_id`	String	Unique ArXiv identifier
`title`	String	Research paper title
`cluster_topic_name`	String	1 of 8 AI Topological Semantic Clusters
`audio_speech_task_paradigm`	String	Task paradigm (Full-Duplex Voice LLM, Zero-Shot TTS, ASR, Neural Codec)
`speech_latency_profile`	String	Latency profile (<200ms Streaming, Streaming Chunk ASR, <500ms Duplex)
`acoustic_sampling_rates`	List[String]	Supported sampling rates (16 kHz, 24 kHz, 44.1 kHz, 48 kHz Hi-Fi)
`speech_foundation_backbone`	String	Model backbone (Whisper-v3, Moshi, CosyVoice, ChatTTS, F5-TTS, MusicGen)
`tested_benchmarks`	List[String]	Evaluated benchmarks (LibriSpeech, CommonVoice, Seed-TTS, AudioCaps)
`benchmark_leaderboard_json`	List[Struct]	Structured WER, MOS, RTF, and Latency scores
`semantic_nearest_neighbors_top3`	List[Struct]	Pre-calculated top-3 related papers with cosine similarity
`commercial_ip_safety_score`	Integer	0–100 commercial compliance index (95% Enterprise Safe)
`tldr_neural_summary`	String	15-word executive summary of key innovation
`title_vector_384d`	List[Float]	384d PyTorch embedding (`all-MiniLM-L6-v2`)
`abstract_vector_384d`	List[Float]	384d dense contextual PyTorch embedding
`reproduction_recipe`	String	1-line bash setup command

🧩 8 AI Semantic Clusters Breakdown

Neural Audio Codecs & Discrete Acoustic Tokenizers (375 papers)
End-to-End Multilingual Speech Recognition (ASR & Whisper) (254 papers)
Speech Enhancement, Denoising & Source Separation (247 papers)
Audio-Visual Speech Synthesis, Talking Heads & Lip-Sync (233 papers)
Generative Music, Audio Diffusion & Sound Synthesis (198 papers)
Low-Latency Zero-Shot TTS & Instant Voice Cloning (191 papers)
Spoken Dialogue Evaluation & Acoustic Benchmarking (137 papers)
Full-Duplex Voice Agents & Speech-to-Speech LLMs (87 papers)

📊 Interactive OpenAngels Visual Dashboard Included

Open DATASET_ANALYTICS_DASHBOARD_100_SAMPLE.html directly in your browser (Chrome/Edge/Safari) to explore the interactive visual intelligence directory with real-time filtering, search, and audio latency metrics.

💻 1-Click Python Quickstart

import pyarrow.parquet as pq

# Load 100-Sample Teaser
table = pq.read_table("AUDIO_SPEECH_FOUNDATION_MODELS_REALTIME_VOICE_AGENTS_2026_100_SAMPLE.parquet")
df = table.to_pandas()

print(f"Loaded {len(df)} sample Audio & Speech AI papers.")
print(f"Top Paper: {df['title'].iloc[0]}")
print(f"Task Paradigm: {df['audio_speech_task_paradigm'].iloc[0]}")
print(f"Latency Profile: {df['speech_latency_profile'].iloc[0]}")
print(f"Top-3 Nearest Neighbors: {df['semantic_nearest_neighbors_top3'].iloc[0]}")

🚀 Get the Full 1,722-Paper Enterprise Edition

The complete commercial production dataset (1,722 papers in Parquet with 384d vectors, SQLite DB, Clean CSV, Interactive OpenAngels HTML Dashboard, and JSON) is available here:

👉 BeatsProm Audio, Speech & Real-Time Voice Agents Dataset Full Edition

Downloads last month: 50