High-Quality Training Datasets

French Natural Conversational Speech Synthesis Corpus. Recorded by two native voice talents (one male, one female), this corpus comprises two modules: free dialogue on specified topics and scripted emotional/paralinguistic text performances. It covers single-emotion, multi-emotion, and paralinguistic expressions (including a variety of interjections). Professional linguists have provided three-tier annotation—on text content, emotion labels, and paralinguistic phenomena. The total duration is approximately 11 hours, with audio specifications of 48 kHz sampling rate, 24‑bit depth, WAV format, mono channel. This dataset is designed to fully meet the diverse requirements of TTS research and development in terms of naturalness, expressiveness, and precise annotation.

Indonesian Natural Conversational Speech Synthesis Corpus. Recorded by two native voice talents (one male, one female), this corpus comprises two modules: free dialogue on specified topics and scripted emotional/paralinguistic text performances. It covers single-emotion, multi-emotion, and paralinguistic expressions (including a variety of modal particles). Professional linguists have provided three-tier annotation—on text content, emotion labels, and paralinguistic phenomena. The total duration is approximately 11 hours, with audio specifications of 48 kHz sampling rate, 24‑bit depth, WAV format, mono channel. This dataset is designed to fully meet the diverse requirements of TTS research and development in terms of naturalness, expressiveness, and precise annotation.

Saudi Arabian Arabic Natural Conversational Speech Synthesis Corpus. Recorded by two native voice talents (one male, one female), this corpus comprises two modules: free dialogue on specified topics and scripted emotional/paralinguistic text performances. It covers single-emotion, multi-emotion, and paralinguistic expressions (including a variety of interjections). Professional linguists have provided three-tier annotation—on text content, emotion labels, and paralinguistic phenomena. The total duration is approximately 11 hours, with audio specifications of 48 kHz sampling rate, 24‑bit depth, WAV format, mono channel. This dataset is designed to fully meet the diverse requirements of TTS research and development in terms of naturalness, expressiveness, and precise annotation.

Modern Standard Arabic Natural Conversational Speech Synthesis Corpus. Recorded by two native voice talents (one male, one female), this corpus comprises two modules: free dialogue on specified topics and scripted emotional/paralinguistic text performances. It covers single-emotion, multi-emotion, and paralinguistic expressions (including a variety of interjections). Professional linguists have provided three-tier annotation—on text content, emotion labels, and paralinguistic phenomena. The total duration is approximately 11 hours, with audio specifications of 48 kHz sampling rate, 24‑bit depth, WAV format, mono channel. This dataset is designed to fully meet the diverse requirements of TTS research and development in terms of naturalness, expressiveness, and precise annotation.

This dataset contains 1,042 egocentric video segments (approximately 35 seconds each) collected across 34 locations and 6 real-world environments, including homes, offices, and retail scenarios. Powered by self-developed VSLAM system, achieving millimeter-level positioning, multi-sensor hard-triggered synchronization (≤1ms), and a high frame rate of 60fps for RGB. It includes multi-view videos, calibrations, point clouds, SLAM trajectories, and gesture recognition results in standard formats. Designed for embodied AI, spatial perception, and 3D reconstruction, it offers high precision, diverse scenarios, and out-of-the-box usability, making it ideal for training robust perception-action models.

Japanese Natural Conversational Speech Synthesis Corpus. Recorded by two native voice talents (one male, one female), this corpus comprises two modules: free dialogue on specified topics and scripted emotional/paralinguistic text performances. It covers single-emotion, multi-emotion, and paralinguistic expressions (including a variety of modal particles). Professional linguists have provided three-tier annotation—on text content, emotion labels, and paralinguistic phenomena. The total duration is approximately 11 hours, with audio specifications of 48 kHz sampling rate, 24‑bit depth, WAV format, mono channel. This dataset is designed to fully meet the diverse requirements of TTS research and development in terms of naturalness, expressiveness, and precise annotation.

Italian Natural Conversational Speech Synthesis Corpus. Recorded by two native voice talents (one male, one female), this corpus comprises two modules: free dialogue on specified topics and scripted emotional/paralinguistic text performances. It covers single-emotion, multi-emotion, and paralinguistic expressions (including a variety of interjections). Professional linguists have provided three-tier annotation—on text content, emotion labels, and paralinguistic phenomena. The total duration is approximately 11 hours, with audio specifications of 48 kHz sampling rate, 24‑bit depth, WAV format, mono channel. This dataset is designed to fully meet the diverse requirements of TTS research and development in terms of naturalness, expressiveness, and precise annotation.

Brazilian Portuguese Natural Conversational Speech Synthesis Corpus. Recorded by two native voice talents (one male, one female), this corpus comprises two modules: free dialogue on specified topics and scripted emotional/paralinguistic text performances. It covers single-emotion, multi-emotion, and paralinguistic expressions (including a variety of interjections). Professional linguists have provided three-tier annotation—on text content, emotion labels, and paralinguistic phenomena. The total duration is approximately 11 hours, with audio specifications of 48 kHz sampling rate, 24‑bit depth, WAV format, mono channel. This dataset is designed to fully meet the diverse requirements of TTS research and development in terms of naturalness, expressiveness, and precise annotation.

German Natural Conversational Speech Synthesis Corpus. Recorded by two native voice talents (one male, one female), this corpus comprises two modules: free dialogue on specified topics and scripted emotional/paralinguistic text performances. It covers single-emotion, multi-emotion, and paralinguistic expressions (including a variety of interjections). Professional linguists have provided three-tier annotation—on text content, emotion labels, and paralinguistic phenomena. The total duration is approximately 11 hours, with audio specifications of 48 kHz sampling rate, 24‑bit depth, WAV format, mono channel. This dataset is designed to fully meet the diverse requirements of TTS research and development in terms of naturalness, expressiveness, and precise annotation.

Indian English Natural Conversational Speech Synthesis Corpus. Recorded by two native voice talents (one male, one female), this corpus comprises two modules: free dialogue on specified topics and scripted emotional/paralinguistic text performances. It covers single-emotion, multi-emotion, and paralinguistic expressions (including a variety of interjections). Professional linguists have provided three-tier annotation—on text content, emotion labels, and paralinguistic phenomena. The total duration is approximately 11 hours, with audio specifications of 48 kHz sampling rate, 24‑bit depth, WAV format, mono channel. This dataset is designed to fully meet the diverse requirements of TTS research and development in terms of naturalness, expressiveness, and precise annotation.

The INTERSPEECH 2026 MLC-SLM Challenge Dataset, curated by Datatang, is derived from fifteen proprietary conversational speech corpora. Distinguished by exceptional annotation accuracy and operational reliability, this dataset is engineered to address critical challenges in multilingual automatic speech recognition (ASR) and long-context comprehension. It meticulously replicates real-world complexities including spontaneous interruptions and speaker overlaps, thereby providing robust training resources for developing world-ready ASR systems. All data collection and processing strictly comply with international privacy regulations including GDPR, CCPA and PIPL, with rigorous protocols ensuring participant anonymity and ethical data usage throughout the lifecycle.

This dataset covers office-based scenarios such as in-depth searches, data analysis, and industry research, encompassing complete multi-turn reasoning trajectories and tool-calling chains. It is designed to support the analysis of agent planning capabilities, research into tool selection strategies, and quality assessment, providing a structured benchmark for agent training and evaluation.