SeniorTalk - A Chinese dialogue speech dataset for the very elderly, jointly developed by Zhiyuan and Nankai University.
SeniorTalk is the world's first Chinese-language dialogue voice dataset for the extremely elderly, jointly developed by the Academy of Artificial Intelligence of Technology (AIAI) and the Human Language Technology Lab (HLT Lab) of the School of Computer Science at Nankai University. The dataset contains 202 individuals aged 75 and above...
What is SeniorTalk?
SeniorTalk is the world's first Chinese-language dialogue voice dataset for the very elderly, jointly launched by the Beijing Academy of Artificial Intelligence (BAAI) and the Human Language Technology Laboratory (HLT Lab) of the School of Computer Science at Nankai University. The dataset contains voice data from 202 individuals aged 75 and above, totaling 55.53 hours. Data collection covered 16 provinces and municipalities, encompassing different regional accents. Based on spontaneous pair conversations, topics included retirement, health, and daily life, closely resembling real-life communication scenarios. The dataset features multi-dimensional, detailed annotations, such as speaker information, dialogue transcription, timestamps (sentence-level and word-level), and accent category labels. SeniorTalk provides valuable support for in-depth research on elderly speech signals and optimization of elderly voice interaction systems, promoting the development of related industries such as age-friendly devices, health management, and assistive elderly care robots.
SeniorTalk's main functions
- Speech recognitionTo improve the accuracy of voice recognition for the very elderly, and to help develop a more accurate voice recognition system that makes it easier for the elderly to use voice interaction.
- Speaker verificationSupport research on speaker verification technology to ensure the security and reliability of voice interaction.
- Speaker separationIt provides multi-speaker dialogue data to support speaker separation technology research and help accurately identify the speech of different speakers in complex environments.
- Voice editingIt provides natural dialogue data to support research on speech editing technology and improve speech synthesis and editing results.
- Health monitoring and assisted communication: Analyze the speech characteristics of the very elderly to support research on health monitoring and assistive communication technologies, and provide data support for elderly care and health management.
SeniorTalk's technical principles
- Data collectionBased on spontaneous pair conversations, this method simulates real-life communication scenarios to ensure the naturalness and authenticity of the voice data. Recordings are taken using various smartphones (including Android and Apple devices) to ensure data diversity and applicability. Strict legal and ethical guidelines are followed to ensure the data collection process is legal, secure, and protects participant privacy.
- Data labelingThis includes speaker information (such as age, gender, region, device, etc.), transcripts of the dialogue, timestamps (sentence-level and word-level), and accent category tags. Data accuracy and completeness are ensured through manual annotation and proofreading.
- Data processingBased on the 16kHz sampling rate WAV file format, audio quality is ensured. The dataset is divided into training, validation, and test sets to support the needs of different research tasks.
- Technology ApplicationThis research focuses on improving speech recognition performance through training with advanced models such as Transformer, Conformer, and E-Branchformer. It also employs models like X-vector, ResNet-TDNN, and ECAPA-TDNN for speaker verification and separation. Furthermore, it utilizes methods such as CampNet, EditSpeech, and A3T to investigate speech editing techniques and improve speech synthesis results.
SeniorTalk's project address
- GitHub repository:https://github.com/flageval-baai/SeniorTalk
- HuggingFace model library:https://huggingface.co/datasets/BAAI/SeniorTalk
- arXiv technical paper:https://www.arxiv.org/pdf/2503.16578
SeniorTalk Application Scenarios
- Smart elderly care systemIt enables elderly people to control home appliances and query information based on voice commands, improving their convenience; it also monitors their health status in real time and issues warnings.
- Auxiliary communication equipmentIt helps elderly people with language impairments to express themselves naturally and accurately recognize specific voice commands in multi-person conversations.
- Health Management PlatformIt analyzes voice characteristics to assess health status and provides health consultation and reminder functions through voice interaction.
- Intelligent voice assistant: Optimize the performance of the voice assistant for elderly users, and provide more natural and easy-to-understand voice feedback.
- Research and development of age-friendly productsSupport the development of smart devices suitable for the elderly, and ensure that voice interaction functions are adapted to the usage habits of the elderly.