Good ideas deserve
to be discovered.
An independent research companion to Interspeech 2026.
Made by Wei-Ren Lan, for people curious about speech.
I’m Wei-Ren Lan.
AI architect, engineer, and consultant based in Taipei.
I have eight years of experience in speech AI, building systems that take research into production—from speech recognition, noise reduction, and voice separation to meeting intelligence and real-time multimodal agents.
As a founding AI engineer at DeepWave Intelligence, I built and led the AI team and helped deliver audio products serving more than two million users. Across my work, I have built and deployed more than 15 AI models.
Hiring for speech AI or applied AI, exploring a consulting project, or looking to collaborate? I’d be glad to connect.
Why I built this wiki.
Every year, I turn to Interspeech to keep up with speech research and find ideas worth exploring. This year, I wanted to make that process easier for both people and AI agents. That became the Interspeech 2026 Wiki and this website.
The idea is simple: people choose the research questions and papers that matter to them; agents help explore the evidence. You can browse by topic or institution, select a research scope, and copy its metadata and summaries into your agent. Every paper also has a Markdown export with its digest and provenance.
A conference is more than a collection of PDFs.
With 1,379 papers to explore, finding a useful starting point can be hard. This wiki brings research into a connected, readable format: the problem, the method, the evidence, the limitations, and where to go next.
Built with Claude Code, TypeSafe AI, and Gemini.
Wei-Ren Lan assembled this open knowledge base with AI-assisted tools. Claude Code, TypeSafe AI, and Gemini helped organize the research into structured metadata and connected digests. Related-paper links are based on pairwise assessments by TypeSafe Jev.
Read with the evidence in view.
Every page identifies its source and summary coverage. Full-paper digest means the summary was compiled from the paper’s full text. Abstract-only means only the abstract supports the summary. AI-generated digests can contain mistakes; use the linked DOI and original paper to check important details.
What “code & resources” means.
Resource links come from the repository’s recorded code field. They may point to author implementations, datasets, models, demo pages, or tools referenced by the paper. A listed link does not by itself establish that the authors released their own implementation. Consult the paper’s Code section and the linked resource for context.
One open source of knowledge.
This website is built directly from the repository’s Markdown wiki pages and YAML metadata. There is no separate content database. You can browse here, clone the repository for your own research, or use a coding agent to ask questions across the wiki.
Help make the next read better.
Spotted an error, a missing affiliation, or a new code release? Authors and readers are welcome to suggest corrections on GitHub. See the contribution guide to get started.
This is an independent community project, not an official ISCA or Interspeech website. Original papers remain with their respective rights holders; this site hosts compiled research digests and links to the source papers.
Explore the research