Interspeech 2025

TL;DR

• Interspeech 2025, the premier conference for Automatic Speech Recognition (ASR), was held in Rotterdam, Netherlands during August 17-21
• Argmax had 1 oral presentation at the Speaker Diarization track
• Argmax showed live demos to hundreds of on-device AI enthusiasts at the expo

‍

SDBench

SDBench is our open-source and reproducible benchmark suite for popular speaker diarization systems, commercial and academic. It includes 13 datasets covering various languages and use cases, standardized test splits, and evaluation metric configurations. Here is the link to the PDF.

We recently renamed SDBench to OpenBench to reflect the expanded benchmarking scope. OpenBench now benchmarks real-time transcription and keyword boosting as well. The link to the GitHub repository is here.

‍

‍

‍

Expo

The number of people inquiring about Frontier Models On Device (FMOD) far exceeded our expectations! We also welcomed friends of Argmax like Hugging Face, Nvidia, pyannoteAI and aicoustics as well as many first-timers who were excited to see frontier models like Nvidia Parakeet v3 (released 1 day before the conference) running on-device in real-time.

‍

‍

‍

‍