Fluently benchmarks streaming speech-to-text on 580K hours of accented English from 1M+ learners
HeyAmit_ · x · 2026-10-28
Fluently, a language-learning app with 1.7M users across 180+ countries and over 580,000 hours of AI-tutor conversations, published a benchmark of streaming speech-to-text systems on spontaneous, accented English recorded by its learners. The finding: STT is far from solved for the people who need it most — accented, non-fluent speakers — even though it works well on clean native audio. Leaderboard and methodology are public, and the team notes its learners also study 14 languages beyond English.
More from Research
- OptIDC: Optimal Transport Unifies Sample, Feature and Embedding Selection for CTR Prediction — _reachsumit · 2026-10-28
- Phantom Correlation: User Activity Alone Fabricates Item Similarity in Recommender Systems — _reachsumit · 2026-10-28
- ACARE: A 4B Reranker Scores Documents Against the Agent's Full Reasoning Context — _reachsumit · 2026-10-28
- Mengdi Wang unveils LabOS: an AI intelligence layer that watches and guides lab experiments — MengdiWang10 · 2026-10-28
- text-to-3d-with-llm-code dataset trends on Hugging Face (10K-100K rows, CC-BY-4.0) — Rapidata · 2026-10-28
- Salesforce launches CRM-Agent-Bench: contract-based benchmark for agents in enterprise environments — salesforceaicloud · 2026-10-28