Salesforce speech model records 2.3% error rate, ranks No. 1 among open-weight models
Original headline2.3% 오류율 기록한 세일즈포스 음성 모델, 오픈웨이트 1위
Not investment advice. Muhabbit scores how much a story may matter, not which way to trade.
Salesforce released SOV-Transcribe-1, an open-weight English speech transcription model that posted a 2.3% word error rate, ranking seventh among 44 evaluated models and first among open-weight entries. Weights are published on Hugging Face for self-hosted, non-real-time inference. The AA-WER benchmark drew on roughly eight hours of English audio, weighted 50% AA-AgentTalk, 25% VoxPopuli and 25% Earnings22. The result matters because it puts an openly downloadable model at the front of the open-weight field, giving enterprises a self-hostable alternative to proprietary transcription APIs. Next, watch whether Salesforce extends the family to real-time and multilingual transcription and how quickly the open-weight lead holds as rivals publish new checkpoints.
Muhabbit's AI wrote this summary in its own words from TokenPost's report and translated it from Korean. It can get details wrong, so check the source.
Named in this story
SOV-Transcribe-1, Salesforce (CRM), Hugging Face, Artificial Analysis
A linked name opens that asset's sentiment page.