Salesforce has released SOV-Transcribe-1, an open-weight model designed for automatic speech recognition (ASR). This model excels in English transcription accuracy and offers multilingual support, along with natural language control over transcription output. It is built upon Salesforce's Sovereign Intelligence initiative and integrates a Whisper large-v3 audio encoder with a Gemma 4 26B A4B MoE language model. AI
IMPACT Provides an open-weight option for advanced speech-to-text capabilities, potentially improving transcription accuracy and control for developers.
RANK_REASON Release of an open-weight model with benchmark results. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →