AI BriefingSalesforce releases SOV-Transcribe-1, an open-weight speech recognition model
·2026.11.14 13:06
Key point
The 26.6B parameter model supports seven languages and offers natural-language control over transcription style and context biasing.
Details
- Salesforce released SOV-Transcribe-1, an open-weight speech-augmented language model for non-streaming automatic speech recognition (ASR).
- The model ranks #1 among open-weight models for non-streaming speech-to-text on Artificial Analysis and #2 of 23 systems on VoiceCodeBench for entity accuracy and task success.
- Architecture uses a Whisper large-v3 audio encoder and Gemma 4 26B A4B MoE with 26.6B total parameters (4.6B active) in BF16 precision.
- Supports English, French, German, Spanish, Italian, Portuguese, and Dutch with up to 30 minutes of audio input.
- Features include natural-language control for verbatim or cleaned-up transcripts, context biasing for domain-specific vocabulary, and robustness to accents and background noise.
- Available under Apache 2.0 license with inference support for vLLM and Hugging Face Transformers, plus a DFlash draft model for speculative decoding.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
GitHub Next releases Ace, a local collaborative coding environment for humans and AI agents