Updated

SOV-Transcribe-1

Salesforce released SOV-Transcribe-1, an open-weight model for non-streaming speech recognition. It supports multilingual transcription, natural-language formatting instructions, and context biasing for domain terms.

By No Code MBA · Facts checked

Provider
Salesforce
Announcement date
Status at last check
Generally available

What changed

Salesforce introduced its first open-weight speech-augmented language model, combining speech recognition with instruction-controlled transcript formatting and context biasing. The company says the Apache 2.0 weights are available for download on Hugging Face.

Access and availability

Download the SOV-Transcribe-1 model weights from Salesforce Sovereign Intelligence's Hugging Face repository. The model is licensed under Apache 2.0.

Official access documentation

What you can build

Evaluate it on representative audio containing names, numbers, and domain vocabulary, and review critical transcripts before feeding them into downstream workflows.

Limitations

This is a non-streaming automatic speech recognition model, not a general-purpose conversational model. Transcription accuracy and benchmark comparisons reported by Salesforce should be independently evaluated for the intended audio and use case.

Sources and release history