Artificial Analysis ranks Salesforce's SOV-Transcribe-1 top open-weights speech-to-text model
Article timeline
The articles in this story. Times are ET.
Artificial AnalysisOfficialArtificial Analysis reports SOV-Transcribe-1 transcribes 60x real time on SalesforceAIArtificial Analysis reports that Salesforce's hosted endpoint runs SOV-Transcribe-1 at a Speed Factor of 60x. At that rate, one minute of audio transcribes in about one second.
- Artificial AnalysisOfficialArtificial Analysis finds SOV-Transcribe-1 leads open-weights speech models on two datasets
AIArtificial Analysis reports that Sovereign AI's SOV-Transcribe-1 leads open-weights models on Earnings22 earnings calls with a 4.1% error rate, versus 4.8% for Voxtral Small. It is also slightly ahead on VoxPopuli parliamentary speech at 1.5% versus 1.8%. On AA-AgentTalk, its voice agent conversations dataset, it scores 1.8%, level with Nemotron 3.5 Transcribe.
- Artificial AnalysisOfficialSOV-Transcribe-1 beats larger transcription models on accuracy with 4.6B active parameters
AIArtificial Analysis says SOV-Transcribe-1, with 4.6B active parameters (26.6B total), sits on the open weights frontier of accuracy versus model size. It is more accurate than Voxtral Small, which has 24B active parameters, and more accurate than Nemotron 3.5 Transcribe, which has 4B active parameters.
- Artificial AnalysisOfficialArtificial Analysis ranks SOV-Transcribe-1 top of open-weights speech models
AIArtificial Analysis reports that SOV-Transcribe-1 has a 2.3% word error rate, 0.5 points ahead of the next open-weights models, Mistral's Voxtral Small and Nvidia's Nemotron 3.5 Transcribe, both at 2.8%.
- Artificial AnalysisOfficialArtificial Analysis ranks SOV-Transcribe-1 top open-weights speech-to-text model
AIArtificial Analysis says Salesforce's SOV-Transcribe-1 is the top open-weights speech-to-text model, scoring 2.3% on the AA-WER Index with 4.6B active parameters. It ranks seventh of 44 models overall, including proprietary models.
