Google LLC and Cohere Inc. have unveiled advanced AI models enhancing audio processing. Google’s Gemini 3.1 Flash Live automates customer service interactions, adapting responses to user emotions like frustration. It supports voice agents for tasks such as product returns and can process overlapping inputs like images for troubleshooting smart devices. Notably, it excelled in the ComplexFuncBench Audio benchmark with a score of 90.8%, indicating a substantial improvement over previous versions.
Cohere Transcribe focuses on high-accuracy transcription with a 5.42% average word error rate, ranking it the best on the Hugging Face Open ASR Leaderboard. Utilizing a Conformer algorithm, it efficiently converts audio to transcripts in over a dozen languages. Both models are available under open-source licenses, enabling companies to utilize them in their infrastructure or via Cohere’s Model Vault.
This launch highlights significant advancements in AI-driven audio processing, offering powerful solutions for businesses across industries.
Source link
