Google Launches Gemini 3.5 Speech-to-Text Model with Emotion Detection

Google has unveiled its Gemini 3.5 speech-to-text model, marking a significant advancement in its artificial intelligence capabilities. This model, detailed in a dedicated model card, supports various features beyond simple transcription, including timestamps, speaker identification, translation, summarization, and notably, emotion detection.

The Gemini 3.5 model can process extended audio sessions of up to 96,000 tokens, making it suitable for lengthy recordings such as business meetings and legal depositions. Users can upload common audio formats like MP3 through the Gemini API or the Gemini macOS application. The model enhances the user experience by cleaning up filler words and producing polished text directly from natural speech.

Emotion detection is a key differentiator for this release, allowing for a more nuanced understanding of spoken content. For example, a summary that notes a customer's frustration during a call provides deeper insights than a simple transcript. This capability is particularly beneficial for professionals generating significant volumes of spoken content, including journalists, lawyers, and academics.

Google's strategic move with Gemini 3.5 comes amid fierce competition in the AI landscape, particularly against companies like OpenAI and Anthropic. The release aims to strengthen Google's position in the AI model rankings as it continues to innovate and expand its AI portfolio.

As the market watches for performance benchmarks and updates from competitors, Gemini 3.5's success will be closely monitored, especially in the context of its standing on the Chatbot Arena LLM Leaderboard.