Paul Daniels
About
No profileSessions
Poster Session Google Colab and Gemini for batch speech analysis more
This practical poster session will demonstrate how to configure Google Colab with a Gemini API to efficiently assess speaking assignments in bulk. Participants will learn how to record student pairwork or presentation tasks, upload them to Google Drive, and use AI to transcribe and evaluate the audio. Sample Gemini AI prompts incorporating speech assessment rubrics will be introduced, covering both text-based criteria (e.g., grammar accuracy and content) and speech-based criteria (e.g., intonation, stress, and rhythm).
Presentation Gemini listens: Analyzing speaking tasks more
Generative AI is transforming language teaching and learning in areas such as translation, feedback, and evaluation. This presentation examines AI’s ability to analyze speaking tasks in the language learning classroom. Most generative AI tools, such as ChatGPT, first convert speech to text and then analyze the transcript—an approach that overlooks important prosodic features. However, Google’s Gemini 2.0 can process raw audio directly, capturing intonation, stress, rhythm, and loudness without relying on text-based transcription. This study compared the accuracy and efficiency of human and AI ratings of pair-work speaking tasks, focusing on Gemini 2.0’s multimodal ability to analyze natural prosody and intonation. The findings revealed a moderate positive correlation between human and AI ratings of speaking tasks, indicating that Gemini 2.0 aligns well with human judgments of intonation and rhythm in language learner speech.