gemini-audio_skill

Overview

This skill guides implementing Google Gemini audio capabilities to transcribe, analyze, summarize, and generate speech from text. It supports long-form audio (up to 9.5 hours per request), multiple input formats, and controllable TTS voice, pace, and tone. The goal is practical, repeatable workflows for processing podcasts, meetings, interviews, and non-speech sounds.

How this skill works

The skill uses the Gemini API to upload or inline audio, then runs model calls for transcription, summarization, speaker separation, and non-speech analysis. For long or repeated inputs it recommends the File API; for small clips it uses inline binary parts. TTS generation returns audio blobs you can save with selectable models and style controls.

When to use it

• Transcribing meetings, interviews, podcasts with timestamps
• Summarizing long recordings into key points or bullet lists
• Identifying speakers and extracting dialogue segments
• Analyzing non-speech audio: music, ambient sounds, sirens, birdsong
• Generating natural-sounding speech from text with voice/style control
• Processing long-form audio up to 9.5 hours in a single request

Best practices

• Store API key in environment or project .env for automatic detection
• Use File API for files >20MB or for reuse; small clips can be inline
• Provide explicit prompts and timestamps for segment analysis (MM:SS)
• Prefer gemini-2.5-flash for most tasks; upgrade to pro for high-complexity needs
• Validate file format, size and implement exponential backoff for rate limits
• Track token usage and prefer summarization to reduce downstream processing costs

Example use cases

• Automate meeting minutes: transcribe, identify speakers, and produce action items
• Podcast production: generate transcripts, show notes, and short episode summaries
• Research interviews: extract quotes, tag topics, and timestamp highlights
• Monitoring: analyze ambient recordings for sound events and classify non-speech audio
• Content creation: convert blog posts or episode notes into narrated audio with custom voice style