Sonic Kinetic: An AI-Powered Workout Timer That Yells at You Using Gemini and ElevenLabs
These articles are AI-generated summaries. Please check the original sources for full details.
I made an AI yell my workouts at me (Sonic Kinetic)
Marouane K built Sonic Kinetic, a workout timer that generates unique audio coaching for each routine. It uses Gemini to create interval scripts and ElevenLabs to produce real-time audio clips, with stability settings lowered to 0.30 for maximum-intensity segments.
Why This Matters
Most workout timers rely on static beeps or pre-recorded tracks, offering no personalization or adaptive intensity. Sonic Kinetic demonstrates a fully dynamic pipeline where AI generates both the workout structure and the audio coaching on the fly, with ElevenLabs’ stability knob tuned per segment to create a genuinely aggressive voice for high-effort intervals.
Key Insights
- Gemini 3.1 Flash Lite generates workout timelines with JSON schema enforcement, ensuring structured output without parsing errors (Marouane K, 2026).
- ElevenLabs’ eleven_flash_v2_5 stability setting is dropped to 0.30 for Maximum-intensity segments, creating a noticeably more aggressive voice (Marouane K, 2026).
- Go backend uses goroutines with sync.WaitGroup to parallelize ElevenLabs API calls, so a 12-segment routine doesn’t take 12x longer than a 1-segment one (Marouane K, 2026).
- Gemini 2.5 Flash was deprecated for new API keys, causing a 404 error during development (Marouane K, 2026).
- ElevenLabs free accounts cannot access stock library voices via API, resulting in 402 errors until a custom voice was used (Marouane K, 2026).
Practical Applications
- Fitness apps can use AI-generated audio coaching to create personalized, dynamic workout experiences without pre-recorded content.
- Audio generation pipelines can parallelize API calls using goroutines and sync.WaitGroup to avoid linear latency scaling with segment count.
- Pitfall: Relying on deprecated AI model versions (e.g., Gemini 2.5 Flash) can cause unexpected 404 errors; always check API deprecation status.
- Pitfall: Assuming free-tier API access includes all features (e.g., ElevenLabs library voices) can lead to 402 payment errors during development.
References:
Continue reading
Next article
Mem0 vs TurboMem: Embedded Memory Layer for TypeScript AI Agents - 2026 Comparison
Related Content
JumpLander Launches AI Engineering Ecosystem for Software Development with Coding Agents and Open Datasets
JumpLander introduces a transparent AI engineering platform featuring coding agents, programming datasets like JumpTrace-1K, and Persian developer support.
RAG App Fails Two Basic Questions: Chunking Bug vs Model Capacity Limits
Phase 1 RAG pipeline reveals two distinct failure modes: chunk dilution and small model indecision, with similarity score of 0.46 just below threshold.
Revise: A Spaced-Repetition Learning Tool Using LLM Prompting
Developer Samot released Revise, a free app utilizing spaced-repetition and LLM integration to prevent academic knowledge decay over summer breaks.