Describe Rename Sound Files (SoundTag AI) is a skill that addresses the problem of cryptic, auto-generated audio filenames by analyzing actual audio content and renaming files with human-readable, descriptive titles following a consistent Title_Case_With_Underscores convention. For example, a filename like audio_track_v2_final_99.wav becomes Bright_Trumpet_Fanfare.wav, and an ElevenLabs-generated file with a timestamp-heavy name becomes Elevenlabs_George_Voice_Speech.mp3.
The skill operates through a two-step workflow. It first runs a local classification using the MIT Audio Spectrogram Transformer (AST), which categorizes sounds into 527 classes — covering Speech, Music, Explosion, and more — entirely offline, avoiding API costs and preserving privacy. For sounds where local classification produces ambiguous results, it makes an improvement pass using the Google Gemini API to generate nuanced descriptions suited to cinematic sound effects, moods, and complex audio textures.
The skill also inspects hints embedded in the original filename so that contextual information already present is not discarded during renaming. Batch processing is supported across .wav, .mp3, .ogg, .flac, .aac, and .m4a formats. Dependency constraints for Torch and Transformers are handled automatically to operate within sandboxed resource limits. The end result is a consistently named, searchable directory of sound files ready for use in sample libraries or field recording archives.