Audio Stem Extractor

Assumption: The tool targets music audio source separation to extract stems from stereo tracks. This is intended for music producers, remixers, educators, and sound engineers who need isolated components for remixing, practice, or analysis. The specification below describes the tool's purpose, domain logic, value proposition, audience, and unique differentiators, without UI or marketing fluff.\n\nPurpose & User Intent: The tool solves the problem of obtaining discrete audio stems (vocals, drums, bass, and other instruments) from a mixed track. Users are typically producers refining remixes, educators teaching audio concepts, and artists preparing stems for practice sessions. The tool operates as a backend service or API that accepts audio input, a requested set of stems, and an output format, returning usable separated files and metadata.\n\nCore Logic & Features:\nRequired features: (1) audio_input: a supported file (mp3, wav, flac) up to a defined size, (2) stems_requested: list of stems to extract (e.g., ['vocals','drums','bass','other']), (3) output_format: ZIP archive or per-stem audio files, (4) processing_pipeline: neural source separation with post-processing to reduce artifacts.\nAdvanced features (optional): (a) separation_quality: standard or high fidelity, (b) batch_mode: queue and process multiple tracks, (c) metadata_export: duration, sample_rate, bit_depth, and stem labels.\n\nInputs & Outputs:\nInputs include audio_input (file, required), stems_requested (array, required), output_format (string, required), separation_quality (string, optional), and target_sample_rate (int, optional). Outputs are: stems as WAV/AIFF, a ZIP or directory archive, and a metadata JSON with duration and stem mapping. Validation enforces supported formats, file integrity, and reasonable file sizes.\n\nAlgorithms & Calculations:\nThe tool uses spectrogram-based neural source separation with a trained model to estimate maskings for each target stem. It applies inverse STFT to reconstruct time-domain audio, followed by optional artifact suppression and gain normalization. Outputs include stem-specific audio streams and accompanying metadata. The approach emphasizes deterministic labeling of stems and consistent alignment across stems for downstream remixing.\n\nError & Edge Cases:\nHandles unsupported formats with a precise error code, rejects corrupted files, enforces maximum track length and size, and flags low-confidence stems with a confidence metric. If some stems are unavailable due to signal constraints, the system returns available stems with a warning in metadata.\n\nIndustry/Region & Localization:\nUses standard audio formats (WAV/AIFF) and common sample rates (22.05kHz–96kHz). Metadata uses ISO-based duration representations and UTC timestamps. No locale-specific licensing terms are assumed; users must comply with rights for the source track.\n\nAssumptions & Exclusions:\nAssumes input tracks have lawful rights for processing; the tool does not grant licensing for commercial reuse of stems. It excludes real-time streaming separation and excludes non-audio inputs.