egc1egc.one / toolsback to sample processing ↗

EGC1

Sample processing help

The sample processor prepares an unorganised sample folder for the EGC1 instrument. It runs entirely in your browser. There is no local helper to install and no audio-processing server.

Processing a folder

Select a folder or a USB stick. Its subfolders are included. Enable file/folder hints to combine names with acoustic classification, or turn hints off to classify from audio alone. Explicit manual labels always take priority. Silence-separated sounds become ordinary samples; hit kits are extracted only from drum loops when you enable that option.

Chrome and Edge on desktop can save directly into the selected folder. Browsers without that folder-write API use a file-picker and ZIP export: extract the ZIP into the same sample folder when processing finishes. The archive contains .egc1, including metadata and any generated regions or kit hits. Keep it with the original audio when copying a library to a USB stick.

Local processing

Your audio, folder names, classifications and review records are not uploaded. The page, fonts, analysis runtime and model files are downloaded. Inference runs locally in a background worker, so the interface and audition controls remain responsive.

The classifier model is approximately 113 MiB; loop timing adds approximately 80 MiB when needed. Downloads are checked against pinned SHA-256 hashes and cached in your browser where storage permits. Once the page and required models are cached, processing can run offline. Clearing browser storage removes those model downloads, but does not remove analysis saved in your sample folder.

Detailed review

Use “review prepared samples” after processing, or choose an existing prepared folder directly on the review page. No classification run is needed. Review instrument attention flags, choose issue checkboxes, inspect waveform zero crossings and audition your marked loop. Review files, marks and explicit flag resolutions use the same format as the local instrument-review tools. They are separate from category overrides.

With direct folder access, reviews save in .egc1. Download mode also saves judgements in this browser; export the metadata ZIP to carry them with your audio. Existing .toysampler metadata is copied into .egc1, preserving labels, reviews and identities. The legacy folder is retained as a backup.

Saved metadata

Only .egc1 is written. Source audio is not renamed, moved or overwritten. Saved analysis uses content hashes so changed audio is processed again. Compatible finished work is reused after a pause or interruption. Existing manual overrides, instrument favourites, attention flags and review answers are preserved. Processing in another tab is blocked while a run or correction is writing.

Use WAV or AIFF for long multi-region sources: their audio is scanned in small blocks. Playable regions must be shorter than 60 seconds, mono/stereo and 8–192 kHz. FLAC and Ogg Vorbis decoding depends on the browser; very large compressed sources should first be converted to WAV. Individual failures appear in the results and saved report; they do not stop other files.

Classification accuracy

The acoustic model compares each sound with fixed descriptions of drum hits, loops, synth roles, voices and effects. Ambiguous results stay broad or uncertain. Filename hints are useful evidence, not verified labels. Model similarity scores are not probabilities. Audition results and use “edit category” when a suggestion is wrong.

The browser compiles the instrument’s existing pitch, transient, vocal-cut, gate and loop-boundary analysis to WebAssembly. Beat This! supplies loop beat/downbeat predictions. Browser preprocessing has its own versioned identity so it cannot silently reuse incompatible native embeddings. Automatic beat grids can still be wrong; loop and kit cuts need listening checks.

credits & source