CED AudioSet sound-event classifiers โ CoreML (coremlit)
CoreML conversions of the four official CED (Consistent Ensemble Distillation) AudioSet sound-event taggers, one fp16 mel โ logits graph per size, for coremlit's audio::ced module.
| size | embed_dim / heads | source |
|---|---|---|
ced-tiny |
192 / 3 | mispeech/ced-tiny |
ced-mini |
256 / 4 | mispeech/ced-mini |
ced-small |
384 / 6 | mispeech/ced-small |
ced-base |
768 / 12 | mispeech/ced-base |
Each ced-<size>/ced_<size>.mlmodelc takes a 64-bin log-mel spectrogram and emits pre-sigmoid logits over the 527 AudioSet classes (the caller applies the sigmoid). The log-mel front-end runs in Rust (coremlit's MelExtractor), so the graph starts at the mel, not the waveform.
Provenance
Converted from-official by coremlit/crates/coremlit/conversion/ced/ โ pinned source revisions + per-file SHA-256 in each ced-<size>/MANIFEST.json and coremlit's tests/ced/model_io.rs. fp32 parity vs the unmodified PyTorch pre-sigmoid oracle: cosine 1.0000000, max|ฮlogit| ~1e-5; shipped fp16 โฅ 0.99999. The weight VALUES are unchanged; the stated graph changes are the mel-front-end lift and the dropped final sigmoid.
License & attribution
Apache-2.0. Upstream: Xiaomi Corporation โ CED (Dinkel et al., 2023; arXiv 2308.11957; RicherMans/CED). Retain the Apache-2.0 license and copyright notices, and state changes, when redistributing.