docs: v2 files, host mean normalization, AMI benchmark and Changelog (FluidAudio #985)

Card for b7be41fc: FluidAudio PR https://github.com/FluidInference/FluidAudio/pull/1099; removes the incorrect 'CAM++ normalizes the fbank internally'.

README.md

@@ -18,20 +18,34 @@ on-device speaker verification / diarization on Apple Silicon. Upstream:
 
 | File | Precision | Compute unit | Role |
 |------|-----------|--------------|------|
-| `CamPlusPreprocessor.mlmodelc` | FP32 | CPU | waveform → 80-d fbank features |
-| `CamPlusPlus.mlmodelc` | FP16 | ANE | fbank → 192-d speaker embedding |
+| `CamPlusPreprocessor_v2.mlmodelc` | FP32 | CPU | waveform → 80-d Kaldi fbank (Povey window, log floor after the mel filterbank) |
+| `CamPlusPlus_v2.mlmodelc` | FP32 | CPU / GPU | mean-normalized fbank → 192-d speaker embedding |
+| `CamPlusPreprocessor.mlmodelc` | FP32 | CPU | v1, superseded by `_v2` (Hamming window); kept for older FluidAudio releases |
+| `CamPlusPlus.mlmodelc` | FP16 | CPU / GPU | v1, superseded by `_v2` (see Changelog); kept for older FluidAudio releases |
 
 ## Pipeline
 
 ```
-waveform → [Preprocessor fp32/CPU] → fbank [1,T,80]
-        → [CAM++ fp16/ANE] → embedding [1,192]  (L2-normalize, then cosine for verification/clustering)
+waveform → [CamPlusPreprocessor_v2 fp32/CPU] → fbank [1,T,80]
+        → subtract the per-utterance mean (host)
+        → [CamPlusPlus_v2 fp32/CPU+GPU] → embedding [1,192]  (L2-normalize, then cosine for verification/clustering)
 ```
 
-CAM++ normalizes the fbank internally. The 192-d embedding is used with cosine
-similarity for speaker verification and diarization clustering.
+Neither model normalizes the fbank: the host subtracts each mel band's mean over the utterance,
+as 3D-Speaker and FunASR do (FluidAudio's `CampPlusEmbedder`). The 192-d embedding is used with
+cosine similarity for speaker verification and diarization clustering.
 
-## Benchmark — AISHELL-1 speaker verification
+## Benchmark — AMI speaker verification (v2)
+
+413 single-speaker AMI Mix-Headset segments (3–8 s, 16 participants, 3–4 meetings each), cosine scoring:
+
+| | v1 + host mean | v2 | PyTorch 3D-Speaker CAMPPlus |
+|---|---|---|---|
+| EER, within meeting (1,332 target trials) | 3.07% | 2.85% | 2.85% |
+| EER, across meetings (4,079 target trials) | 3.48% | 3.55% | 3.55% |
+| Cosine to PyTorch, 357 lengths 0.35–60 s | as low as −0.01 | 1.0000 | 1 |
+
+## Benchmark — AISHELL-1 speaker verification (v1)
 
 | Metric | Value |
 |--------|-------|
@@ -41,6 +55,12 @@ similarity for speaker verification and diarization clustering.
 
 AISHELL-1 (clean read Mandarin) is easier than the official CN-Celeb (~6-7%). CoreML↔torch embedding cosine 0.9997-0.99999.
 
+## Changelog
+
+| Date | Files | Change | Issue | Recipe | FluidAudio PR | HF commit |
+|---|---|---|---|---|---|---|
+| 2026-11-01 | `CamPlusPreprocessor_v2.mlmodelc`, `CamPlusPlus_v2.mlmodelc` | Kaldi Povey fbank with the log floor after the mel filterbank; each segment-pooling tail averaged over its own frames (v1 divided by 100); fp32 (v1's fp16 pointwise convs gave unrelated embeddings at some lengths, e.g. 17.9–20.4 s) | [FluidAudio #985](https://github.com/FluidInference/FluidAudio/issues/985) | model-lab `models/campplus-coreml-v2` | [#1099](https://github.com/FluidInference/FluidAudio/pull/1099) | [b7be41fc](https://huggingface.co/FluidInference/campplus-coreml/commit/b7be41fcf4bf7697d879ca71c687dc532854e8a2) |
+
 ## License
 
 Weights derive from FunASR's CAM++; upstream license applies. Format conversion only.