feat: add senselab VAD, eGeMAPS, and embedding processors (#566)

docs/api/processors/pyhealth.processors.SenselabEGeMAPSProcessor.rst

@@ -0,0 +1,7 @@
+pyhealth.processors.SenselabEGeMAPSProcessor
+=============================================
+
+.. autoclass:: pyhealth.processors.SenselabEGeMAPSProcessor
+   :members:
+   :undoc-members:
+   :show-inheritance:

docs/api/processors/pyhealth.processors.SenselabEmbeddingProcessor.rst

@@ -0,0 +1,7 @@
+pyhealth.processors.SenselabEmbeddingProcessor
+===============================================
+
+.. autoclass:: pyhealth.processors.SenselabEmbeddingProcessor
+   :members:
+   :undoc-members:
+   :show-inheritance:

docs/api/processors/pyhealth.processors.SenselabVADProcessor.rst

@@ -0,0 +1,7 @@
+pyhealth.processors.SenselabVADProcessor
+=========================================
+
+.. autoclass:: pyhealth.processors.SenselabVADProcessor
+   :members:
+   :undoc-members:
+   :show-inheritance:

examples/senselab_processor_tutorial.py

@@ -0,0 +1,70 @@
+"""
+Senselab Processor Tutorial
+============================
+Demonstrates PyHealth's senselab processor wrappers for voice
+biomarker extraction: VAD, eGeMAPS features, and speaker embeddings.
+
+Use cases include depression detection, Parkinson's monitoring,
+and cognitive assessment through voice.
+
+Requires: pip install senselab
+"""
+import os
+import tempfile
+
+import torch
+import torchaudio
+
+
+def make_test_wav(path: str, sr: int = 16000, duration: float = 3.0) -> None:
+    """Write a synthetic sine-wave audio file for demonstration."""
+    t = torch.linspace(0, duration, int(sr * duration))
+    waveform = torch.sin(2 * torch.pi * 440 * t).unsqueeze(0)
+    torchaudio.save(path, waveform, sr)
+
+
+with tempfile.TemporaryDirectory() as tmpdir:
+    audio_path = os.path.join(tmpdir, "sample.wav")
+    make_test_wav(audio_path)
+
+    # ------------------------------------------------------------------
+    # 1. SenselabVADProcessor
+    #    Strips silence and noise, returns speech-only waveform tensor.
+    # ------------------------------------------------------------------
+    from pyhealth.processors import SenselabVADProcessor
+
+    vad = SenselabVADProcessor(sample_rate=16000)
+    speech = vad.process(audio_path)
+    print(f"VAD output shape:     {speech.shape}")
+    # Expected: torch.Size([1, N]) where N <= original samples
+
+    # ------------------------------------------------------------------
+    # 2. SenselabEGeMAPSProcessor
+    #    Extracts 88 hand-crafted eGeMAPS acoustic features widely used
+    #    in clinical voice research.
+    # ------------------------------------------------------------------
+    from pyhealth.processors import SenselabEGeMAPSProcessor
+
+    egemaps = SenselabEGeMAPSProcessor(sample_rate=16000, apply_vad=True)
+    features = egemaps.process(audio_path)
+    print(f"eGeMAPS output shape: {features.shape}")
+    # Expected: torch.Size([88])
+
+    # ------------------------------------------------------------------
+    # 3. SenselabEmbeddingProcessor
+    #    Extracts deep neural speaker embeddings using WavLM or
+    #    ECAPA-TDNN for voice biomarker tasks.
+    # ------------------------------------------------------------------
+    from pyhealth.processors import SenselabEmbeddingProcessor
+
+    # WavLM embedding (768-dim)
+    wavlm = SenselabEmbeddingProcessor(model="wavlm", sample_rate=16000)
+    embedding = wavlm.process(audio_path)
+    print(f"WavLM embedding shape: {embedding.shape}")
+    # Expected: torch.Size([768])
+
+    # ECAPA-TDNN embedding (192-dim)
+    ecapa = SenselabEmbeddingProcessor(model="ecapa", sample_rate=16000)
+    embedding_ecapa = ecapa.process(audio_path)
+    print(f"ECAPA embedding shape: {embedding_ecapa.shape}")
+    # Expected: torch.Size([192])

pyhealth/processors/__init__.py

@@ -18,6 +18,11 @@ def get_processor(name: str):
 
 
 # Import all processors so they register themselves
+from .senselab_processor import (
+    SenselabVADProcessor,
+    SenselabEGeMAPSProcessor,
+    SenselabEmbeddingProcessor,
+)
 from .image_processor import ImageProcessor
 from .label_processor import (
     BinaryLabelProcessor,
@@ -82,4 +87,7 @@ def get_processor(name: str):
     "TupleTimeTextProcessor",
     "CehrProcessor",
     "ConceptVocab",
+    "SenselabVADProcessor",
+    "SenselabEGeMAPSProcessor",
+    "SenselabEmbeddingProcessor",
 ]