Publish artifacts for audio-event-triage-20260823

README.md

@@ -0,0 +1,61 @@
+---
+license: mit
+library_name: custom
+pipeline_tag: audio-classification
+datasets:
+- RKB109/audio-event-triage-20260823-dataset
+tags:
+- synthetic-data
+- transparent-baseline
+- audio-ml
+- audio-classification
+- automatic-speech-recognition
+- feature-extraction
+- audio-to-audio
+metrics:
+- accuracy
+---
+
+# Audio Event Triage Baseline Baseline Model
+
+## Model Description
+
+This repository contains a small, transparent prototype model for
+**Operations teams need an explainable starting point for classifying alarms, machinery noise, and speech-like events.**
+
+The model combines per-label token weights with IDF-weighted evidence
+retrieval. It was generated for reproducible architecture demonstrations and
+does not call a hosted LLM.
+
+## Evaluation
+
+- Held-out synthetic examples: 4
+- Accuracy: 1
+- Intended metrics: classification_accuracy, macro_recall, review_coverage
+
+## Intended Use
+
+- Architecture prototyping
+- CI and evaluation examples
+- Local baseline comparisons
+- Educational experimentation
+
+## Hugging Face Task Coverage
+
+- `audio-classification`
+- `automatic-speech-recognition`
+- `feature-extraction`
+- `audio-to-audio`
+
+## Limitations and Risks
+
+The included records are synthetic feature vectors and do not replace evaluation on licensed real audio.
+
+The dataset is synthetic and small. Do not use this model for consequential
+decisions without representative data, expert review, and production-grade
+evaluation.
+
+## Reproducibility
+
+The linked GitHub repository includes `train.py`, the exact dataset split,
+evaluation code, and the model JSON format.

evaluation.json

@@ -0,0 +1,5 @@
+{
+  "test_examples": 4,
+  "accuracy": 1,
+  "synthetic_evaluation": true
+}

inference.py

@@ -0,0 +1,80 @@
+"""Transparent baseline pipeline for the generated AI project."""
+
+from __future__ import annotations
+
+import json
+import math
+import re
+from pathlib import Path
+
+
+def tokenize(value: str) -> list[str]:
+    return re.findall(r"[a-z0-9]+", value.lower())
+
+
+class Pipeline:
+    def __init__(self, model: dict):
+        self.model = model
+
+    @classmethod
+    def from_file(cls, path: str | Path) -> "Pipeline":
+        return cls(json.loads(Path(path).read_text(encoding="utf-8")))
+
+    def classify(self, text: str) -> tuple[str, float]:
+        tokens = tokenize(text)
+        scores = {
+            label: sum(weights.get(token, 0) for token in tokens)
+            for label, weights in self.model["prototypes"].items()
+        }
+        ranked = sorted(scores.items(), key=lambda item: (-item[1], item[0]))
+        label, best = ranked[0]
+        total = sum(max(score, 0) for _, score in ranked) or 1
+        return label, best / total
+
+    def search(self, query: str, limit: int = 3) -> list[dict]:
+        query_tokens = set(tokenize(query))
+        ranked = []
+        for document in self.model["documents"]:
+            document_tokens = set(tokenize(document["text"]))
+            lexical = sum(
+                self.model["idf"].get(token, 1.0)
+                for token in query_tokens & document_tokens
+            )
+            ranked.append({**document, "score": round(lexical, 6)})
+        return sorted(ranked, key=lambda item: (-item["score"], item["id"]))[:limit]
+
+    def graph_evidence(self, text: str) -> list[dict]:
+        tokens = set(tokenize(text))
+        matches = []
+        for subject, relation, target in self.model.get("graph_edges", []):
+            edge_tokens = set(tokenize(f"{subject} {relation} {target}"))
+            overlap = len(tokens & edge_tokens)
+            if overlap:
+                matches.append(
+                    {
+                        "subject": subject,
+                        "relation": relation,
+                        "target": target,
+                        "overlap": overlap,
+                    }
+                )
+        return sorted(matches, key=lambda item: -item["overlap"])
+
+    def run(self, text: str) -> dict:
+        label, confidence = self.classify(text)
+        evidence = self.search(text)
+        result = {
+            "prediction": label,
+            "confidence": round(confidence, 4),
+            "requires_review": confidence < self.model["confidence_threshold"],
+            "evidence": evidence,
+        }
+        if self.model["mode"] == "graph":
+            result["graph_evidence"] = self.graph_evidence(text)
+        if self.model["mode"] == "agent":
+            result["proposed_tool"] = label
+            result["approval_required"] = label in {
+                "request-approval",
+                "request-human-help",
+            }
+        return result

model.json

@@ -0,0 +1,163 @@
+{
+  "format": "daily-project-prototype-v1",
+  "project": "audio-event-triage",
+  "title": "Audio Event Triage Baseline",
+  "domain": "audio-ml",
+  "mode": "classifier",
+  "labels": [
+    "alarm",
+    "machinery",
+    "speech"
+  ],
+  "prototypes": {
+    "alarm": {
+      "2100": 3,
+      "high": 3,
+      "energy": 3,
+      "repeating": 5,
+      "tone": 5,
+      "peak": 3,
+      "frequency": 3,
+      "synthetic": 5,
+      "alarm": 3,
+      "feature": 5,
+      "summary": 5,
+      "in": 2,
+      "an": 3,
+      "operations": 2,
+      "review": 2,
+      "for": 1,
+      "evaluation": 1,
+      "case": 1,
+      "pulse": 2,
+      "bright": 2,
+      "spectrum": 2,
+      "urgent": 2,
+      "alert": 2
+    },
+    "machinery": {
+      "in": 2,
+      "an": 4,
+      "operations": 2,
+      "review": 2,
+      "low": 5,
+      "frequency": 2,
+      "continuous": 2,
+      "vibration": 5,
+      "high": 2,
+      "zero": 2,
+      "crossing": 2,
+      "stability": 2,
+      "synthetic": 5,
+      "rotating": 2,
+      "equipment": 2,
+      "feature": 5,
+      "summary": 5,
+      "for": 2,
+      "evaluation": 2,
+      "case": 2,
+      "steady": 3,
+      "hum": 3,
+      "harmonic": 3,
+      "pitch": 3,
+      "motor": 3
+    },
+    "speech": {
+      "variable": 2,
+      "pitch": 4,
+      "moderate": 2,
+      "energy": 2,
+      "speech": 2,
+      "cadence": 2,
+      "synthetic": 4,
+      "spoken": 2,
+      "segment": 2,
+      "feature": 4,
+      "summary": 4,
+      "for": 2,
+      "an": 3,
+      "evaluation": 2,
+      "case": 2,
+      "in": 1,
+      "operations": 1,
+      "review": 1,
+      "syllabic": 2,
+      "rhythm": 2,
+      "changing": 2,
+      "voice": 4,
+      "activity": 2
+    }
+  },
+  "idf": {
+    "synthetic": 1,
+    "alarm": 2.252763,
+    "feature": 1,
+    "summary": 1,
+    "rotating": 2.252763,
+    "equipment": 2.252763,
+    "spoken": 2.252763,
+    "segment": 2.252763,
+    "alert": 2.252763,
+    "motor": 2.252763,
+    "voice": 2.252763
+  },
+  "documents": [
+    {
+      "id": "audio-01",
+      "label": "alarm",
+      "text": "Synthetic alarm feature summary.",
+      "metadata": {
+        "synthetic": true,
+        "domain": "audio-ml"
+      }
+    },
+    {
+      "id": "audio-02",
+      "label": "machinery",
+      "text": "Synthetic rotating equipment feature summary.",
+      "metadata": {
+        "synthetic": true,
+        "domain": "audio-ml"
+      }
+    },
+    {
+      "id": "audio-03",
+      "label": "speech",
+      "text": "Synthetic spoken segment feature summary.",
+      "metadata": {
+        "synthetic": true,
+        "domain": "audio-ml"
+      }
+    },
+    {
+      "id": "audio-04",
+      "label": "alarm",
+      "text": "Synthetic alert feature summary.",
+      "metadata": {
+        "synthetic": true,
+        "domain": "audio-ml"
+      }
+    },
+    {
+      "id": "audio-05",
+      "label": "machinery",
+      "text": "Synthetic motor feature summary.",
+      "metadata": {
+        "synthetic": true,
+        "domain": "audio-ml"
+      }
+    },
+    {
+      "id": "audio-06",
+      "label": "speech",
+      "text": "Synthetic voice feature summary.",
+      "metadata": {
+        "synthetic": true,
+        "domain": "audio-ml"
+      }
+    }
+  ],
+  "graph_edges": [],
+  "confidence_threshold": 0.18,
+  "trained_on_synthetic_data": true
+}

project.json

@@ -0,0 +1,52 @@
+{
+  "name": "Audio Event Triage Baseline",
+  "problem": "Operations teams need an explainable starting point for classifying alarms, machinery noise, and speech-like events.",
+  "domain": "audio-ml",
+  "architecture": "classifier",
+  "hugging_face_tasks": [
+    "audio-classification",
+    "automatic-speech-recognition",
+    "feature-extraction",
+    "audio-to-audio"
+  ],
+  "recommended_stack": [
+    "FastAPI for audio metadata and prediction APIs",
+    "Transformers with Wav2Vec2 or audio spectrogram models",
+    "librosa or torchaudio for feature extraction",
+    "MLflow for experiment and model registry",
+    "Object storage for licensed audio",
+    "OpenTelemetry for inference latency and error traces"
+  ],
+  "real_world_data_sources": [
+    {
+      "name": "Freesound API",
+      "url": "https://freesound.org/apiv2/search/text/?query=alarm&page_size=5",
+      "purpose": "Licensed public audio search when an API key is configured"
+    },
+    {
+      "name": "Hugging Face Datasets API",
+      "url": "https://huggingface.co/api/datasets?search=audio-classification&limit=5",
+      "purpose": "Discover public audio datasets and metadata"
+    }
+  ],
+  "job_description_skills": [
+    "Audio feature pipelines and model fine-tuning",
+    "Class imbalance and macro-metric evaluation",
+    "Streaming inference and confidence calibration",
+    "Dataset licensing and provenance controls",
+    "Model monitoring and human review workflows"
+  ],
+  "impact_targets": [
+    "Reach macro recall >= 0.85 on a licensed evaluation set",
+    "Review every prediction below the confidence threshold",
+    "Process one minute of audio in under five seconds",
+    "Track class-level drift and false-negative rates"
+  ],
+  "baseline_evaluation": {
+    "test_examples": 4,
+    "accuracy": 1,
+    "synthetic_evaluation": true
+  },
+  "estimated_delivery": "8-12 weeks for one engineer",
+  "generated_baseline_is_production_ready": false
+}