voice-to-report

Voice to Report

Overview

Field workers prefer talking over typing. This skill converts voice recordings into structured construction reports using speech-to-text and LLM processing.

Why Voice?

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                    VOICE TO REPORT PIPELINE                      │
├─────────────────────────────────────────────────────────────────┤
│                                                                  │
│  🎤 Voice      →    📝 Transcribe    →    🤖 Structure    →    📊 Report │
│  Recording         Whisper API           GPT-4o               Formatted  │
│                                                                  │
│  "We finished      "We finished         {                    Daily Report │
│   the foundation    the foundation       "activity":         ──────────── │
│   pour today,       pour today,          "foundation",       Foundation   │
│   about 500         about 500            "quantity": 500,    pour: 500m³  │
│   cubic meters"     cubic meters"        "unit": "m³"        Complete ✓   │
│                                          }                               │
└─────────────────────────────────────────────────────────────────┘

Quick Start

from openai import OpenAI
import json

client = OpenAI()

def voice_to_report(audio_path: str, report_type: str = "daily") -> dict:
    """Convert voice recording to structured report"""

    # Step 1: Transcribe audio
    with open(audio_path, "rb") as audio_file:
        transcript = client.audio.transcriptions.create(
            model="whisper-1",
            file=audio_file,
            language="en"
        )

    # Step 2: Structure with LLM
    schema = get_report_schema(report_type)

    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[
            {
                "role": "system",
                "content": f"""You are a construction report assistant.
                Convert the voice transcript into a structured report.
                Extract all relevant information and format as JSON.

                Report type: {report_type}
                Schema: {json.dumps(schema, indent=2)}

                Rules:
                - Extract quantities with units
                - Identify activities and locations
                - Note any issues or concerns
                - Capture weather if mentioned
                - List workers/trades if mentioned
                """
            },
            {
                "role": "user",
                "content": f"Transcript:\n{transcript.text}"
            }
        ],
        response_format={"type": "json_object"}
    )

    return {
        "transcript": transcript.text,
        "structured_report": json.loads(response.choices[0].message.content)
    }