EmoAlignBench Raw Dataset: Multimodal Financial Earnings Calls

Welcome to the raw dataset repository for EmoAlignBench. This dataset contains foundational multimodal data (audio, transcripts, and dual-track emotion annotations) from corporate earnings calls. It is specifically curated for analyzing emotion incongruence—the discrepancy between textual semantics and vocal prosody—within financial discourse.

📂 Repository Structure

The dataset is organized into parallel directories named by corporate Tickers. Each directory contains the raw audio, a metadata index, emotion-tagged transcripts (JSONL), and historical stock data.

.
├── TICKER/
│   ├── {uid}_Index.json            # Metadata index linking all modal files
│   ├── {uid}_Emotion.txt           # JSONL transcript with textual & vocal emotion tags
│   ├── {uid}.m4a                   # Raw audio file (or .mp3 / .wav)
│   └── {uid}_Stockprice.json       # Daily historical stock prices and volumes
└── README.md

📄 Data Formats & Fields

1. Metadata Index (*_Index.json)

Links the disparate data files for a specific earnings call event.

{
  "uid": "ABBV_20250425_EC",            // Unique event identifier
  "ticker": "ABBV",                     // Stock ticker symbol
  "event_start_et": "2025-04-25 09:00", // Event start time (US Eastern Time)
  "sound_file": "ABBV_20250425.m4a",    // Reference to the audio file
  "transcript_sound": "...Emotion.txt", // Reference to the emotion transcript
  "stock_price": "...Stockprice.json"   // Reference to the stock data
}

2. Emotion Transcripts (*_Emotion.txt)

A JSON-Lines file where each line is an utterance with synchronized timestamps and emotion labels.

{
  "speaker": "Analyst Name",            // Identity of the speaker
  "timestamp": "2025-04-25 09:15:20",   // Start time for audio alignment
  "text": "The margin pressure is...",  // Raw transcript text
  "text_emotion": "fear",               // Textual Emotion Label (fear, joy, none, etc.)
  "emotion": "😰",                      // Vocal Emotion emoji (😰, 😊, none, etc.)
  "events": ""                          // Auxiliary symbolic markers
}

3. Stock Price Data (*_Stockprice.json)

Standard daily time series data containing historical context around the call date.

{
  "Meta Data": { "2. Symbol": "ABBV", ... }, 
  "Time Series (Daily)": {                   
    "2025-04-25": {
      "1. open": "165.20",
      "2. high": "168.50",
      "3. low": "164.10",
      "4. close": "167.30",
      "5. volume": "5200300"
    }
  }
}

📝 License

This dataset is licensed under the Apache License 2.0.