nyu-dice-lab/wave-pulse-transcripts

Multi-Speaker Audio Transcripts Dataset

Dataset Summary

This dataset consists of 15844201 JSON files, where each file contains structured audio transcript data. Each entry includes:

• Timestamps (start, end): Marking the start and end of the audio segment.
• Transcription (text): The text corresponding to the audio segment.
• Speaker Identification (speaker): The speaker label for each segment.

The dataset is suitable for tasks such as:

• Speech recognition
• Speaker diarization
• Natural language processing
• Audio-text alignment

---

Supported Tasks and Leaderboards

The dataset supports the following tasks:

• Speech-to-Text: Generating transcriptions from audio data.
• Speaker Diarization: Identifying "who spoke when."
• Text Segmentation: Splitting text into meaningful units based on timestamps.

---

Languages

The dataset is primarily in English (en).

---

Dataset Structure

Data Fields

Each JSON file contains a list of objects with the following fields:

• start (float32): Start time of the segment in seconds.
• end (float32): End time of the segment in seconds.
• text (string): Transcribed text for the segment.
• speaker (string): Speaker label (e.g., SPEAKER_05).

Example Data

[
  {
    "start": 946.93,
    "end": 948.391,
    "text":"Here 's Anna with headlines .",
    "speaker":"SPEAKER_04"
  },
  {
    "start": 948.532,
    "end": 959.54,
    "text":"Pope Francis today has accepted the
    retirement of the longtime Archbishop of
    Boston , Cardinal Sean O' Malley and chosen
    Providence Bishop Richard Henning to be his
    successor .",
    "speaker":"SPEAKER_19"
  },
  {
    "start": 960.601,
    "end": 964.004,
    "text":"Tropical Storm Debbie is now a hurricane
    .",
    "speaker":"SPEAKER_19"
  },
  {
    "start": 964.585,
    "end": 965.045,
    "text":"What , Matt ?",
    "speaker":"SPEAKER_19"
  },
  {
    "start": 966.766,
    "end": 968.087,
    "text":"Oh , no , my mic is still alive .",
    "speaker":"SPEAKER_04"
  }
]