zant-os/zant-echo-golden

license: cc0-1.0 task_categories:

• —audio-classification language:
• —en tags:
• —speaker-diarization
• —test-dataset size_categories:
• —n<1K ---

# ZantOS Golden Test Set

Human-recorded meeting audio with ground truth speaker annotations for acceptance testing.

## Dataset Details

• —Version: 1.0.0
• —Clips: 3 meetings (2-4 minutes each)
• —Speakers: 2-4 per clip
• —Format: 16kHz mono WAV
• —Annotation: RTTM format (Rich Transcription Time Marked)

## Structure

v1.0.0/ ├── metadata.json # Dataset version info ├── audio/ │ ├── meeting001.wav # 3 min, 2 speakers, clear audio │ ├── meeting002.wav # 4 min, 3 speakers, some cross-talk │ └── meeting003.wav # 2 min, 4 speakers, challenging └── reference/ ├── meeting001.rttm # Ground truth annotations ├── meeting002.rttm └── meeting003.rttm

## RTTM Format

Each line represents a speaker segment: SPEAKER meeting001 1 0.00 5.32 speaker0 SPEAKER meeting001 1 5.32 12.15 speaker1

Fields: type file channel start duration _ _ speaker_id _ _

## Usage

Download via Python:

  from datasets import load_dataset

  dataset = load_dataset("zant-os/zant-echo-golden", split="v1.0.0")

  Or clone directly:
  git clone https://huggingface.co/datasets/zant-os/zant-echo-golden

  Quality Targets

  - DER (Diarization Error Rate): ≤18%
  - Speaker detection accuracy: ≥95%
  - Overlap handling: Graceful degradation

  License

  CC0 1.0 Universal (Public Domain)