---
dataset_info:
  features:
  - name: audio
    dtype: audio
  splits:
  - name: train
    num_bytes: 109272060.0
    num_examples: 4
  download_size: 106357055
  dataset_size: 109272060.0
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---

# Audio Only Eval Dataset
This repo contains a few audio examples for kotoba-whisper evaluation (no transcription). Currently, it includes 1-hour presentation audio, and three manzai recordings (five min each).


| path                 |  array size |   sampling rate |      duration (min) |
|:---------------------|----------:|----------------:|---------:|
| long_interview_1.mp3 | 133,197,167 |           44,100 | 50.3 |
| manzai3.mp3          |   5,343,599 |           16,000 |  5.6 |
| manzai2.mp3          |   4,703,434 |           16,000 |  4.9 |
| manzai1.mp3          |   5,383,343 |           16,000 |  5.6 |