hishab/MegaBNSpeech_Test_Data

---
language:
- bn
license: cc-by-nc-4.0
task_categories:
- automatic-speech-recognition
dataset_info:
features:
- name: audio
dtype: audio
- name: text
dtype: string
- name: duration
dtype: float64
- name: category
dtype: string
- name: source
dtype: string
splits:
- name: train
num_bytes: 219091915.875
num_examples: 1753
download_size: 214321460
dataset_size: 219091915.875
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---

# MegaBNSpeech Test Data

To evaluate the performance of the models, we used four test sets. Two of these were developed as part of the MegaBNSpeech corpus, while the remaining two (Fleurs and Common Voice) are commonly used test sets that are widely recognized by the speech community.

## Use dataset library:
```python
from datasets import load_dataset
dataset = load_dataset("hishab/MegaBNSpeech_Test_Data")
```
## Reported Word error rate (WER) /character error rate (CER) on four test sets using four ASR systems

| Category | Duration (hr) | Hishab BN Fastconformer | Google MMS | OOD-speech |
|-------------------- | -------------- | ------------ | ---------- | ----------- |
| MegaBNSpeech-YT | 8.1 | 6.4/3.39 | 28.3/18.88 | 51.1/23.49 |
| MegaBNSpeech-Tel | 1.9 | ∗40.7/24.38 | ∗59/41.26 | ∗76.8/39.36 |

## Reported Word error rate (WER) /character error rate (CER) on different categories present in Hishab BN FastConformer

| Category | Duration (hr) | Hishab BN FastConformer | Google MMS | OOD-speech |
|-------------------- | -------------- | ------------ | ---------- | ----------- |
| News | 1.21 | 2.5/1.21 | 18.9/10.46 | 52.2/21.65 |
| Talkshow | 1.39 | 6/3.29 | 28/18.71 | 48.8/21.5 |
| Courses | 3.81 | 6.8/3.79 | 30.8/21.64 | 50.2/23.52 |
| Drama | 0.03 | 10.3/7.47 | 37.3/27.43 | 64.3/32.74 |
| Science | 0.26 | 5/1.92 | 20.6/11.4 | 45.3/19.93 |
| Vlog | 0.18 | 11.3/6.69 | 33/22.9 | 57.9/27.18 |
| Recipie | 0.58 | 7.5/3.29 | 26.4/16.6 | 53.3/26.89 |
| Waz | 0.49 | 9.6/5.45 | 33.3/23.1 | 57.3/27.46 |
| Movie | 0.1 | 8/4.64 | 35.2/23.88 | 64.4/34.96 |

原始信息汇总

MegaBNSpeech Test Data

数据集信息

语言

• 孟加拉语 (bn)

许可

• CC BY-NC 4.0

任务类别

• 自动语音识别 (automatic-speech-recognition)

特征

• 音频 (audio)
• 文本 (text)
• 持续时间 (duration)
• 类别 (category)
• 来源 (source)

数据分割

• 训练集 (train)
  • 字节数: 219091915.875
  • 样本数: 1753

下载大小

• 214321460 字节

数据集大小

• 219091915.875 字节

配置

• 默认配置 (default)
  • 数据文件:
    • 训练集 (train): data/train-*

测试集性能

四个测试集的词错误率 (WER) 和字符错误率 (CER)

类别	持续时间 (小时)	Hishab BN Fastconformer	Google MMS	OOD-speech
MegaBNSpeech-YT	8.1	6.4/3.39	28.3/18.88	51.1/23.49
MegaBNSpeech-Tel	1.9	∗40.7/24.38	∗59/41.26	∗76.8/39.36

Hishab BN FastConformer 在不同类别上的词错误率 (WER) 和字符错误率 (CER)

类别	持续时间 (小时)	Hishab BN FastConformer	Google MMS	OOD-speech
News	1.21	2.5/1.21	18.9/10.46	52.2/21.65
Talkshow	1.39	6/3.29	28/18.71	48.8/21.5
Courses	3.81	6.8/3.79	30.8/21.64	50.2/23.52
Drama	0.03	10.3/7.47	37.3/27.43	64.3/32.74
Science	0.26	5/1.92	20.6/11.4	45.3/19.93
Vlog	0.18	11.3/6.69	33/22.9	57.9/27.18
Recipie	0.58	7.5/3.29	26.4/16.6	53.3/26.89
Waz	0.49	9.6/5.45	33.3/23.1	57.3/27.46
Movie	0.1	8/4.64	35.2/23.88	64.4/34.96