Datasets:
Update README.md
Browse files
README.md
CHANGED
|
@@ -27,25 +27,66 @@ tags:
|
|
| 27 |
- mongolian
|
| 28 |
- speech
|
| 29 |
- dataset
|
|
|
|
|
|
|
|
|
|
|
|
|
| 30 |
pretty_name: mbspeech_mn
|
| 31 |
---
|
| 32 |
|
| 33 |
-
Mongolian Biblical
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 34 |
|
| 35 |
-
Usage for LJSpeech speech like directory :
|
| 36 |
```python
|
| 37 |
import os
|
| 38 |
import csv
|
| 39 |
import soundfile as sf
|
| 40 |
from datasets import load_dataset
|
| 41 |
|
|
|
|
| 42 |
DATASET_NAME = "btsee/mbspeech_mn"
|
| 43 |
OUTPUT_DIR = "dataset"
|
| 44 |
WAV_DIR = os.path.join(OUTPUT_DIR, "wavs")
|
| 45 |
os.makedirs(WAV_DIR, exist_ok=True)
|
| 46 |
|
|
|
|
| 47 |
ds = load_dataset(DATASET_NAME, split="train")
|
| 48 |
|
|
|
|
| 49 |
with open(os.path.join(OUTPUT_DIR, "metadata.csv"), "w", newline="", encoding="utf-8") as f:
|
| 50 |
writer = csv.writer(f, delimiter="|")
|
| 51 |
for idx, item in enumerate(ds):
|
|
|
|
| 27 |
- mongolian
|
| 28 |
- speech
|
| 29 |
- dataset
|
| 30 |
+
- text-to-speech
|
| 31 |
+
- audio
|
| 32 |
+
- tts
|
| 33 |
+
- biblical
|
| 34 |
pretty_name: mbspeech_mn
|
| 35 |
---
|
| 36 |
|
| 37 |
+
# MBSpeech MN: Mongolian Biblical Speech Dataset
|
| 38 |
+
|
| 39 |
+
MBSpeech MN is a Mongolian text-to-speech (TTS) dataset derived from biblical texts. It consists of aligned audio recordings and corresponding sentences in Mongolian. The dataset is suitable for training TTS models and other speech processing applications.
|
| 40 |
+
|
| 41 |
+
## Dataset Summary
|
| 42 |
+
|
| 43 |
+
Language: Mongolian (mn)
|
| 44 |
+
|
| 45 |
+
Task: Text-to-Speech (TTS)
|
| 46 |
+
|
| 47 |
+
License: MIT
|
| 48 |
+
|
| 49 |
+
Size:
|
| 50 |
+
|
| 51 |
+
Download size: ~721 MB
|
| 52 |
+
|
| 53 |
+
Dataset size: ~822 MB
|
| 54 |
+
|
| 55 |
+
Examples: 3,846
|
| 56 |
+
|
| 57 |
+
## Dataset Structure
|
| 58 |
+
|
| 59 |
+
### Features
|
| 60 |
+
Name Type Description
|
| 61 |
+
audio Audio Audio data sampled at 16 kHz
|
| 62 |
+
sentence string Transcription in Mongolian
|
| 63 |
+
|
| 64 |
+
### Splits
|
| 65 |
+
Split Examples Size
|
| 66 |
+
train 3,846 ~822 MB
|
| 67 |
+
|
| 68 |
+
|
| 69 |
+
|
| 70 |
+
## Usage
|
| 71 |
+
|
| 72 |
+
To convert the dataset into an LJSpeech–style format for TTS model training:
|
| 73 |
|
|
|
|
| 74 |
```python
|
| 75 |
import os
|
| 76 |
import csv
|
| 77 |
import soundfile as sf
|
| 78 |
from datasets import load_dataset
|
| 79 |
|
| 80 |
+
# Dataset and output configuration
|
| 81 |
DATASET_NAME = "btsee/mbspeech_mn"
|
| 82 |
OUTPUT_DIR = "dataset"
|
| 83 |
WAV_DIR = os.path.join(OUTPUT_DIR, "wavs")
|
| 84 |
os.makedirs(WAV_DIR, exist_ok=True)
|
| 85 |
|
| 86 |
+
# Load dataset
|
| 87 |
ds = load_dataset(DATASET_NAME, split="train")
|
| 88 |
|
| 89 |
+
# Export audio files and metadata
|
| 90 |
with open(os.path.join(OUTPUT_DIR, "metadata.csv"), "w", newline="", encoding="utf-8") as f:
|
| 91 |
writer = csv.writer(f, delimiter="|")
|
| 92 |
for idx, item in enumerate(ds):
|