metadata

dataset_info:
  features:
    - name: audio
      dtype:
        audio:
          sampling_rate: 16000
    - name: sentence_orig
      dtype: string
    - name: sentence_norm
      dtype: string
  splits:
    - name: train
      num_bytes: 822408859.068
      num_examples: 3846
  download_size: 755846406
  dataset_size: 822408859.068
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
license: mit
language:
  - mn
task_categories:
  - text-to-speech
tags:
  - mongolian
  - speech
  - dataset
  - text-to-speech
  - audio
  - tts
  - biblical
pretty_name: mbspeech_mn

MBSpeech MN: Mongolian Biblical Speech Dataset

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.

Dataset Summary

Language: Mongolian (mn)

Task: Text-to-Speech (TTS)

License: MIT

Size:

  Download size: ~721 MB

  Dataset size: ~822 MB

  Examples: 3,846

Dataset Structure

Features

Name Type Description audio Audio Audio data sampled at 16 kHz sentence string Transcription in Mongolian

Splits

Split Examples Size train 3,846 ~822 MB

Usage

To convert the dataset into an LJSpeech–style format for TTS model training:

import os
import csv
import soundfile as sf
from datasets import load_dataset

# Dataset and output configuration
DATASET_NAME = "btsee/mbspeech_mn"
OUTPUT_DIR = "dataset"
WAV_DIR = os.path.join(OUTPUT_DIR, "wavs")
os.makedirs(WAV_DIR, exist_ok=True)

# Load dataset
ds = load_dataset(DATASET_NAME, split="train")

# Export audio files and metadata
with open(os.path.join(OUTPUT_DIR, "metadata.csv"), "w", newline="", encoding="utf-8") as f:
    writer = csv.writer(f, delimiter="|")
    for idx, item in enumerate(ds):
        array = item["audio"]["array"]
        sr = item["audio"]["sampling_rate"]
        text = item["sentence"]
        fname = f"{idx:05d}"
        path = os.path.join(WAV_DIR, f"{fname}.wav")
        sf.write(path, array, sr, subtype="PCM_16")
        writer.writerow([fname, text])