---
dataset_info:
  features:
  - name: ID
    dtype: string
  - name: speaker_id
    dtype: string
  - name: Language
    dtype: string
  - name: audio
    dtype: 
      audio:
        sampling_rate: 16000
  - name: transcript
    dtype: string
  - name: length
    dtype: float32
  - name: dataset_name
    dtype: string
  - name: confidence_score
    dtype: float64
  splits:
  - name: train
    num_examples: 100
  download_size: 0
  dataset_size: 0
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train/*.parquet
---

# Thanarit/Thai-Voice

Combined Thai audio dataset from multiple sources

## Dataset Details

- **Total samples**: 100
- **Total duration**: 0.11 hours
- **Language**: Thai (th)
- **Audio format**: 16kHz mono WAV
- **Volume normalization**: -20dB

## Sources

Processed 1 datasets in streaming mode

## Source Datasets

1. **GigaSpeech2**: Large-scale multilingual speech corpus

## Usage

```python
from datasets import load_dataset

# Load with streaming to avoid downloading everything
dataset = load_dataset("Thanarit/Thai-Voice-10000000", streaming=True)

# Iterate through samples
for sample in dataset['train']:
    print(sample['ID'], sample['transcript'][:50])
    # Process audio: sample['audio']
    break
```

## Schema

- `ID`: Unique identifier (S1, S2, S3, ...)
- `speaker_id`: Speaker identifier (SPK_00001, SPK_00002, ...)
- `Language`: Language code (always "th" for Thai)
- `audio`: Audio data with 16kHz sampling rate
- `transcript`: Text transcript of the audio
- `length`: Duration in seconds
- `dataset_name`: Source dataset name (e.g., "GigaSpeech2", "ProcessedVoiceTH", "MozillaCommonVoice")
- `confidence_score`: Confidence score of the transcript (0.0-1.0)
  - 1.0: Original transcript from source dataset
  - <1.0: STT-generated transcript
  - 0.0: Fallback transcript (e.g., [NO_TRANSCRIPT])

## Processing Details

This dataset was created using streaming processing to handle large-scale data without requiring full downloads.
Audio has been standardized to 16kHz mono with -20dB volume normalization.