ntkuhn/mlx_voice_commands

Ntkuhn/Mlx Voice Commands

Dataset Description

This dataset contains audio commands with voice control labels generated using text-to-speech synthesis.

Dataset Summary

• Total samples: 2745
• Number of classes: 13
• Audio format: MP3, 16kHz sampling rate
• Text-to-speech engine: Google Text-to-Speech (gTTS)

Class Distribution

Class Label	Count
	201
	200
	307
	100
	307
	100
	200
	200
	207
	206
	100
	309
	308

Dataset Structure

Data Fields

• audio (Audio): Audio file containing the spoken command
• original_sentence (string): Original sentence with special tokens
• clean_sentence (string): Clean sentence without special tokens (used for TTS)
• class_label (string): Command class label (e.g., <volume_up>, <volume_down>)
• info (string): Additional information about the audio generation

Example

{
    'audio': {'path': 'audio_0001_volume_up_16k.mp3', 'array': [...], 'sampling_rate': 16000},
    'original_sentence': 'Please <volume_up> the music',
    'clean_sentence': 'Please the music',
    'class_label': '<volume_up>',
    'info': 'gTTS, 16kHz mp3'
}

Usage

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("ntkuhn/mlx_voice_commands")

# Access audio and labels
for sample in dataset["train"]:
    audio = sample["audio"]
    label = sample["class_label"]
    clean_text = sample["clean_sentence"]
    print(f"Label: {label}, Text: {clean_text}")

Dataset Creation

This dataset was created using the generate_audio.py script, which:

1. Processes command sentences containing special tokens
2. Removes special tokens to create clean sentences
3. Generates audio using Google Text-to-Speech (gTTS)
4. Resamples audio to 16kHz for consistency

Licensing

Please refer to the original dataset license and Google's terms of service for gTTS usage.