PixVoice-YTMed

French Medical Speech Dataset

Developed and maintained by Pixemantic AI Lab

1. Overview

PixVoice-YTMed is a French medical speech dataset developed and maintained by Pixemantic AI Lab.

The dataset supports research and development in French medical speech processing, Automatic Speech Recognition (ASR), audio-to-text transcription, and healthcare-oriented artificial intelligence applications.

It contains 13,480 audio chunks associated with their corresponding textual transcriptions.

The dataset is organized into training and testing subsets and distributed using Apache Arrow storage.

2. Dataset Information

Property	Value
Dataset name	PixVoice-YTMed
Version	1.0.0
Organization	Pixemantic
Project	PixVoice
Language	French
Domain	Medical Speech
Modality	Audio and Text
Total samples	13,480
Training samples	12,137
Testing samples	1,343
Total Arrow shards	18
Storage format	Apache Arrow
License metadata	CC-BY-4.0

3. Dataset Composition

3.1 Training Split

The training split contains 12,137 samples distributed across 16 Arrow files.

train/
├── data-00000-of-00016.arrow
├── data-00001-of-00016.arrow
├── data-00002-of-00016.arrow
├── data-00003-of-00016.arrow
├── data-00004-of-00016.arrow
├── data-00005-of-00016.arrow
├── data-00006-of-00016.arrow
├── data-00007-of-00016.arrow
├── data-00008-of-00016.arrow
├── data-00009-of-00016.arrow
├── data-00010-of-00016.arrow
├── data-00011-of-00016.arrow
├── data-00012-of-00016.arrow
├── data-00013-of-00016.arrow
├── data-00014-of-00016.arrow
└── data-00015-of-00016.arrow

Training fingerprint:

73e475b5f71bf9d2

3.2 Testing Split

The testing split contains 1,343 samples distributed across 2 Arrow files.

test/
├── data-00000-of-00002.arrow
└── data-00001-of-00002.arrow

Testing fingerprint:

d329a55f82eb0ece

3.3 Split Summary

Split	Samples	Arrow shards
Train	12,137	16
Test	1,343	2
Total	13,480	18

4. Dataset Features

The dataset provides two principal features:

Feature	Data type	Description
chunk_files	Audio	Medical speech audio segments
chunk_texts	String	Corresponding textual transcriptions

Each sample associates an audio segment with its corresponding text transcription.

5. Loading the Dataset

Install the required libraries:

pip install -U datasets[audio]

Load the complete dataset:

from datasets import load_dataset

dataset = load_dataset(
    "Pixemantic/PixVoice-YTMed"
)

train_dataset = dataset["train"]
test_dataset = dataset["test"]

print("Training samples:", len(train_dataset))
print("Testing samples:", len(test_dataset))
print("Total samples:", len(train_dataset) + len(test_dataset))

print("Features:", train_dataset.features)

Expected output:

Training samples: 12137
Testing samples: 1343
Total samples: 13480

6. Streaming Access

For applications where downloading the complete dataset is not desirable:

from datasets import load_dataset

dataset = load_dataset(
    "Pixemantic/PixVoice-YTMed",
    streaming=True
)

for sample in dataset["train"]:
    print(sample)
    break

Streaming can be useful for large-scale experiments and memory-efficient data processing.

7. Research Applications

PixVoice-YTMed supports research and development in:

• French medical Automatic Speech Recognition.
• Medical audio transcription.
• Healthcare speech processing.
• Medical terminology recognition.
• Speech-to-text applications.
• Healthcare-oriented conversational AI.
• Speech recognition model training and evaluation.

8. PixVoice Research Directions

The dataset contributes to Pixemantic's PixVoice research initiative, focusing on intelligent voice technologies.

Potential research directions include:

• Whisper model fine-tuning.
• French medical ASR.
• Speech segmentation and preprocessing.
• Medical vocabulary recognition.
• LoRA and parameter-efficient fine-tuning.
• Word Error Rate (WER) and Character Error Rate (CER) evaluation.
• Deployment of speech recognition services.

9. Responsible Use

This dataset is intended for research and development purposes.

Users should consider:

• Data privacy and confidentiality.
• Applicable data protection regulations.
• Responsible handling of potentially sensitive medical information.
• Appropriate validation before healthcare deployment.

This dataset is not a certified clinical diagnostic resource.

10. Citation

If you use PixVoice-YTMed in academic research, please cite:

@dataset{touati_2026_pixvoice_ytmed,
  author = {Touati, Mohamed},
  organization = {Pixemantic},
  title = {PixVoice-YTMed},
  year = {2026},
  version = {1.0.0},
  url = {
    https://huggingface.co/datasets/Pixemantic/PixVoice-YTMed
  }
}

11. Organization

Pixemantic AI Lab

Artificial Intelligence Research and Development.

Hugging Face Organization: https://huggingface.co/Pixemantic

Dataset Repository: https://huggingface.co/datasets/Pixemantic/PixVoice-YTMed

---

Pixemantic AI Lab | PixVoice

Advancing Medical Speech AI.