burkimbia/sidbi-ziri-dataset

Sid Bi Ziri — Moore Language Speech Dataset

Dataset Overview

56,992 audio segments of Moore language speech extracted from the Sid Bi Ziri TV show (SAVANE TV, Burkina Faso). Transcriptions were automatically generated by BIA-WHISPER (burkimbia/BIA-WHISPER-LARGE-SACHI_V2), a Whisper model fine-tuned on Moore.

Note on transcription quality: This dataset is designed for TTS fine-tuning (SparkTTS). Since SparkTTS does not need to semantically understand the language — it learns the mapping between text tokens and BiCodec audio tokens (semantic + global) — automatic ASR transcriptions of good quality are sufficient for mono-speaker TTS fine-tuning.

---

Key Statistics

Metric	Value
Total segments	56,992
Audio format	FLAC, 16kHz mono
Segment duration	1-15 seconds
Language	Moore (Mooré) - mos
Transcription	Automatic (BIA-WHISPER)
Source	YouTube - SAVANE TV playlist

---

Data Fields

• audio - FLAC audio file (16kHz)
• text - Moore transcription (ASR, BIA-WHISPER)
• duration - Duration in seconds
• language - "moore"
• video_id - Source YouTube video ID

---

Technical Processing

Audio Pipeline:

• Source: YouTube playlist (Sid Bi Ziri - SAVANE TV)
• Segmentation: 1-15s chunks using VAD
• Format: FLAC 16kHz mono
• Storage: S3 (Tigris) -> HuggingFace parquet shards

Transcription:

• Model: burkimbia/BIA-WHISPER-LARGE-SACHI_V2
• Language: Moore (mos)
• Method: Batch inference, automatic

Pipeline Tools:

• https://github.com/BurkimbIA/tts-dataset-gen

---

Usage

from datasets import load_dataset
import os

dataset = load_dataset(
    "sawadogosalif/sidbi-ziri-dataset",
    split="train",
    token=os.environ["HF_TOKEN"]
)

Research Applications:

• Text-to-Speech (TTS) for Moore
• Automatic Speech Recognition (ASR)
• Low-resource African language speech technology
• Cross-lingual speech processing

---

Contributors and Credits

Dataset Preparation: @sawadogosalif Salif SAWADOGO - BurkimbIA

Related Projects:

• sawadogosalif/speech-dataset-public - Moore speech corpus (150h, manually validated)
• burkimbia/BIA-WHISPER-LARGE-SACHI_V2 - ASR model used for transcription
• burkimbia/BIA-SPARKTTS-V4 - TTS model trained on this data

---

Usage Guidelines

Permitted Uses

• Academic research and education
• Non-commercial AI model development
• Open-source community projects

Prohibited Uses

• Commercial speech recognition or TTS services
• Proprietary product development
• Revenue-generating applications

By using this dataset, you agree to respect these terms and contribute positively to the Burkina Faso AI community.

---

Advancing AI technologies for African languages through collaborative, high-quality linguistic resources.