Natiq-Studio/Natiq-dz-ARG
Text-to-Speech • Updated • 244 • 7
Datasets:
License:
Request dataset access
Access is reviewed manually. Please describe your organization and intended use. Submitting this form does not grant access or a license.
Commercial licensing is available through Natiq Studio. We will use your submitted information to review your request and contact you about scope, licensing and pricing. Any required agreement and payment are handled separately from Hugging Face.
Log in or Sign Up to review the conditions and access this dataset content.
Algerian Darija Speech Dataset
Production-grade speech corpus for Algerian Darija Text-to-Speech (TTS), Automatic Speech Recognition (ASR), and voice cloning.
By Natiq Studio 🇩🇿
Dataset Overview & Key Metrics
A comprehensive, multi-speaker Algerian Darija speech collection standardized for high-fidelity speech synthesis and conversational AI research, featuring rich French-Arabic code-switching and complete acoustic metadata.
| Metric | Specification |
|---|---|
| Total Audio Duration | ~303.1 Hours |
| Total Utterances / Clips | 101,034 clips |
| Audio Sampling Rate | 24,000 Hz (24 kHz Mono) |
| Audio Format | Standard Parquet with embedded 24 kHz WAV audio |
| Primary Language | Algerian Darija (الدارجة الجزائرية) |
| Linguistic Characteristic | Authentic conversational speech with natural French code-switching |
| Speaker Coverage | Diverse multi-speaker representation across distinct acoustic environments |
| Acoustic Annotations | Paired text transcripts, duration, SNR, energy levels, speaker embeddings, and zero-shot reference pairs |
| Applications | Text-to-Speech (TTS), Speech Recognition (ASR), Zero-Shot Voice Cloning |
| Availability | Gated manual approval; commercial licensing via Natiq Studio |
Corpus Breakdown
The dataset is organized into four complementary subsets covering diverse acoustic conditions and speaking registers:
| Subset | Clean Utterances | Duration | Speech Profile & Acoustic Context |
|---|---|---|---|
subset_01 |
4,685 clips | ~8.6 Hours | Single-speaker clean narrative voice with high acoustic signal-to-noise ratio and zero background artifacts. |
subset_02 |
11,069 clips | ~52.1 Hours | Multi-speaker dynamic dialogue and natural conversational turn-taking in spontaneous settings. |
subset_03 |
23,750 clips | ~115.7 Hours | Single-speaker clean female voice with highly consistent vocal timbre, ideal for acoustic base modeling. |
subset_04 |
61,530 clips | ~126.7 Hours | Multi-speaker long-form discussions covering professional, technical, academic, financial, and healthcare topics with heavy code-switching. |
| Total | 101,034 clips | ~303.1 Hours | Unified 24 kHz Speech Corpus |
Audio Quality & Listening Evaluation
Audio Evaluation Samples: The clips below are just preview samples provided to illustrate acoustic fidelity and natural French code-switching across topics. They are short excerpts for listening evaluation only; the complete ~303.1-hour proprietary dataset is gated and accessible upon licensing review.
1. Technology & Engineering
7.10 seconds • 24 kHz audio
Transcript: Développeur؟ لا لا لا، physicien de base، كيما كان يقول واحد le prof تاع physique، qui peut le plus peut le moins، بلا ما نعايروا حتى واحد.
2. Finance & Markets
8.23 seconds • 24 kHz audio
Transcript: يدخل للموقع الرسمي تاع la COSOB، cosob point dz, point org point dz، يدخل IOB ولا promoteur en bourse يلقى، كاين معلومات تما.
3. Academic Research & Science
8.16 seconds • 24 kHz audio
Transcript: ودرست thin film semiconductor في التطبيق في الطاقة الشمسية ودرت master و PhD.
Commercial Licensing & Access
Commercial licensing is available. Submit an access request above and describe your organization and intended use. We will contact you regarding licensing, pricing and delivery.
- Submit Access Request: Use the Hugging Face access request form above.
- Review & Scope: We review intended use and requirements (academic, evaluation, or commercial deployment).
- Licensing Agreement: Agree on terms, permitted use scope, and delivery specifications.
- Access Approval: Repository access is granted manually upon completion of the agreement.
Licensing Terms
This is a proprietary dataset, not an open-data release. A public dataset page, submitted request, or technical download permission does not by itself grant permission to train models or commercialize resulting weights without an agreed license.
Redistribution, public hosting, or sublicensing of recordings, transcripts, or dataset shards is strictly prohibited without prior written agreement. See LICENSE.md.
Contact
For licensing inquiries and dataset specifications:
- Email: abdelhak.kadouci@ensia.edu.dz
- Organization: Natiq Studio
- Downloads last month
- 99
Models trained or fine-tuned on Natiq-Studio/Algeria-TTS-dataset