Building AI Speech for Algerian Darija: Introducing Natiq-dz-ARG

Description and use

Natiq-dz-ARG is an open‑weight TTS model for Algerian Darija that addresses naturalness and code‑switching issues, offering zero‑shot voice cloning.

Problem

Existing TTS models produce unnatural speech for Algerian Darija, especially with French code‑switching.

Outcome

Natiq-dz-ARG delivers realistic Algerian Darija speech with code‑switching support and zero‑shot voice cloning, enabling developers to integrate local dialect TTS into applications.

Workflow

• Collect 300+ hours of Algerian speech from 100+ speakers
• Clean and correct mixed‑language transcripts to preserve French words
• Fine‑tune Higgs TTS 3.4B with LoRA on the curated corpus
• Deploy the LoRA adapter on Hugging Face
• Provide a REST API via Natiq Studio for easy integration
• Offer a hosted platform with pre‑built voices and a free token tier

Evidence

Most text-to-speech (TTS) models today can generate impressive speech in English, French, and Modern Standard Arabic. But what happens when you try generating speech in Algerian Darija?

Source

Algerian Darija has unique vocabulary, pronunciation patterns, and linguistic influences from Arabic, French, and Amazigh.

Source

300+ hours of Algerian speech used for fine‑tuning. 100+ speakers to capture different voices, accents, and speaking styles.

Source

We used Low‑Rank Adaptation (LoRA) to fine‑tune the existing bosonai/higgs-tts-3-4b model.

Source

The inference process accepts: a reference audio recording, its transcript, and new Darija text to synthesize.

Source