saudi-tts

Saudi TTS

A high-quality Text-to-Speech model for Saudi Arabic dialect.

Sample Audio

<audio controls> <source src="https://huggingface.co/AhmedEladl/saudi-tts/resolve/main/sample_audio.wav" type="audio/wav"> </audio>

Quick Start

Dependencies to install

pip install git+https://github.com/coqui-ai/TTS
pip install transformers
pip install deepspeed

Inference

Load the model

import os
import torch
import torchaudio
from TTS.tts.configs.xtts_config import XttsConfig
from TTS.tts.models.xtts import Xtts

CONFIG_FILE_PATH = 'path/to/config.json'
VOCAB_FILE_PATH = 'path/to/vocab.json'
MODEL_PATH = 'path/to/model'
SPEAKER_AUDIO_PATH = 'path/to/speaker.wav'

print("Loading model...")

config = XttsConfig()
config.load_json(CONFIG_FILE_PATH)

model = Xtts.init_from_config(config)
model.load_checkpoint(config, checkpoint_dir=MODEL_PATH, use_deepspeed=True, vocab_path=VOCAB_FILE_PATH)
model.cuda()

print("Computing speaker latents...")
gpt_cond_latent, speaker_embedding = model.get_conditioning_latents(audio_path=[SPEAKER_AUDIO_PATH])

Run the model

from IPython.display import Audio, display

text = ".هذا مثال على النص العربي باللهجة السعودية"
print("Inference...")
out = model.inference(
    text,
    "ar",
    gpt_cond_latent,
    speaker_embedding,
    temperature=0.75,
)

AUDIO_OUTPUT_PATH = "path/to/output_audio.wav"
torchaudio.save("tts_audio.wav", torch.tensor(out["wav"]).unsqueeze(0), 24000)
display(Audio(AUDIO_OUTPUT_PATH, autoplay=True))

Files

• config.json - Model configuration
• vocab.json - Vocabulary file
• speaker.wav - Reference speaker audio
• model.pth - Model weights

Citation

@misc{eladl-sauditts,
      author = {Ahmed Eladl},
      title = {Saudi TTS: A High-Quality Text-to-Speech Model for Saudi Arabic Dialect},
      year = {2025},
      url = {https://huggingface.co/AhmedEladl/saudi-tts}
}

---