ModelListedExternal

Saudi Xtts v2

A fine-tuned XTTS-v2 model for Saudi Arabic / English code-switching text-to-speech, with a single female speaker voice ("Hoda").

Sourcehuggingface.co/Rabe3/saudi-xtts-v2/tree/be0e80198bdb21b2095563197e9194753ce08209

KindOtherBase modelcoqui/XTTS-v2VersionRevision be0e801

Model source

Kind: Other
File size: 1.04 KB–5.22 GB
Base model: coqui/XTTS-v2
Version: vRevision be0e801
Updated: Sep 21, 2026
Downloads: 246
Tasks: Audio
Source: Hugging Face

Source description

Saudi XTTS-v2

A fine-tuned XTTS-v2 model for Saudi Arabic / English code-switching text-to-speech, with a single female speaker voice ("Hoda").

Model Details

Detail	Value
Base model	coqui/XTTS-v2
Fine-tuned on	~50 000 synthetic Saudi Arabic sentences
Speaker	Hoda (female, Saudi)
Languages	Arabic (`ar`) + English (`en`)
Sample rate	24 000 Hz
Training dialect	Najdi / Saudi colloquial Arabic (`ars`)
Precision	fp16
Checkpoint step	320 190

Training Data

The model was trained on ~50 000 synthetically generated sentences covering:

Saudi dialect corpus segments (TaghreedT/SDC, ~60%)
Code-switching templates mixing Saudi Arabic and English (~24%)
Number verbalization sentences — digits rendered in Saudi colloquial words (~16%)

Audio was synthesized with the lahgtna-omnivoice-v2 TTS system using the Najdi Arabic (ars) voice.

Files

File	Description
`model.pth`	Fine-tuned GPT weights (upload this as the XTTS checkpoint)
`config.json`	Training / inference configuration
`vocab.json`	XTTS-v2 tokenizer vocabulary
`dvae.pth`	Discrete VAE (from XTTS-v2 base, unchanged)
`mel_stats.pth`	Mel spectrogram normalisation stats (from XTTS-v2 base, unchanged)
`speakers_xtts.pth`	Speaker embedding library (from XTTS-v2 base)
`reference_audios/hoda.wav`	Reference audio for voice cloning
`references.json`	Speaker metadata

Quick Start

from TTS.api import TTS

tts = TTS(model_path="Rabe3/saudi-xtts-v2", progress_bar=True)

tts.tts_to_file(
    text="كيف الحال؟ وش قاعد تسوي اليوم؟",
    speaker_wav="reference_audios/hoda.wav",
    language="ar",
    file_path="output.wav",
)

Or using the XTTS model directly:

from TTS.tts.configs.xtts_config import XttsConfig
from TTS.tts.models.xtts import Xtts

config = XttsConfig()
config.load_json("config.json")

model = Xtts.init_from_config(config)
model.load_checkpoint(
    config,
    checkpoint_dir=".",   # directory containing model.pth, dvae.pth, etc.
    use_deepspeed=False,
)
model.cuda()

outputs = model.synthesize(
    text="أبغى أقول لك عن the new project اللي قاعدين نشتغل عليه",
    config=config,
    speaker_wav="reference_audios/hoda.wav",
    language="ar",
)

License

CC BY-NC 4.0 — non-commercial use only, consistent with the XTTS-v2 base model license.

Sources

1 source

Hugging FacePrimaryVersion be0e80198bdb21b2095563197e9194753ce08209

huggingface.co/Rabe3/saudi-xtts-v2/tree/be0e80198bdb21b2095563197e9194753ce08209

Verified Sep 21

Model artifacts

4 artifacts

dvae.pth

pth · 201 MB · SHA-256 b29bc227d410…d97a · Hugging Face

Download

mel_stats.pth

pth · 1.04 KB · SHA-256 1f69422a8a8f…119c · Hugging Face

Download

model.pth

pth · 5.22 GB · SHA-256 8b2a11d7e97a…ceff · Hugging Face

Download

speakers_xtts.pth

pth · 7.40 MB · SHA-256 f0f6137c19a4…943b · Hugging Face

Download

Source excerpts

2 excerpts

ReadmeSep 21huggingface.co/Rabe3/saudi-xtts-v2

--- language: - ar - en license: cc-by-nc-4.0 library_name: coqui pipeline_tag: text-to-speech tags: - text-to-speech - tts - xtts - xtts-v2 - arabic - saudi-arabic - najdi-arabic - code-switching - voice-cloning base_model: - coqui/XTTS-v2 --- # Saudi XTTS-v2 A fine-tuned [XTTS-v2](https://huggingface.co/coqui/XTTS-v2) model for **Saudi Arabic / English code-switching** text-to-speech, with a single female speaker voice ("Hoda"). ## Model Details | Detail | Value | |---|---| | Base model | coqui/XTTS-v2 | | Fine-tuned on | ~50 000 synthetic Saudi Arabic sentences | | Speaker | Hoda (female, Saudi) | | Languages | Arabic (`ar`) + English (`en`) | | Sample rate | 24 000 Hz | | Training dialect | Najdi / Saudi colloquial Arabic (`ars`) | | Precision | fp16 | | Checkpoint step | 320 190 | ## Training Data The model was trained on ~50 000 synthetically generated sentences covering: - **Saudi dialect corpus** segments (TaghreedT/SDC, ~60%) - **Code-switching templates** mixing Saudi Arabic and English (~24%) - **Number verbalization** sentences — digits rendered in Saudi colloquial words (~16%) Audio was synthesized with the `lahgtna-omnivoice-v2` TTS system using the Najdi Arabic (`ars`) voice. ## Files | File | Description | |---|---| | `model.pth` | Fine-tuned GPT weights (upload this as the XTTS checkpoint) | | `config.json` | Training / inference configuration | | `vocab.json` | XTTS-v2 tokenizer vocabulary | | `dvae.pth` | Discrete VAE (from XTTS-v2 base, unchanged) | | `mel_stats.pth` | Mel spectrogram normalisation stats (from XTTS-v2 base, unchanged) | | `speakers_xtts.pth` | Speaker embedding library (from XTTS-v2 base) | | `reference_audios/hoda.wav` | Reference audio for voice cloning | | `references.json` | Speaker metadata | ## Quick Start ```python from TTS.api import TTS tts = TTS(model_path="Rabe3/saudi-xtts-v2", progress_bar=True) tts.tts_to_file( text="كيف الحال؟ وش قاعد تسوي اليوم؟", speaker_wav="reference_audios/hoda.wav", language="ar", file_path="output.wav", ) ``` Or using the XTTS model directly: ```python from TTS.tts.configs.xtts_config import XttsConfig from TTS.tts.models.xtts import Xtts config = XttsConfig() config.load_json("config.json") model = Xtts.init_from_config(config) model.load_checkpoint( config, checkpoint_dir=".", # directory containing model.pth, dvae.pth, etc. use_deepspeed=False, ) model.cu...

NoteSep 21huggingface.co/Rabe3/saudi-xtts-v2

Source context: 246 downloads · 0 likes · Pipeline text-to-speech · Library coqui · Repo Rabe3/saudi-xtts-v2