README.md

metadata

title: Arabic TTS (MMS)
emoji: 🗣️
colorFrom: indigo
colorTo: green
sdk: streamlit
sdk_version: 1.38.0
app_file: app.py
pinned: false

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

Arabic TTS Streamlit App

Arabic text-to-speech demo powered by open-source Hugging Face checkpoints:

• facebook/mms-tts-ara (MMS Modern Standard Arabic)
• wasmdashai/vits-ar-sa-A (community VITS Saudi Arabic voice)
• MBZUAI/speecht5_tts_clartts_ar (SpeechT5 CLAra for Classical Arabic)
• AhmedEladl/saudi-tts (community Saudi Arabic TTS model)

Features

• Local inference for MMS, community VITS, SpeechT5, and XTTS engines (GPU-aware via PyTorch)
• Sidebar model download button for caching checkpoints under models_cache/
• Downloadable app.log for easy debugging/support
• Hugging Face token input if you later add hosted models requiring authentication

Prerequisites

sudo apt-get update && sudo apt-get install -y python3-venv espeak-ng
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Optional: Install PyTorch w/ CUDA

Pick the command for your driver from https://pytorch.org/get-started/locally/

Example (CUDA 12.1):

pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu121

CPU-only fallback:

pip install torch --index-url https://download.pytorch.org/whl/cpu

Run the App

streamlit run app.py

Open the printed URL (usually http://localhost:8501).

Using the UI

1. Pick an Arabic model in the sidebar.
2. Expand Model assets to download weights locally (recommended for MMS/VITS/SpeechT5).
3. Enter Arabic text and click 🔊 Generate Speech.
4. Download app.log from the sidebar if you need to inspect or share errors.

Troubleshooting

• StopIteration from hosted inference: current models run locally only.
• Message "النص المدخل لم ينتج أي رموز…": the tokenizer stripped everything. Add Arabic letters or a longer phrase.
• SpeechT5 downloads HiFi-GAN + CMU x-vectors the first time; keep the tab open.
• SpeechT5 also requires sentencepiece (already listed in requirements.txt). If you see an import error, re-run pip install sentencepiece and restart the app.
• For AhmedEladl/saudi-tts (XTTS-style), provide local paths in the sidebar: config.json, vocab.json, the checkpoint directory containing the .pth file, and a speaker.wav. If you don't have the Coqui TTS library installed, run pip install TTS.
• CUDA OOM: shorten the text input or set CUDA_VISIBLE_DEVICES="" to force CPU inference.

License

• Each Hugging Face model keeps its own license; review before redistribution.
• App scaffolding may be reused freely in your projects.