HAMMALE/speecht5-darija

Moroccan Darija Text-to-Speech Model

This model is a fine-tuned version of SpeechT5 for Moroccan Darija Text-to-Speech synthesis.

Model Details

• Base Model: Microsoft SpeechT5
• Fine-tuned on: DODa audio dataset
• Languages: Moroccan Darija (Latin script)
• Features: Multiple voice support (male/female)
• Release Date: April 2025

Usage

from transformers import SpeechT5Processor, SpeechT5ForTextToSpeech, SpeechT5HifiGan
import torch
import soundfile as sf

# Load models
processor = SpeechT5Processor.from_pretrained("HAMMALE/speecht5-darija")
model = SpeechT5ForTextToSpeech.from_pretrained("HAMMALE/speecht5-darija")
vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")

# Load speaker embedding (replace with your own speaker embedding)
speaker_embedding = torch.randn(1, 512)  # Example embedding

# Process text
text = "Salam, kifach nta lyoum?"
inputs = processor(text=text, return_tensors="pt")

# Generate speech
speech = model.generate_speech(inputs["input_ids"], speaker_embedding, vocoder=vocoder)

# Save audio file
sf.write("output.wav", speech.numpy(), 16000)

Demo

A live demo is available at Hugging Face Spaces

License

This model is available under the MIT License.

Acknowledgments

• The DODa audio dataset creators
• Microsoft Research for the SpeechT5 model architecture