|
|
--- |
|
|
language: en |
|
|
license: mit |
|
|
tags: |
|
|
- speech |
|
|
- audio |
|
|
- vocoder |
|
|
- hifigan |
|
|
- tts |
|
|
--- |
|
|
|
|
|
# HiFiGAN Arabic Vocoder |
|
|
|
|
|
A standalone implementation of HiFiGAN vocoder for Arabic text-to-speech, based on the paper "HiFi-GAN: Generative Adversarial Networks for Efficient and High Fidelity Speech Synthesis" (https://arxiv.org/pdf/2010.05646.pdf). |
|
|
|
|
|
## Usage |
|
|
|
|
|
```python |
|
|
from hifigan_ar_v2 import HiFiGANArabicGenerator |
|
|
import torch |
|
|
|
|
|
# Load the model |
|
|
model = HiFiGANArabicGenerator.from_pretrained("generator.ckpt", "config.json") |
|
|
|
|
|
# Generate audio from mel spectrogram |
|
|
mel = torch.rand(1, 80, 122) # Example mel spectrogram |
|
|
audio = model(mel) # Shape: [1, 1, 8448] |
|
|
``` |
|
|
|
|
|
## Model Details |
|
|
- Sample Rate: 22050 Hz |
|
|
- Input: Mel spectrogram (80 channels) |
|
|
- Output: Audio waveform (1 channel) |
|
|
|
