Juanfa/F5-Spanish-MLX-Compat

This model is a fork of Spanish-F5 renamed to work with https://github.com/lucasnewman/f5-tts-mlx

Example command to run: python3 -m f5_tts_mlx.generate --model "Juanfa/F5-Spanish-MLX-Compat" --text "Mi nombre es manuel" --duration 10

---

GitHub

F5-TTS Spanish Language Model (MLX Compat)

Overview

The F5-TTS model is finetuned specifically for Spanish language speech synthesis. This project aims to deliver high-quality, regionally diverse speech synthesis capabilities for Spanish speakers.

License

This model is released under the CC0-1.0 license, which allows for free usage, modification, and distribution.

Datasets

The following datasets were used for training:

• Voxpopuli Dataset, with mainly Peninsular Spain accents
• Crowdsourced high-quality Spanish speech data:
  • Argentinian Spanish
  • Chilean Spanish
  • Colombian Spanish
  • Peruvian Spanish
  • Puerto Rican Spanish
  • Venezuelan Spanish
• TEDx Spanish Corpus

Additional sources:

• Crowdsourced high-quality Argentinian Spanish speech data set
• Crowdsourced high-quality Chilean Spanish speech data set
• Crowdsourced high-quality Colombian Spanish speech data set
• Crowdsourced high-quality Peruvian Spanish speech data set
• Crowdsourced high-quality Puerto Rico Spanish speech data set
• Crowdsourced high-quality Venezuelan Spanish speech data set
• • TEDx Spanish Corpus

Model Information

Base Model: SWivid/F5-TTS
Total Training Duration: 218 hours of audio
Training Configuration:

• Batch Size: 3200
• Max Samples: 64
• Training Steps: 1,200,000

Usage Instructions

Method 0: HuggingFace space (https://huggingface.co/spaces/jpgallegoar/Spanish-F5)

Method 1: Manual Model Replacement

1. Run the F5-TTS Application: Start the F5-TTS application and observe the terminal for output indicating the model file path. It should appear similar to:

   model : C:\Users\thega\.cache\huggingface\hub\models--SWivid--F5-TTS\snapshots\995ff41929c08ff968786b448a384330438b5cb6\F5TTS_Base\model_1200000.safetensors
   

2. Replace the Model File:

   • Navigate to the displayed file location.
   • Rename the existing model file to model_1200000.safetensors.bak.
   • Download model_1200000.safetensors from this repository and save it to the same location.

3. Restart the Application: Relaunch the F5-TTS application to load the updated model.

Alternative Methods

• GitHub Repository: Clone the Spanish-F5 repository and follow the provided installation instructions.
• Google Colab: Use the model via Google Colab.
  • Runtime -> Change Runtime Type -> T4 GPU
  • Runtime -> Run all
  • Click on the link shown in "Running on public URL: https://link.gradio.live" when it loads
• Jupyter Notebook: Run the model through the Spanish_F5.ipynb notebook.

Contributions and Recommendations

This model may benefit from further fine-tuning to enhance its performance across different Spanish dialects. Contributions from the community are encouraged. For optimal output quality, preprocess the reference audio by removing background noise, balancing audio levels, and enhancing clarity.