Add standalone notebook for TTS inference (Python 3.10+ / Coqui-TTS v0.22)

[osinkolu]

Summary

I've added a standalone Jupyter Notebook (Digital_Umuganda_TTS_Inference_Tutorial.ipynb) that allows users to run the Kinyarwanda TTS model in modern environments, such as Google Colab or local Python 3.10+ setups.

Problem

The current repository setup relies on older dependencies (specifically a 2021 commit of coqui-tts) that do not compile easily on newer Python versions. Running the model with the latest coqui-tts library normally causes crashes due to:

1. Config Mismatches: Missing keys like output_sample_rate and phoneme_language.
2. Path Errors: Hardcoded local paths in the original config.json.
3. Runtime Errors: A logic conflict in newer library versions where the speaker embedding is dropped if use_d_vector_file is set to True during inference.

Solution

This notebook provides a "one-click" solution that:

• Automates Setup: Handles git lfs pull and installs modern dependencies.
• In-Memory Patching: Fixes the configuration errors dynamically without modifying the existing config.json on disk (preserving the original repo structure).
• Runtime Fix: Includes a runtime patch to force the model to accept speaker embeddings, resolving the crash in modern Coqui-TTS versions.

Tested successfully on Google Colab.