Uganda-lang/OSAL

inference package for the OSAL models **OSAL** This is an extension api for the ease access of the models we've developed. All the models can be accessed on the huggingface hub ## Features - **High-Quality Synthesis**: Generates natural-sounding speech. - **Multi-Lingual Support**: Create and manage multiple TTS models for different languages simultaneously in the same application. - **High-Performance**: Fast C++ backend with multi-threading support. - **GPU Acceleration**: Offload model layers to the GPU for even faster performance. - **Offline Inference**: After the initial model download, no internet connection is required. ## Prerequisites To download the required TTS models, you need a **Hugging Face Hub API Token**. 1. If you don't have one, create a Hugging Face account here. 2. Navigate to your account settings and find the "Access Tokens" page, or go directly to huggingface.co. 3. Create a new token with at least `read` permissions. ## Installation You can install Orpheus TTS via pip: ```bash git clone github.com cd OSAL pip install -r requirements.txt ``` ## Quick Start: Multiple Language Instances We recommend storing your token as an environment variable for security. ```bash export HUGGING_FACE_TOKEN="hf_YourTokenGoesHere" ``` Now, you can run the following Python code to create both English and Luganda synthesizers: ```python import os from osal import tts # 1. Get Hugging Face Token --- hf_token = os.environ.get("HUGGING_FACE_TOKEN") if not hf_token: raise ValueError("Hugging Face token not found. Please set the HUGGING_FACE_TOKEN environment variable.") # 2. Create TTS Instances for Different Languages --- # This will download the models on the first run. print("Initializing English TTS model...") tts_en = tts.OrpheusCpp(huggingface_token=hf_token, lang="en") # Other languages print("\nInitializing Luganda TTS model with GPU acceleration...") tts_lug = tts.OrpheusCpp( huggingface_token=hf_token, lang="lug", n_gp …