Nimalesh/nlp_project_tts

Building Text to speech model for Swahili(East African language) # NLP Project: Text-to-Speech (TTS) for Swahili using tacotron2 ## Overview This project builds a **Text-to-Speech (TTS) model for Swahili**, an East African language. ## Getting Started ### 1. Install Dependencies Before running the project, install the required dependencies by running: ```bash pip install -r requirements.txt ``` ### 2. Dataset Preprocessing Download the dataset from drive.google.com and place it in your project directory. Then, run the **NLP_TTS_preprocessing.py** script to perform basic preprocessing on the dataset. ```bash NLP_TTS_preprocessing.ipynb ``` This script processes the dataset and generates **tts_dataset.csv**, which will be used for training the model. ### 3. Train the Model Once preprocessing is complete, proceed to ```bash nlp_project_model.ipynb ``` **Update the dataset path** in the notebook to point to the generated **tts_dataset.csv** file. Since speech synthesis models require significant training, **set the number of epochs to 100** for better performance. Run the notebook to train the model. ### 4. Inference After training, you can perform inference using the saved model. The model takes a Swahili text prompt as input and generates a **Mel spectrogram**, which is then converted into an audio file. The generated audio will be saved automatically. ## Notes - Use **Google Colab** to get results - Ensure that the dataset path is correctly updated in the notebook before training. - Training on **100 epochs** is recommended for optimal results. - The output audio files will be stored in the specified directory after inference. ## Thank you