Text-to-Speech (TTS)

The Text-to-Speech (TTS) component is responsible for converting the final Igbo text response into an audio file that can be played back to the user. This is the last step in generating the voice assistant's reply.

Process

1. Text Input: Receives the translated Igbo text from the Translation component.
2. Model Inference: The Igbo text is fed into a TTS model specifically trained or fine-tuned for the Igbo language. The model generates audio data (e.g., WAV, MP3).
3. Audio Output: The generated audio data is saved as an audio file.
4. Link Generation: A link to the generated audio file is created and sent back to the frontend.

Technology Considerations

• Model Choice: We will explore pre-trained TTS models available on platforms like Hugging Face that support Igbo speech synthesis. The user specifically mentioned YarnGPT as a potential candidate.
• Integration: Transformers.js will be used to interact with the chosen TTS model, allowing for seamless integration within the Hono.js backend.
• Voice Quality: The selection of TTS models will prioritize natural-sounding speech with clear pronunciation and appropriate intonation for the Igbo language.
• Audio Format: The generated audio files will be in a widely supported format (e.g., MP3, WAV) for easy playback on the frontend.
• Storage: The generated audio files will need to be stored temporarily or permanently, depending on the requirements. A simple file storage system or a cloud storage solution could be used.