What is this project about?

This project focuses on building a high-quality voice recording and synthesis workflow, enabling realistic text-to-speech (TTS) experiences using Sinhala and Latin-script inputs. It involves studio-grade voice capture, neural network modeling, and accessible feedback interfaces.

Who is the target audience?

Cooperates and government institutes interested in TTS generation

Voice artists and linguists involved in dataset creation

AI researchers and developers working with speech synthesis

End users of voice assistant and accessibility technologies

Product teams deploying Sinhala TTS in mobile or cloud apps

What is the recording setup used? 

We utilize a professional voice studio with:

• Directional condenser microphones

• Acoustic treatment for clarity

• Digital tablets for script delivery

• Clean and modern lighting for ease of use and comfort

How does the transcription process work? 

Audio captured from the voice artist is transcribed into aligned text using a digital interface. Transcripts and waveform data are visualized and verified through structured metadata, ensuring high accuracy. 

What is the neural network model used for? 

The model transforms Latin input text into a Mel Spectrogram, which is then converted into an audio waveform. This allows seamless voice synthesis while preserving pronunciation and intonation.

How is quality controlled?

A feedback interface lets listeners:

• Rate voice output (e.g., using star ratings)

• Submit written comments

• Help train and fine-tune models for better performance

Can this system run on low-spec devices? 

Yes. The generated voice models are optimized for cloud, mobile, and edge deployment. A sample smartphone interface shows the TTS output, with branding and visual cues consistent with SDT (Susinidu's Digital Technology) design language. 

What languages are supported?

Currently, the system supports Sinhala and Latin-based Sinhala inputs, with room for expansion into other local dialects and languages. Sinhala unicode will be mapped to these syllables for better alignment with modern use of Sinhala.

Is the voice artist’s data secure? 

Absolutely. All recordings, transcripts, and feedback data are managed under strict privacy policies and are only used for training under agreed-upon consent. 

How can I contribute or collaborate?  

You can:

• Provide annotated recordings

• Join the feedback loop

• Collaborate as a research or product partner

• Contact the project team for API or deployment options