Developing a Bidirectional English-Idoma Speech-to-Text (STT) and Text-to-Speech (TTS) System for Enhanced Communication and Language Preservation
đź“– **Abstract**
Most speech technology exists for high-resource languages, leaving low-resource languages like Idoma digitally underrepresented. This project bridges that gap by developing a fully functional Bidirectional English-Idoma Speech System.
The system utilizes an Applied Research Design to solve data scarcity and tonal complexity issues inherent in the Idoma language. It integrates three fine-tuned transformer models to enable natural conversation flow:
- **ASR (Automatic Speech Recognition):** Fine-tuned Wav2Vec 2.0 XLS-R.
- **NMT (Neural Machine Translation):** Fine-tuned NLLB-200.
- **TTS (Text-to-Speech):** Fine-tuned VITS-based MMS-TTS.
**Note: This repository contains just the modularized version of the inference code found in the Jupyter Notebooks. For full implementation codes, check the Juypter Notebooks.**
**Jupyter Notebooks**
- wav2vec2-xls-r-1b-finetuned-idoma
- idoma-mms-tts-eng
- idu-eng-translator
- Inference code for the bidirectional TTS, STT and NMT
🏗️ **System Architecture**
The application follows a Service-Oriented Architecture (SOA) where the UI is decoupled from the inference logic. The processing pipeline for English → Idoma (and vice-versa) follows this flow:
```mermaid
graph LR
A[Input Audio] --> B[ASR Service]
B --> C[NMT Service]
C --> D[TTS Service]
D --> E[Synthesized Speech]
```
**Core Components**
- **Speech-to-Text (STT):** Converts input audio to text using Wav2Vec2 (Idoma) and Whisper Large v3 (English).
- **Machine Translation (NMT):** Translates text between English and Idoma using NLLB-200.
- **Text-to-Speech (TTS):** Synthesizes the translated text into speech using VITS (Idoma) and SpeechT5 (English).
📊 **Performance Results**
The system was rigorously tested against a held-out test set. Below are the quantitative evaluation metrics reported in the thesis:
| Component | Model Architecture | Metric | Score |
| --------------|---------------------------|-------------- …