Open Speech 🎙️

Open Speech is an open-source AI English speaking coach Android app built with native Jetpack Compose and powered by Google Gemini AI. Grounded in peer-reviewed Second Language Acquisition (SLA) research, it moves beyond simple scoring to actively guide language learners toward authentic fluency.

🌐 Try the Live Web Demo: https://sunnydev07.github.io/Open-Speech/
📥 Download APK: Direct Download (v1.1 • 23.4 MB) | GitHub Releases

---

📱 Screenshots (Native Pixel 8 Compose Renders)

🏠 Dashboard	🎙️ Live Recording	⚡ AI Analyzing	🏆 Fluency Results	🔁 "Fix & Re-Say" Drill

---

🔬 Grounded in Second Language Acquisition (SLA) Science

Most language learning apps only provide a generic score. Open Speech closes the cognitive learning loop by implementing methods validated by modern SLA literature:

1. Noticing Hypothesis (Schmidt 1990): Learners cannot correct phonetic deviations they do not perceive. With in-app audio playback, learners listen back to their own recording immediately after speaking with clickable timeline seeking and speed controls.
2. Task Repetition with Delta Tracking (Zhang 2023): Immediate re-attempts on the same speaking prompt show marked gains in articulation rate, lexical access, and filler reduction. Open Speech tracks attempt-to-attempt progress with side-by-side metric comparisons (Attempt 2 of 2: +6 pts, +14 WPM, -2 pauses).
3. Metacognitive Self-Assessment (Dörnyei 2005 & Oxford 1990): Before seeing AI scores, learners rate their own flow, clarity, and confidence. The app then calculates their Calibration Gap, reassuring anxious speakers who underestimate their ability and pinpointing blind spots.
4. Target Acoustic Modeling (Foote & McDonough 2017): Hearing target acoustic models directly after an error is flagged accelerates phoneme accuracy. Each phonetic tip includes dual-speed native TTS models (1.0x natural cadence and 0.7x slow articulation).
5. "Say It Again" Form-Focused Re-Drills (Lyster & Saito 2010): Rather than overwhelming learners with ten generic suggestions, the app extracts exact sentence diffs and launches a focused 30-second micro-drill on a single corrected sentence.

---

✨ Key Features

• ⏱️ Flexible Timed Speaking Sessions: Choose presets of 30s, 60s, 120s, or Free-Flow with live countdown rings and waveform visualizers.
• 📚 Adaptive Prompt Library: 20+ prompts categorized across CEFR levels (A2 to C1), covering interview scenarios, opinion framing, storytelling, and professional triage.
• 🤖 Gemini AI Evaluation: Strict-JSON parsing evaluating:
  • Overall Fluency Score (0–100) & CEFR level mapping (A1–C2).
  • Speech Rate in Words Per Minute (WPM with 120–150 target thresholds).
  • Hesitation pauses (>1s) and filler word frequency (um, uh, like).
  • Grammar and syntax accuracy percentage.
• 🗣️ Pronunciation & Accent Profiling:
  • Detected accent cadence (e.g., General American).
  • Phonetic tips complete with IPA spelling, syllable stress guidelines, and audio reference models.
• 🔁 Sentence Corrections & Re-Drill Loop:
  • Highlights Original Sentence vs. Corrected Sentence with inline word-level diffs.
  • One-tap "Practice this fix" micro-session that links back to the parent session in the database.
• 🔒 Local-First Privacy & Persistence (Room DB v3):
  • All session records, streaks, and daily targets are stored on-device using SQLite/Room.
  • Zero login required; no personal identification, email, or third-party tracking.
• 📤 Share Fluency Report: Native Android share sheet integration to celebrate practice milestones and streaks with peers or mentors.
• 📱 Responsive Web Companion: Test the app directly in your desktop or mobile browser via our GitHub Pages simulator at sunnydev07.github.io/Open-Speech.

---

🏗️ Architecture & Tech Stack

Open-Speech/
├── app/src/main/java/com/example/
│   ├── MainActivity.kt                  # AppState machine, SpeechViewModel & Composable screens
│   ├── ai/
│   │   ├── GeminiPronunciationService.kt # Gemini 2.5 Flash REST client, prompt engineering & parsers
│   │   ├── PromptLibrary.kt             # CEFR-graded prompt rotation library
│   │   └── CefrMapper.kt                # Scientific rubric mapping fluency metrics to CEFR bands
│   ├── audio/
│   │   └── AudioRecorderManager.kt      # MediaRecorder (AAC/MPEG-4, 44.1kHz) & amplitude StateFlow
│   ├── data/
│   │   ├── SessionEntity.kt             # Room DB schema with parent-child drill hierarchy
│   │   └── AppDatabase.kt               # Room database instance (v3)
│   ├── ui/
│   │   ├── components/
│   │   │   ├── AudioPlaybackCard.kt     # In-app recording playback with slider seek
│   │   │   ├── ImprovementComparisonCard.kt # Side-by-side prompt repetition deltas
│   │   │   ├── SelfAssessmentComponent.kt # Metacognitive 3-star rating & calibration card
│   │   │   ├── RedrillCard.kt           # "Fix & re-say" sentence corrections & drill target banner
│   │   │   ├── PronunciationAccentCard.kt # Dual score meters & TTS phonetic models
│   │   │   ├── DailyGoalComponent.kt    # Daily target tracking & streak calculation
│   │   │   └── MilestoneBadgeComponent.kt # Gamified milestones & consistency badges
│   │   ├── effects/
│   │   │   └── EffectComponents.kt      # Canvas waveform audio visualizer & subtle clicks
│   │   └── theme/                       # Modern Material3 Dark/Light typography & palettes
│   └── util/
│       ├── TtsHelper.kt                 # Android TextToSpeech engine with dual-speed playback
│       ├── WordDiff.kt                  # Word-level diff highlighter for grammar corrections
│       ├── ShareProgressHelper.kt       # Native Android share sheet report generator
│       ├── StreakCalculator.kt          # Calendar-date streak computation
│       └── HapticFeedbackHelper.kt      # Tactile haptic feedback
├── preview/                             # GitHub Pages interactive site & prebuilt APK
└── .github/workflows/pages.yml          # Automated CI/CD Pages deployment

Core Technologies

• UI Framework: Jetpack Compose with Material 3
• Language: Kotlin (100%) + Coroutines + StateFlow
• Database: Android Room 2.6 with parent-child relational sessions
• AI Processing: Google Gemini 2.5 Flash API with strict JSON schema response mode
• Audio Engine: Android MediaRecorder + MediaPlayer + TextToSpeech
• Testing: JUnit 4 + Robolectric + Roborazzi Compose screenshot tests

---

🚀 Getting Started

Prerequisites

• Android Studio Koala (2024.1.1) or newer
• JDK 17
• Android SDK 35 / 36 (Minimum SDK: 24 / Android 7.0+)

Setup Instructions

1. Clone the repository:

   git clone https://github.com/sunnydev07/Open-Speech.git
   cd Open-Speech

2. Configure Gemini API Key: Copy .env.example to .env in the root folder:

   cp .env.example .env

   Add your Google Gemini API key:

   GEMINI_API_KEY=your_actual_gemini_api_key_here

   (Note: If no API key is provided, the app runs in Diagnostic Demo Mode with realistic templated feedback).

3. Build and Run:

   • Open the project in Android Studio.
   • Select an emulator or physical device.
   • Click Run or use the Gradle wrapper:

     ./gradlew assembleDebug

4. Run Unit & Screenshot Tests:

   ./gradlew testDebugUnitTest

---

📦 Install Prebuilt APK

You can install the app directly on your Android phone without building from source:

adb install preview/app-debug.apk

Or download it directly from your mobile browser:
👉 Download Latest APK (v1.1)

---

🗺️ Fluency Roadmap

• In-App Audio Playback: Listen back to your recorded voice on the results screen (Schmidt 1990).
• Task Repetition & Comparison: Retry prompts with side-by-side metric deltas (Zhang 2023).
• Metacognitive Self-Assessment: Pre-result confidence calibration (Dörnyei 2005).
• Target Phonetic Acoustic Modeling: Normal & 0.7x slow native TTS drills (Foote 2017).
• Form-Focused Sentence Corrections: "Fix & re-say" micro-drills with word diffs (Lyster 2010).
• Local-First SQLite/Room DB: Persistent session logs, streaks, and progress.
• Interactive Conversation Mode: Two-way streaming turn-taking using Gemini Live audio.
• IELTS / TOEFL Tracks: Official band-descriptor rubrics for Part 1/2/3 tasks.
• Listen-and-Repeat / Shadowing Mode: Word-level pronunciation alignment.

---

🤝 Contributing

Contributions are always welcome! Whether it's reporting a bug, suggesting a new speaking exercise, or adding a new SLA feature:

1. Fork the repository.
2. Create a feature branch: git checkout -b feature/amazing-feature.
3. Commit your changes: git commit -m "feat: add amazing feature".
4. Push to your branch: git push origin feature/amazing-feature.
5. Open a Pull Request.

---

📄 License

This project is open-source software licensed under the MIT License.

Copyright (c) 2026 Sunny Kumar Dev