Get Supertonic for Android

Supertonic is a Text-to-Speech (TTS) engine available across multiple platforms. This repository is for the Systemwide TTS implementation on Android specifically, supporting most/all ARM ABIs on Play Store and GitHub releases APKs. On F-Droid release currently only arm64-v8a is implemented.

Please note that functionality and bug fixes may vary slightly between the F-Droid and Play Store builds as there can be version difference as they get submitted and approved following different timetables.

Download Options

• Stable Release: Available now on F-Droid
• Stable Release: Available on Google Play Store

---

Note: This repository currently tracks both versions of the application.

---

Adding or Improving Language Support (Submit a PR)

We welcome community contributions to support new languages or improve existing ones. The text-processing pipeline works as follows:

1. Kotlin Text Normalizer: Expands shorthand notations (numbers, currency, percentages, ranges) into full spoken words (in the target language).
2. JNI / Rust Chunker: Receives the normalized text and splits it into optimal sentences and audio chunks (maximum 300 characters, or 120 for CJK languages).

To add or improve support for a language, follow these steps:

1. High-Level Text Normalization (Kotlin)

Text normalization prevents the engine from reading symbols literally (e.g. pronouncing "3" as English "three" instead of Hindi "तीन").

• Path: app/src/main/java/com/brahmadeo/supertonic/tts/utils/TextNormalizer.kt
• Action: Add a routing check for your language prefix inside the normalize function, and implement your language-specific normalization logic:

// Inside TextNormalizer.normalize()
if (lowerLang.startsWith("hi")) {
    return normalizeHindi(processedText)
}

// Example: Implementation of normalizeHindi
private fun normalizeHindi(text: String): String {
    // 1. Convert native scripts digits to latin digits (e.g. ०-९ -> 0-9)
    var normalized = convertDevanagariDigits(text)
    
    // 2. Format ranges (e.g. "10-15" -> "10 से 15")
    val rangePattern = Pattern.compile("\\b(\\d+)\\s*[-–—]\\s*(\\d+)\\b")
    // Replace logic ...

    // 3. Format currency (e.g. "₹500" -> "500 रुपये")
    val currencyPattern = Pattern.compile("(?:\\bINR|₹)\\s*(\\d+(?:\\.\\d+)?)\\b")
    // Replace logic ...

    // 4. Convert remaining digits to words using NumberUtils
    val numberPattern = Pattern.compile("\\b(\\d+(?:\\.\\d+)?)\\b")
    // Replace each match using NumberUtils.convertHindi or NumberUtils.convertHindiDouble
    return normalized
}

2. Number to Word Expansion (Kotlin)

• Path: app/src/main/java/com/brahmadeo/supertonic/tts/utils/NumberUtils.kt
• Action: Implement utility functions to convert integer and decimal values into spoken words:

// Example signature:
fun convertHindi(n: Long): String
fun convertHindiDouble(d: Double): String

3. Sentence Splitting and Chunking (Rust)

The JNI layer delegates text chunking to Rust, ensuring the display UI in PlaybackActivity matches the underlying audio chunks.

• Paths:
  • rust/src/lang/mod.rs (Language Normalizer Registry)
  • rust/src/lang/<lang_code>.rs (Language normalizer implementation)
  • rust/src/lang/configs/<lang_code>.json (JSON config for abbreviations)
• Action:
  1. Add your language configuration JSON containing punctuation splits, abbreviations, and rules.
  2. Implement the LanguageNormalizer trait for your language:

     pub struct HindiNormalizer;
     impl LanguageNormalizer for HindiNormalizer {
         fn preprocess(&self, text: &str) -> String { ... }
         fn split_sentences(&self, text: &str) -> Vec<String> { ... }
         fn max_chunk_len(&self) -> usize { 300 }
         fn should_wrap_tags(&self) -> bool { false }
     }

  3. Register your normalizer in the get_normalizer factory function in rust/src/lang/mod.rs.

4. Language Selection UI and Resources (Kotlin)

• Paths:
  • app/src/main/res/values/strings.xml
  • app/src/main/java/com/brahmadeo/supertonic/tts/MainActivity.kt
• Action:
  1. Declare the language display string in strings.xml:

     <string name="lang_hindi">Hindi</string>

  2. Add the string ID and language code mapping to the languages map in MainActivity.kt:

     R.string.lang_hindi to "hi"

5. Writing and Running Tests

Always write tests for the normalization expansions and chunking rules to prevent regressions.

• Kotlin Tests Path: app/src/test/java/com/brahmadeo/supertonic/tts/utils/TextNormalizerTest.kt
• Run Rust Tests:

  cd rust
  cargo test

• Run Kotlin Tests:

  .\gradlew.bat testDebugUnitTest

---

Credits

Supertonic for creating lightweight and great-sounding TTS models for edge compute.