Demo App v3.0.1 - Compatibility Fix

WhisperGF Demo App v3.0.1

Compatibility Fix

• Fixed: supports-any-density: false - now supports all screen densities
• Fixed: Screen size restricted to normal - now supports small/normal/large/xlarge
• Added: armeabi-v7a (32-bit ARM) support alongside arm64-v8a
• Fixed: extractNativeLibs=true for proper native library extraction
• Added: uses-feature-not-required for microphone and touchscreen (tablet support)

Supported Devices

• ARM64-v8a (64-bit): Modern Android phones (Android 8.0+)
• armeabi-v7a (32-bit): Older/budget Android phones (Android 8.0+)
• All screen sizes and densities

How to Install

1. Download whispergf-test.apk
2. Enable "Install from unknown sources" in Settings
3. Open the APK to install
4. Grant microphone permission when prompted
5. Download a model (or pick a GGUF file) and start transcribing!

v3.0.0 — IQ-quants, Diarize, Speculative Decoding, Thermal

whisperGF v3.0.0

تقليل استهلاك الذاكرة (RAM)

• IQ-quants (IQ2_XXS, IQ3_S, IQ4_XS, TQ1_0, TQ2_0) with imatrix
• KV Cache Quantization (INT8/INT4)
• Model Sharding (gguf-split) auto-handled
• K-quants auto-detect from GGUF

تسريع الاستدلال (Inference Speed)

• Speculative Decoding (draft + verify, 2x speedup)
• Qualcomm Hexagon HTP/QNN backend
• Thermal-Aware Threading
• Flash Attention enabled by default

تحسين تجربة المطور (DX)

• Tinydiarize (speaker diarization)
• Integrated Model Downloader (HuggingFace)
• Stateless VAD Reset
• RAII wrappers for JNI

البناء

• ABI: arm64-v8a only
• Size: 5.3 MB

whisperGF v2.0.0

whisperGF v2.0.0 - Major Update

New Features

Feature	Description
Quantized Models	Auto-detect q4_0, q5_0, q8_0 - up to 45% less RAM
VAD	Voice Activity Detection - skip silence, faster processing
Context Size	Configurable n_max_text_ctx (default 448, reduce to 256 for low RAM)
Thread Control	User-controllable with smart defaults
Streaming	Chunked processing with transcribeStream() - ~40% less peak RAM
x86_64 ABI	Emulator and x86 device support

API Highlights

// Advanced transcription with full control
val config = TranscribeConfig(
    numThreads = 4,
    maxTextCtx = 256,
    language = "ar",
    enableVad = true,
    vadThreshold = 0.4f
)

val ctx = WhisperContext.createContextFromFile("model-q4_0.bin")
val info = ctx.getModelInfo()
println("Quantization: ${info.quantizationName}")

val result = ctx.transcribeStream(audioData, config)
for (seg in result.segments) {
    println("[${seg.t0Ms}ms -> ${seg.t1Ms}ms] ${seg.text}")
}
ctx.release()

Files

• whisper-android.aar - Android Library (12 MB, arm64-v8a + x86_64)
• whisper-android.zip - Compressed archive

whisperGF v1.0.0

whisperGF v1.0.0 - Initial Release

whisper.cpp Android AAR Library for Speech-to-Text

Features:

• Speech-to-Text on Android (arm64-v8a)
• ARMv8.2 FP16 optimized variant
• Kotlin API with coroutines support
• Model loading from file, assets, or InputStream

Files:

• whisper-android.aar - Android Library
• whisper-android.zip - Compressed archive

Requirements:

• Min SDK: 26 (Android 8.0+)
• Kotlin 1.9.0+
• kotlinx-coroutines-android 1.7.1+