Release v0.3.2

README.md

@@ -8,15 +8,19 @@
   <img src="https://github.com/user-attachments/assets/1be5e31c-de42-40ab-9b85-790cb911ed47" alt="WhisperKit" width="20%" />
 </a>
 
-# WhisperKit Android (Beta)
+# WhisperKit Android
+
+[![Tests](https://github.com/argmaxinc/whisperkitandroid/actions/workflows/pr-checks.yml/badge.svg)](https://github.com/argmaxinc/whisperkitandroid/actions/workflows/pr-checks.yml)
+[![License](https://img.shields.io/github/license/argmaxinc/whisperkitandroid?logo=github&logoColor=969da4&label=License&labelColor=353a41&color=32d058)](LICENSE.md)
+[![Maven Central](https://img.shields.io/maven-central/v/com.argmaxinc/whisperkit?logo=sonatype&logoColor=969da4&label=Maven%20Central&labelColor=353a41&color=32d058)](https://central.sonatype.com/artifact/com.argmaxinc/whisperkit)
+[![Discord](https://img.shields.io/discord/1171912382512115722?style=flat&logo=discord&logoColor=969da4&label=Discord&labelColor=353a41&color=32d058&link=https%3A%2F%2Fdiscord.gg%2FG5F5GZGecC)](https://discord.gg/G5F5GZGecC)
 
-[![Maven Central](https://img.shields.io/maven-central/v/com.argmaxinc/whisperkit?color=32d058)](https://central.sonatype.com/artifact/com.argmaxinc/whisperkit)
 </div>
 
 WhisperKit Android brings Foundation Models On Device for Automatic Speech Recognition. It extends the performance and feature set of [WhisperKit](https://github.com/argmaxinc/WhisperKit) from Apple platforms to Android and Linux.  The current feature set is a subset of the iOS counterpart,
 but we are continuing to invest in Android and now welcome contributions from the community.
 
-[Example App (Coming Soon)] [[Blog Post]](https://takeargmax.com/blog/android) [[Python Tools Repo]](https://github.com/argmaxinc/whisperkittools)
+[[Example App]](https://play.google.com/store/apps/details?id=com.argmaxinc.whisperax) [[Blog Post]](https://takeargmax.com/blog/android) [[Python Tools Repo]](https://github.com/argmaxinc/whisperkittools)
 
 ## Table of Contents
 
@@ -37,7 +41,7 @@ To use WhisperKit in your Android app, you need to:
 ```kotlin
 dependencies {
    // 1. WhisperKit SDK
-   implementation("com.argmaxinc:whisperkit:0.3.0")  // Check badge above for latest version
+   implementation("com.argmaxinc:whisperkit:0.3.2")  // Check badge above for latest version
 
    // 2. QNN dependencies for hardware acceleration
    implementation("com.qualcomm.qnn:qnn-runtime:2.34.0")
@@ -73,18 +77,22 @@ class YourActivity : AppCompatActivity() {
       whisperKit = WhisperKit.Builder()
          .setModel(WhisperKit.OPENAI_TINY_EN)
          .setApplicationContext(applicationContext)
-         .setCallback { what, timestamp, msg ->
+         .setCallback { what, result ->
             // Handle transcription output
             when (what) {
                WhisperKit.TextOutputCallback.MSG_INIT -> {
                   // Model initialized successfully
                }
                WhisperKit.TextOutputCallback.MSG_TEXT_OUT -> {
                   // New transcription available
-                  val text = msg
-                  val time = timestamp
+                  val fullText = result.text
+                  val segments = result.segments
                   // Process the transcribed text as it becomes available
                   // This callback will be called multiple times as more audio is processed
+                  segments.forEach { segment ->
+                     // Process each segment
+                     val segmentText = segment.text
+                  }
                }
                WhisperKit.TextOutputCallback.MSG_CLOSE -> {
                   // Cleanup complete

android/config/detekt.yml

@@ -14,7 +14,7 @@ complexity:
     thresholdInObjects: 10
   LongParameterList:
     functionThreshold: 8
-    constructorThreshold: 7
+    constructorThreshold: 8
   CyclomaticComplexMethod:
     threshold: 20
   NestedBlockDepth:

android/examples/WhisperAX/build.gradle.kts

@@ -12,7 +12,7 @@ android {
         applicationId = "com.argmaxinc.whisperax"
         minSdk = 26
         targetSdk = 35
-        versionCode = 3
+        versionCode = 6
         versionName = "0.1.0"
 
         testInstrumentationRunner = "androidx.test.runner.AndroidJUnitRunner"

android/examples/WhisperAX/src/main/java/com/argmaxinc/whisperax/ComputeUnitsView.kt

@@ -31,6 +31,7 @@ import androidx.compose.material3.Surface
 import androidx.compose.material3.Text
 import androidx.compose.runtime.Composable
 import androidx.compose.runtime.collectAsState
+import androidx.compose.runtime.derivedStateOf
 import androidx.compose.runtime.getValue
 import androidx.compose.runtime.mutableStateOf
 import androidx.compose.runtime.remember
@@ -40,6 +41,7 @@ import androidx.compose.ui.Modifier
 import androidx.compose.ui.draw.alpha
 import androidx.compose.ui.draw.rotate
 import androidx.compose.ui.unit.dp
+import com.argmaxinc.whisperax.WhisperViewModel.Companion.MODELS_SUPPORTING_NPU
 import com.argmaxinc.whisperkit.ExperimentalWhisperKit
 import com.argmaxinc.whisperkit.WhisperKit
 
@@ -50,11 +52,18 @@ enum class ComputeUnits(val displayName: String, val backendValue: Int) {
     CPU_AND_NPU("NPU", WhisperKit.Builder.CPU_AND_NPU),
 }
 
+@OptIn(ExperimentalWhisperKit::class)
 @Composable
 fun ComputeUnitsView(viewModel: WhisperViewModel) {
     val modelState by viewModel.modelState.collectAsState()
     val encoderState by viewModel.encoderState.collectAsState()
     val decoderState by viewModel.decoderState.collectAsState()
+    val selectedModel by viewModel.selectedModel.collectAsState()
+    val shouldEnableNPUForEncoderDecoder by remember {
+        derivedStateOf {
+            selectedModel in MODELS_SUPPORTING_NPU
+        }
+    }
     val isEnabled = modelState == ModelState.LOADED || modelState == ModelState.UNLOADED
 
     var whisperKitExpanded by remember { mutableStateOf(true) }
@@ -75,6 +84,7 @@ fun ComputeUnitsView(viewModel: WhisperViewModel) {
                     currentState = encoderState,
                     currentUnit = viewModel.encoderComputeUnits.collectAsState().value,
                     onUnitSelected = { viewModel.setEncoderComputeUnits(it) },
+                    shouldEnableNPU = shouldEnableNPUForEncoderDecoder,
                     enabled = isEnabled,
                 )
 
@@ -85,6 +95,7 @@ fun ComputeUnitsView(viewModel: WhisperViewModel) {
                     currentState = decoderState,
                     currentUnit = viewModel.decoderComputeUnits.collectAsState().value,
                     onUnitSelected = { viewModel.setDecoderComputeUnits(it) },
+                    shouldEnableNPU = shouldEnableNPUForEncoderDecoder,
                     enabled = isEnabled,
                 )
             }
@@ -185,6 +196,7 @@ fun ComputeUnitRow(
     currentState: ModelState,
     currentUnit: ComputeUnits,
     onUnitSelected: (ComputeUnits) -> Unit,
+    shouldEnableNPU: Boolean = true,
     enabled: Boolean = true,
 ) {
     val infiniteTransition = rememberInfiniteTransition(label = "loading animation")
@@ -248,7 +260,11 @@ fun ComputeUnitRow(
                 expanded = expanded,
                 onDismissRequest = { expanded = false },
             ) {
-                ComputeUnits.values().forEach { unit ->
+                if (shouldEnableNPU) {
+                    listOf(ComputeUnits.CPU_ONLY, ComputeUnits.CPU_AND_GPU, ComputeUnits.CPU_AND_NPU)
+                } else {
+                    listOf(ComputeUnits.CPU_ONLY, ComputeUnits.CPU_AND_GPU)
+                }.forEach { unit ->
                     DropdownMenuItem(
                         text = { Text(unit.displayName) },
                         onClick = {

android/examples/WhisperAX/src/main/java/com/argmaxinc/whisperax/ModelSelectorView.kt

@@ -34,6 +34,7 @@ import androidx.compose.material3.Icon
 import androidx.compose.material3.IconButton
 import androidx.compose.material3.LinearProgressIndicator
 import androidx.compose.material3.MaterialTheme
+import androidx.compose.material3.MenuAnchorType
 import androidx.compose.material3.OutlinedTextField
 import androidx.compose.material3.Surface
 import androidx.compose.material3.Text
@@ -111,7 +112,8 @@ fun ModelSelectorView(viewModel: WhisperViewModel) {
                     },
                     modifier = Modifier
                         .fillMaxWidth()
-                        .weight(1f),
+                        .weight(1f)
+                        .menuAnchor(MenuAnchorType.PrimaryNotEditable),
                 )
 
                 ExposedDropdownMenu(

android/examples/WhisperAX/src/main/java/com/argmaxinc/whisperax/WhisperViewModel.kt

@@ -14,6 +14,8 @@ import androidx.compose.runtime.mutableStateListOf
 import androidx.lifecycle.ViewModel
 import androidx.lifecycle.viewModelScope
 import com.argmaxinc.whisperkit.ExperimentalWhisperKit
+import com.argmaxinc.whisperkit.TranscriptionResult
+import com.argmaxinc.whisperkit.TranscriptionSegment
 import com.argmaxinc.whisperkit.WhisperKit
 import com.argmaxinc.whisperkit.WhisperKit.TextOutputCallback
 import com.argmaxinc.whisperkit.WhisperKitException
@@ -33,22 +35,13 @@ import java.text.SimpleDateFormat
 import java.util.Date
 import java.util.Locale
 
-data class TranscriptionSegment(
-    val text: String,
-    val start: Float,
-    val end: Float,
-    val tokens: List<Int> = emptyList(),
-)
-
-data class TranscriptionResult(
-    val text: String = "",
-    val segments: List<TranscriptionSegment> = emptyList(),
-)
-
 @OptIn(ExperimentalWhisperKit::class)
 class WhisperViewModel : ViewModel() {
     companion object {
         const val TAG = "WhisperViewModel"
+
+        // Models currently supporting NPU backend, don't enable NPU for other models
+        val MODELS_SUPPORTING_NPU = listOf(WhisperKit.Builder.QUALCOMM_TINY_EN, WhisperKit.Builder.QUALCOMM_BASE_EN)
     }
 
     private lateinit var appContext: Context
@@ -190,25 +183,25 @@ class WhisperViewModel : ViewModel() {
         cacheDir = context.cacheDir.absolutePath
     }
 
-    fun onTextOutput(what: Int, timestamp: Float, msg: String) {
+    fun onTextOutput(what: Int, result: TranscriptionResult) {
+        val segments = result.segments
         when (what) {
             TextOutputCallback.MSG_INIT -> {
-                Log.i(MainActivity.TAG, "TFLite initialized: $msg")
+                Log.i(MainActivity.TAG, "TFLite initialized: ${result.text}")
                 startTime = System.currentTimeMillis()
                 _pipelineStart.value = startTime.toDouble() / 1000.0
                 _isInitializing.value = false
             }
 
             TextOutputCallback.MSG_TEXT_OUT -> {
                 Log.i(MainActivity.TAG, "TEXT OUT THREAD")
-                if (msg.isNotEmpty()) {
+                if (segments.isNotEmpty()) {
                     if (!firstTokenReceived) {
                         firstTokenReceived = true
                         firstTokenTimestamp = System.currentTimeMillis()
                         _firstTokenTime.value = (firstTokenTimestamp - startTime).toDouble() / 1000.0
                     }
-
-                    val newTokens = msg.length / 4
+                    val newTokens = segments.joinToString("") { it.text }.length / 4
                     totalTokens += newTokens
 
                     val currentTime = System.currentTimeMillis()
@@ -220,14 +213,14 @@ class WhisperViewModel : ViewModel() {
                     }
 
                     lastTokenTimestamp = currentTime
-                    updateTranscript(msg)
+                    updateTranscript(segments)
                 }
             }
 
             TextOutputCallback.MSG_CLOSE -> {
                 Log.i(MainActivity.TAG, "Transcription completed.")
-                if (msg.isNotEmpty()) {
-                    val newTokens = msg.length / 4
+                if (segments.isNotEmpty()) {
+                    val newTokens = segments.joinToString("") { it.text }.length / 4
                     totalTokens += newTokens
 
                     val totalTime = (System.currentTimeMillis() - startTime).toDouble() / 1000.0
@@ -236,8 +229,7 @@ class WhisperViewModel : ViewModel() {
 
                         updateRealtimeMetrics(totalTime)
                     }
-
-                    updateTranscript(msg)
+                    updateTranscript(segments)
                 }
             }
 
@@ -247,25 +239,8 @@ class WhisperViewModel : ViewModel() {
         }
     }
 
-    private fun updateTranscript(chunkText: String, withTimestamps: Boolean = false) {
-        var processedText = chunkText
-
-        val timestamps = if (withTimestamps) {
-            val timestampPattern = "<\\|(\\d+\\.\\d+)\\|>".toRegex()
-            val timestampMatches = timestampPattern.findAll(chunkText).toList()
-            timestampMatches.map { it.groupValues[1].toFloat() }
-        } else {
-            emptyList()
-        }
-
-        if (!withTimestamps) {
-            processedText = processedText
-                .replace("<\\|[^>]*\\|>".toRegex(), "")
-                .trim()
-        } else {
-            processedText = processedText.trim()
-        }
-
+    private fun updateTranscript(segments: List<TranscriptionSegment>) {
+        val processedText = segments.joinToString("") { it.text }
         if (processedText.isNotEmpty()) {
             if (allText.isNotEmpty()) {
                 allText.append("\n")
@@ -284,13 +259,12 @@ class WhisperViewModel : ViewModel() {
     fun listModels() {
         viewModelScope.launch {
             val modelDirs = listOf(
-                // TODO: enable when models are ready
-                // WhisperKit.Builder.OPENAI_TINY_EN,
-                // WhisperKit.Builder.OPENAI_BASE_EN,
-                // WhisperKit.Builder.OPENAI_SMALL_EN,
                 WhisperKit.Builder.QUALCOMM_TINY_EN,
                 WhisperKit.Builder.QUALCOMM_BASE_EN,
-                // WhisperKit.Builder.QUALCOMM_SMALL_EN
+                WhisperKit.Builder.OPENAI_TINY_EN,
+                WhisperKit.Builder.OPENAI_BASE_EN,
+                WhisperKit.Builder.OPENAI_TINY,
+                WhisperKit.Builder.OPENAI_BASE,
             )
             availableModels.clear()
             availableModels.addAll(modelDirs)
@@ -364,6 +338,21 @@ class WhisperViewModel : ViewModel() {
 
     fun selectModel(model: String) {
         _selectedModel.value = model
+        if (model in MODELS_SUPPORTING_NPU) {
+            _encoderComputeUnits.update {
+                ComputeUnits.CPU_AND_NPU
+            }
+            _decoderComputeUnits.update {
+                ComputeUnits.CPU_AND_NPU
+            }
+        } else {
+            _encoderComputeUnits.update {
+                ComputeUnits.CPU_ONLY
+            }
+            _decoderComputeUnits.update {
+                ComputeUnits.CPU_ONLY
+            }
+        }
         _modelState.value = ModelState.UNLOADED
         _encoderState.value = ModelState.UNLOADED
         _decoderState.value = ModelState.UNLOADED

android/whisperkit/build.gradle.kts

@@ -66,7 +66,7 @@ dependencies {
 
 mavenPublishing {
 
-    coordinates("com.argmaxinc", "whisperkit", "0.3.0")
+    coordinates("com.argmaxinc", "whisperkit", "0.3.2")
     pom {
         name.set("WhisperKit")
         description.set("On-device Speech Recognition for Android")

android/whisperkit/detekt-baseline.xml

@@ -2,8 +2,10 @@
 <SmellBaseline>
   <ManuallySuppressedIssues/>
   <CurrentIssues>
+    <ID>LargeClass:ArgmaxModelDownloaderImplTest.kt$ArgmaxModelDownloaderImplTest</ID>
     <ID>ThrowsCount:WhisperKit.kt$WhisperKit.Builder$@Throws(WhisperKitException::class) fun build(): WhisperKit</ID>
     <ID>TooGenericExceptionCaught:KtorHuggingFaceApiImpl.kt$KtorHuggingFaceApiImpl$e: Exception</ID>
     <ID>TooGenericExceptionCaught:WhisperKitImpl.kt$WhisperKitImpl$e: Exception</ID>
+    <ID>UnusedParameter:WhisperKitImpl.kt$WhisperKitImpl$timestamp: Float</ID>
   </CurrentIssues>
 </SmellBaseline>