fix(ios): make feedStreamingAudio work and move to FluidAudio 0.18

README.md

@@ -12,15 +12,17 @@ React Native wrapper for [FluidAudio](https://github.com/FluidInference/FluidAud
 ## Features
 
 - **ASR (Automatic Speech Recognition)** - High-quality speech-to-text using Parakeet TDT models
-- **Streaming ASR** - Real-time transcription from microphone or system audio
+- **Streaming ASR** - Real-time transcription from the microphone or from audio you feed in
 - **VAD (Voice Activity Detection)** - Detect speech segments in audio
 - **Speaker Diarization** - Identify and track different speakers
 - **TTS (Text-to-Speech)** - Natural voice synthesis using Kokoro TTS
 
 ## Requirements
 
 - iOS 17.0+
-- React Native 0.71+ or Expo SDK 50+
+- React Native 0.75+ or Expo SDK 52+
+
+FluidAudio (0.18.0 and later 0.18.x releases) is pulled from GitHub through Swift Package Manager with React Native's `spm_dependency` podspec helper, which needs React Native 0.75. CocoaPods trunk only has FluidAudio 0.7.8 and goes read-only on December 2, 2026.
 
 ## Installation
 
@@ -42,6 +44,8 @@ npm install @fluidinference/react-native-fluidaudio
 cd ios && pod install
 ```
 
+Remove any `pod 'FluidAudio'` line from your Podfile: the package brings FluidAudio in itself. If the app fails to link with static pods, install with `USE_FRAMEWORKS=dynamic pod install` (see [react-native#44627](https://github.com/facebook/react-native/pull/44627#issuecomment-2123119711)).
+
 ## Usage
 
 ### Basic Transcription
@@ -75,20 +79,24 @@ import { StreamingASRManager, onStreamingUpdate } from '@fluidinference/react-na
 
 const streaming = new StreamingASRManager();
 
-// Start streaming with update callback
+// Transcribe the microphone (the module records it)
 await streaming.start({ source: 'microphone' }, (update) => {
-  console.log('Confirmed:', update.confirmed);
-  console.log('Volatile:', update.volatile);
+  console.log('Transcript so far:', update.confirmed);
 });
 
-// Feed audio data (16-bit PCM, 16kHz, base64 encoded)
-await streaming.feedAudio(base64AudioChunk);
+// ...or transcribe audio you supply: little-endian 16-bit mono PCM at 16 kHz
+await streaming.start({ source: 'manual' }, (update) => {
+  console.log('Transcript so far:', update.confirmed);
+});
+await streaming.feedAudio(base64AudioChunk); // or feedBuffer(arrayBuffer)
 
-// Stop and get final result
+// Stop: queued audio is transcribed first, then the final text is returned
 const result = await streaming.stop();
 console.log('Final transcription:', result.text);
 ```
 
+`feedAudio` / `feedBuffer` only work in a `'manual'` session; in a microphone session they reject. The default model is Parakeet EOU 320 ms (English); pick another with `model`, e.g. `{ source: 'manual', model: 'parakeet-eou-160ms' }`. The first start downloads and compiles the model.
+
 ### Voice Activity Detection
 
 ```typescript

__tests__/FluidAudio.test.ts

@@ -115,12 +115,20 @@ describe('FluidAudio', () => {
     });
 
     it('should feed audio while streaming', async () => {
-      await streaming.start();
+      await streaming.start({ source: 'manual' });
       await streaming.feedAudio('base64chunk');
 
       expect(mockModule.feedStreamingAudio).toHaveBeenCalledWith('base64chunk');
     });
 
+    it('should feed an ArrayBuffer as base64', async () => {
+      await streaming.start({ source: 'manual' });
+      // Two 16-bit samples: 1 and -1, little-endian.
+      await streaming.feedBuffer(new Uint8Array([0x01, 0x00, 0xff, 0xff]).buffer);
+
+      expect(mockModule.feedStreamingAudio).toHaveBeenCalledWith('AQD//w==');
+    });
+
     it('should throw when feeding audio without starting', async () => {
       await expect(streaming.feedAudio('base64chunk')).rejects.toThrow(
         'Streaming not started'

__tests__/types.test.ts

@@ -94,18 +94,21 @@ describe('Type Definitions', () => {
       expect(config.source).toBe('microphone');
     });
 
-    it('should support system audio source', () => {
+    it('should support a manual source and a model choice', () => {
       const config: StreamingASRConfig = {
-        source: 'system',
+        source: 'manual',
+        model: 'parakeet-eou-160ms',
       };
 
-      expect(config.source).toBe('system');
+      expect(config.source).toBe('manual');
+      expect(config.model).toBe('parakeet-eou-160ms');
     });
 
     it('should define StreamingUpdate correctly', () => {
       const update: StreamingUpdate = {
-        volatile: 'partial',
+        volatile: '',
         confirmed: 'full',
+        text: 'full',
         isFinal: false,
       };
 

example/App.tsx

@@ -14,6 +14,7 @@ import {
   View,
   ActivityIndicator,
   TextInput,
+  Image,
 } from 'react-native';
 
 import {
@@ -30,6 +31,45 @@ import {
   type StreamingUpdate,
 } from 'react-native-fluidaudio';
 
+/** The PCM data of a 16 kHz mono 16-bit WAV file. */
+function pcm16FromWav(wav: ArrayBuffer): ArrayBuffer {
+  const view = new DataView(wav);
+  const tag = (offset: number) =>
+    String.fromCharCode(
+      view.getUint8(offset),
+      view.getUint8(offset + 1),
+      view.getUint8(offset + 2),
+      view.getUint8(offset + 3)
+    );
+  if (view.byteLength < 12 || tag(0) !== 'RIFF' || tag(8) !== 'WAVE') {
+    throw new Error('Not a WAV file');
+  }
+  let formatOk = false;
+  let offset = 12;
+  while (offset + 8 <= view.byteLength) {
+    const id = tag(offset);
+    const size = view.getUint32(offset + 4, true);
+    const body = offset + 8;
+    if (id === 'fmt ') {
+      formatOk =
+        view.getUint16(body, true) === 1 && // PCM
+        view.getUint16(body + 2, true) === 1 && // mono
+        view.getUint32(body + 4, true) === 16000 &&
+        view.getUint16(body + 14, true) === 16;
+    } else if (id === 'data') {
+      if (!formatOk) {
+        throw new Error('Expected 16 kHz mono 16-bit PCM');
+      }
+      return wav.slice(body, body + size);
+    }
+    offset = body + size + (size % 2);
+  }
+  throw new Error('WAV file has no data chunk');
+}
+
+/** 320 ms of 16 kHz 16-bit audio per feedBuffer call. */
+const FEED_CHUNK_BYTES = 16000 * 2 * 0.32;
+
 const App = () => {
   const [systemInfo, setSystemInfo] = useState<SystemInfo | null>(null);
   const [loading, setLoading] = useState(false);
@@ -129,6 +169,36 @@ const App = () => {
     }
   }, [streaming]);
 
+  // Stream the bundled sample (FluidAudio test fixture, 16.9 s) through feedBuffer,
+  // to check streaming transcription without a microphone.
+  const streamSampleFile = useCallback(async () => {
+    setLoading(true);
+    setResult('');
+    setStreamingText('');
+    setConfirmedText('');
+    try {
+      const asset = Image.resolveAssetSource(require('./assets/speech-16k-mono.wav'));
+      const response = await fetch(asset.uri);
+      const pcm = pcm16FromWav(await response.arrayBuffer());
+
+      await streaming.start({ source: 'manual' }, handleStreamingUpdate);
+      setIsStreaming(true);
+      for (let offset = 0; offset < pcm.byteLength; offset += FEED_CHUNK_BYTES) {
+        await streaming.feedBuffer(pcm.slice(offset, offset + FEED_CHUNK_BYTES));
+      }
+      const stopResult = await streaming.stop();
+      setResult(`File transcription:\n${stopResult.text || '(no speech detected)'}`);
+    } catch (error) {
+      if (streaming.isStreaming()) {
+        await streaming.stop().catch(() => undefined);
+      }
+      setResult(`Error: ${(error as Error).message}`);
+    } finally {
+      setIsStreaming(false);
+      setLoading(false);
+    }
+  }, [streaming, handleStreamingUpdate]);
+
   // ============================================================================
   // VAD Functions
   // ============================================================================
@@ -247,15 +317,26 @@ const App = () => {
         {/* Streaming ASR Section */}
         <View style={styles.section}>
           <Text style={styles.sectionTitle}>Streaming Speech-to-Text</Text>
-          <Text style={styles.sectionDesc}>Real-time transcription from microphone</Text>
+          <Text style={styles.sectionDesc}>
+            Real-time transcription from the microphone, or from a bundled file fed in chunks
+          </Text>
           {!isStreaming ? (
-            <TouchableOpacity
-              style={[styles.button, styles.buttonPrimary, loading && styles.buttonDisabled]}
-              onPress={startStreaming}
-              disabled={loading}
-            >
-              <Text style={styles.buttonText}>🎤 Start Streaming</Text>
-            </TouchableOpacity>
+            <>
+              <TouchableOpacity
+                style={[styles.button, styles.buttonPrimary, loading && styles.buttonDisabled]}
+                onPress={startStreaming}
+                disabled={loading}
+              >
+                <Text style={styles.buttonText}>🎤 Start Streaming</Text>
+              </TouchableOpacity>
+              <TouchableOpacity
+                style={[styles.button, styles.buttonPrimary, loading && styles.buttonDisabled]}
+                onPress={streamSampleFile}
+                disabled={loading}
+              >
+                <Text style={styles.buttonText}>📄 Stream Sample File</Text>
+              </TouchableOpacity>
+            </>
           ) : (
             <TouchableOpacity
               style={[styles.button, styles.buttonDanger]}

example/README.md

@@ -32,6 +32,7 @@ npm run ios
 
 - **System Info**: Shows device capabilities
 - **ASR**: Speech-to-text initialization
+- **Streaming ASR**: Live transcription from the microphone, or of a bundled clip fed in 320 ms chunks through `feedBuffer` (**Stream Sample File**, no microphone needed)
 - **VAD**: Voice activity detection
 - **Diarization**: Speaker identification
 - **TTS**: Text-to-speech synthesis
@@ -40,3 +41,4 @@ npm run ios
 
 - First run downloads ~500MB of ML models
 - Model compilation takes 20-30 seconds on first launch
+- `assets/speech-16k-mono.wav` (16.9 s, 16 kHz mono 16-bit) is `03-diff-explanation-16.9s.wav` from FluidAudio's test fixtures (Apache-2.0)

example/ios/FluidAudioExample/Info.plist

@@ -31,6 +31,8 @@
 		<key>NSAllowsLocalNetworking</key>
 		<true/>
 	</dict>
+	<key>NSMicrophoneUsageDescription</key>
+	<string>Streaming transcription listens to the microphone.</string>
 	<key>NSLocationWhenInUseUsageDescription</key>
 	<string></string>
 	<key>UILaunchStoryboardName</key>

example/ios/Podfile

@@ -17,9 +17,6 @@ end
 target 'FluidAudioExample' do
   config = use_native_modules!
 
-  # FluidAudio native library
-  pod 'FluidAudio', '~> 0.7'
-
   use_react_native!(
     :path => config[:reactNativePath],
     :app_path => "#{Pod::Config.instance.installation_root}/.."