whisper_ggml_plus 1.3.0

Whisper.cpp Flutter plugin with Large-v3-Turbo (128-mel) support.

example/lib/main.dart

import 'dart:io';

import 'package:flutter/material.dart';
import 'package:flutter/services.dart';
import 'package:path_provider/path_provider.dart';
import 'package:whisper_ggml_plus/whisper_ggml_plus.dart';
import 'package:record/record.dart';

/// Whisper GGML Plus Example
///
/// This example demonstrates basic speech recognition with optional CoreML acceleration.
///
/// IMPORTANT: CoreML Acceleration (iOS/macOS only)
/// =================================================
/// For 3x+ faster transcription on Apple Silicon (M1+, A14+), you can add a CoreML encoder.
///
/// ⚠️ CRITICAL: .mlmodelc is a DIRECTORY, not a file!
/// - Cannot be bundled via Flutter assets (directory structure breaks)
/// - Must be deployed via runtime download or native bundle
/// - Must be placed in same directory as .bin model file
///
/// Example structure:
/// ```
/// /app/support/models/
/// ├── ggml-base-q5_0.bin
/// └── ggml-base-encoder.mlmodelc/       ← Directory
///     ├── model.mil                      ← CoreML model IR
///     ├── coremldata.bin                 ← Model weights
///     └── metadata.json                  ← Model config
/// ```
///
/// How to generate .mlmodelc:
/// ```bash
/// git clone https://github.com/ggerganov/whisper.cpp
/// cd whisper.cpp
/// python3.11 -m venv venv && source venv/bin/activate
/// pip install torch==2.5.0 "numpy<2.0" coremltools==8.1 openai-whisper ane_transformers
/// ./models/generate-coreml-model.sh base  # or large-v3-turbo
/// # Output: models/ggml-base-encoder.mlmodelc/ (~800MB directory)
/// ```
///
/// See README.md for detailed deployment options:
/// - Option A: Runtime download (recommended)
/// - Option B: Xcode native bundle (advanced)

void main() {
  runApp(const MyApp());
}

class MyApp extends StatelessWidget {
  const MyApp({super.key});

  @override
  Widget build(BuildContext context) {
    return MaterialApp(
      title: 'Whisper ggml example',
      theme: ThemeData(
        colorScheme: ColorScheme.fromSeed(seedColor: Colors.deepPurple),
        useMaterial3: true,
      ),
      home: const MyHomePage(),
    );
  }
}

class MyHomePage extends StatefulWidget {
  /// Modify this model based on your needs

  const MyHomePage({super.key});

  @override
  State<MyHomePage> createState() => _MyHomePageState();
}

class _MyHomePageState extends State<MyHomePage> {
  /// Change this to WhisperModel.largeV3 for better accuracy (requires more RAM)
  /// For CoreML acceleration, see the header comments on how to deploy .mlmodelc
  final model = WhisperModel.base;

  final AudioRecorder audioRecorder = AudioRecorder();
  final WhisperController whisperController = WhisperController();
  String transcribedText = 'Transcribed text will be displayed here';
  bool isProcessing = false;
  bool isProcessingFile = false;
  bool isListening = false;

  @override
  void initState() {
    initModel();
    super.initState();
  }

  @override
  Widget build(BuildContext context) {
    return Scaffold(
      appBar: AppBar(
        backgroundColor: Theme.of(context).colorScheme.inversePrimary,
        title: Text('Whisper ggml example'),
      ),
      body: SafeArea(
        child: Padding(
          padding: const EdgeInsets.symmetric(horizontal: 24),
          child: Stack(
            fit: StackFit.expand,
            children: [
              Center(
                child: Text(
                  transcribedText,
                  style: Theme.of(context).textTheme.headlineMedium,
                ),
              ),
              Positioned(
                bottom: 24,
                left: 0,
                child: Tooltip(
                  message: 'Transcribe jfk.wav asset file',
                  child: CircleAvatar(
                    backgroundColor: Colors.purple.shade100,
                    maxRadius: 25,
                    child: isProcessingFile
                        ? const CircularProgressIndicator()
                        : IconButton(
                            onPressed: transcribeJfk,
                            icon: Icon(
                              Icons.folder,
                            ),
                          ),
                  ),
                ),
              )
            ],
          ),
        ),
      ),
      floatingActionButton: FloatingActionButton(
        onPressed: record,
        tooltip: 'Start listening',
        child: isProcessing
            ? const CircularProgressIndicator()
            : Icon(
                isListening ? Icons.mic_off : Icons.mic,
                color: isListening ? Colors.red : null,
              ),
      ),
    );
  }

  Future<void> initModel() async {
    try {
      /// Try initializing the model from assets
      ///
      /// NOTE: Only GGML .bin models can be bundled in assets.
      /// CoreML .mlmodelc directories CANNOT be bundled via assets!
      /// See header comments for CoreML deployment methods.
      final bytesBase =
          await rootBundle.load('assets/ggml-${model.modelName}.bin');
      final modelPathBase = await whisperController.getPath(model);
      final fileBase = File(modelPathBase);
      await fileBase.writeAsBytes(bytesBase.buffer
          .asUint8List(bytesBase.offsetInBytes, bytesBase.lengthInBytes));
    } catch (e) {
      /// On error try downloading the model
      await whisperController.downloadModel(model);
    }
  }

  Future<void> record() async {
    if (await audioRecorder.hasPermission()) {
      if (await audioRecorder.isRecording()) {
        final audioPath = await audioRecorder.stop();

        if (audioPath != null) {
          debugPrint('Stopped listening.');

          setState(() {
            isListening = false;
            isProcessing = true;
          });

          /// Transcribe the recorded audio
          ///
          /// If CoreML encoder is present in the same directory as the model,
          /// it will be automatically detected and used for 3x+ faster transcription.
          ///
          /// CoreML detection: ggml-base-*.bin → looks for ggml-base-encoder.mlmodelc/
          /// No code changes needed - detection is automatic!
          final result = await whisperController.transcribe(
            model: model,
            audioPath: audioPath,
            lang: 'en',
          );

          if (mounted) {
            setState(() {
              isProcessing = false;
            });
          }

          if (result?.transcription.text != null) {
            setState(() {
              transcribedText = result!.transcription.text;
            });
          }
        } else {
          debugPrint('No recording exists.');
        }
      } else {
        debugPrint('Started listening.');

        setState(() {
          isListening = true;
        });

        final Directory appDirectory = await getTemporaryDirectory();
        await audioRecorder.start(const RecordConfig(),
            path: '${appDirectory.path}/test.m4a');
      }
    }
  }

  Future<void> transcribeJfk() async {
    final Directory tempDir = await getTemporaryDirectory();
    final asset = await rootBundle.load('assets/jfk.wav');
    final String jfkPath = "${tempDir.path}/jfk.wav";
    final File convertedFile = await File(jfkPath).writeAsBytes(
      asset.buffer.asUint8List(),
    );

    setState(() {
      isProcessingFile = true;
    });

    final result = await whisperController.transcribe(
      model: model,
      audioPath: convertedFile.path,
      lang: 'auto',
    );

    setState(() {
      isProcessingFile = false;
    });

    if (result?.transcription.text != null) {
      setState(() {
        transcribedText = result!.transcription.text;
      });
    }
  }
}