TFlite Audio Plugin for Flutter

Audio classification Tflite package for flutter (iOS & Android). Can also support Google Teachable Machine models.

If you are a complete newbie to audio classification, you can read the tutorial here. Credit to Carolina for writing a comprehensive article.

To keep this project alive, consider giving a star or a like. Pull requests or bug reports are also welcome.

Recording	Inference result

Table of Contents

• About this plugin
• Known Issues/Commonly asked questions
• Please read if you are using Google's Teachable Machine. Otherwise skip.
• How to add tflite model and label to flutter
• How to use this plugin
• Rough guide on parameters
• Android Installation & Permissions
• iOS Installation & Permissions
• References

About This Plugin

The plugin has several features:

1. Audio recognition for stored audio files. (Only mono wav files for now)

2. Audio recognition for recordings.

3. Tunable parameters for recording/inferences

4. Please look a the parameters below for more information.

5. Automatically reshape/transpose audio inputs.

This plugin can support several model types:

1. Google Teachable Machine (Raw audio input)

2. For beginners with little to no machine learning knowledge. You can read can read the tutorial here if you are a newbie.

3. Training can be done here

4. Raw audio input.

5. Can recognize the following inputs: float32[audioLength, 1] or float32[1, audioLength]

6. For more information on how to train your own model, take a look here.

7. Decoded wav input.

8. Supports two inputs: float32[audioLength, 1] and int32[1]

9. For more information on how to train your own model. Take a look here

10. To train a decoded wave with MFCC, take a look here

11. (Experimental feature) Spectogram, melspectrogram, and MFCC inputs.

12. Please note that this feature is experimental, and results may not be accurate compared to raw audio / decoded wav.

13. Spectrogram model can be trained here tutorial.

14. (Currently worked on feature) Multiple input and outputs.

Known Issues/Commonly asked questions

1. How to adjust the recording length/time

There are two ways to reduce adjust recording length/time:

• You can increase the recording time by adjusting the bufferSize to a lower value.

• You can also increase recording time by lowering the sample rate.

Note: That stretching the value too low will cause problems with model accuracy. In that case, you may want to consider lowering your sample rate as well. Likewise, a very low sample rate can also cause problems with accuracy. It is your job to find the sweetspot for both values.

1. How to reduce false positives in my model

To reduce false positives, you may want to adjust the default values of detectionThreshold=0.3 and averageWindowDuration=1000 to a higher value. A good value for both respectively are 0.7 and 1500. For more details about these parameters, please visit this section.

1. I am getting build errors on iOS

There are several ways to fix this:

• Some have reported to fix this issue by replacing the following line:

  ```ruby target 'Runner' do use_frameworks! use_modular_headers! #pod 'TensorFlowLiteSelectTfOps' #Old line pod'TensorFlowLiteSelectTfOps','~> 2.6.0' #New line

  flutter_install_all_ios_pods File.dirname(File.realpath(FILE)) end ```

• Others have fixed this issue building the app without the line: pod 'TensorFlowLiteSelectTfOps. Then rebuilding the app by re-adding the line again.

• Remember to run the following below:

``` 1. cd into iOS folder

 2. Run `flutter pub get` on terminal

 3. Run `pod install` on terminal

 4. Run `flutter clean` on terminal

 5. Run `flutter run` on terminal. All done!

```

1. I am getting TensorFlow Lite Error on iOS. - Regular TensorFlow ops are not supported by this interpreter. Make sure you apply/link the Flex delegate before inference

2. Please make sure that you have enabled ops-select on your podfile - step 4 & Xcode - step 5 and build gradle - step 3

3. If you tried above, please run the example on a device (not emulator). If you still recieved this error, its very likely that theres an issue with cocoapod or Xcode configuration. Please check the issue #7

4. If you recieved this error from your custom model (not GTM), its likely that you're using unsupported tensorflow operators for tflite, as found in issue #5. For more details on which operators are supported, look at the official documentation here

5. Take a looking at issue number 3 if none of the above works.

6. (iOS) App crashes when running Google's Teachable Machine model

Please run your simulation on actual iOS device. Running your device on M1 macs should also be ok.

As of this moment, there's limited support for x86_64 architectures from the Tensorflow Lite select-ops framework. If you absolutely need to run it on an emulator, you can consider building the select ops framework yourself. Instructions can be found here

1. (Android) Fatal signal 11 (SIGSEGV), code 1 (SEGV_MAPERR), fault addr 0xfffffff4 in tid 5403

It seems like the latest tflite package for android is causing this issue. Until this issue is fixed, please run this package on an actual Android Device.

Please Read If You Are Using Google's Teachable Machine. (Otherwise Skip)

BE AWARE: Google's Teachable Machine requires select tensorflow operators to work. This feature is experimental and will cause the following issues:

1. Increase the overall size of your app. If this is unnacceptable for you, it's recommended that you build your own custom model. Tutorials can be found in the About this plugin section

2. Emulators for iOS do not work due to limited support for x86_64 architectures. You need to run your simulation on an actual device. Issue can be found here

3. You will need to manually implement ops-select on your podfile - step 4 & Xcode - step 5 and build gradle - step 3

How to add tflite model and label to flutter:

1. Create an assets folder and then place your custom tflite model and labels inside.

1. In pubsec.yaml, link your tflite model and label under 'assets'. For example:

  assets:
    - assets/decoded_wav_model.tflite
    - assets/decoded_wav_label.txt

How to use this plugin

Please look at the example on how to implement these futures.

1. To add the package in pubspec.yaml, open your terminal and run this line in your flutter project:

flutter pub add tflite_audio

1. Import the plugin. For example:

import 'package:tflite_audio/tflite_audio.dart';

1. To load your model:

```dart //Example for decodedWav models TfliteAudio.loadModel( model: 'assets/conv_actions_frozen.tflite', label: 'assets/conv_actions_label.txt', inputType: 'decodedWav');

//Example for Google's Teachable Machine models TfliteAudio.loadModel( model: 'assets/google_teach_machine_model.tflite', label: 'assets/google_teach_machine_label.txt', inputType: 'rawAudio');

//Example if you want to take advantage of all optional parameters from loadModel() TfliteAudio.loadModel( model: 'assets/conv_actions_frozen.tflite', label: 'assets/conv_actions_label.txt', inputType: 'decodedWav', outputRawScores: false, numThreads: 1, isAsset: true, ); ```

1. To start and listen to the stream for inference results:

   • Declare stream value dart Stream<Map<dynamic, dynamic>> recognitionStream;

   • If you want to use the recognition stream for recording: ```dart //Example values for Google's Teachable Machine models recognitionStream = TfliteAudio.startAudioRecognition( sampleRate: 44100, bufferSize: 22016, )

   //Example values for decodedWav recognitionStream = TfliteAudio.startAudioRecognition( sampleRate: 16000, bufferSize: 2000, )

   //Example for advanced users who want to utilise all optional parameters from this package. //Note the values are default. recognitionStream = TfliteAudio.startAudioRecognition( sampleRate: 44100, bufferSize: 22016, numOfInferences: 5, audioLength = 44032, detectionThreshold: 0.3, averageWindowDuration = 1000, minimumTimeBetweenSamples = 30, suppressionTime = 1500, )

   ```

   • If you want to use the recognition stream for stored audio files.

   ```dart //Example values for Google teachable models recognitionStream = TfliteAudio.startFileRecognition( sampleRate: 44100, audioDirectory: "assets/sampleAudio.wav", );

   //Examples values for decodedWav recognitionStream = TfliteAudio.startFileRecognition( sampleRate: 16000, audioDirectory: "assets/sampleAudio.wav", );

   //Example for advanced users who want to utilise all optional parameters from this package. recognitionStream = TfliteAudio.startFileRecognition( sampleRate: 44100, audioDirectory: "assets/sampleAudio.wav", audioLength: 44032, detectionThreshold: 0.3, averageWindowDuration: 1000, minimumTimeBetweenSamples: 30, suppressionTime: 1500, ); ```

   • Listen for results ```dart String result = ''; int inferenceTime = 0;

   recognitionStream.listen((event){ result = event["inferenceTime"]; inferenceTime = event["recognitionResult"]; }) .onDone( //Do something here when stream closes ); ```

2. To forcibly cancel recognition stream

dart TfliteAudio.stopAudioRecognition();

Rough guide on the parameters

• outputRawScores - Will output the result as an array in string format. For example '[0.2, 0.6, 0.1, 0.1]'

• numThreads - Higher threads will reduce inferenceTime. However, will utilise the more cpu resource.

• isAsset - is your model, label or audio file in the asset file? If yes, set true. If the files are outside (such as external storage), set false.

• numOfInferences - determines how many times you want to loop the recording and inference. For example: numOfInference = 3 will repeat the recording three times, so recording length will be (1 to 2 seconds) x 3 = (3 to 6 seconds). Also the model will output the scores three times.

• sampleRate - A higher sample rate may improve accuracy for recordings. Recommened values are 16000, 22050, 44100

• audioLength - Default is 0 as the plugin will determine the length for you. You can manually adjust this if you wish to shorten or extend the number of audio samples.

• bufferSize - A lower value will lengthen the recording. Likewise, a higehr value will shorten the recording. Make sure this value is equal or bel