Getting "Cannot copy to a TensorFlowLite tensor (serving_default_input_1:0) with 63984 bytes from a Java Buffer with 64000 bytes" error while attempting to pass the Input and Output TensorAudio Buffer to TFLite Interpreter for inference

I am getting the following error while loading the input and output audio buffers to tensorflow-lite interpreter for inferencing. I am using the TensorAudio and TensorAudioFormat packages from the tflite-support.audio package. The tflite-lite model I am using is trained for audio samples with a 16kHz sample rate and mono channel. Please take a look at the code below.

float audioData [] = wavObj.ReadingAudioFile(getExternalFilesDir(Environment.DIRECTORY_DOWNLOADS).getPath()+ "/final_record.wav"); float [] outputBuffer; TensorAudioFormat.Builder audioFormat = TensorAudioFormat.builder(); audioFormat.setChannels(1); audioFormat.setSampleRate(16000); TensorAudio tensorAudio = TensorAudio.create(audioFormat.build(), audioFormat.build().getSampleRate()); tensorAudio.load(audioData); tflite.run(tensorAudio.getTensorBuffer().getBuffer(), outputBuffer);

Here, wavObj.ReadingAudioFile is returning the audio buffer for the saved audio file. I am using TensorAudioFormat to explicitly set the channel type and sample rate to match the trained model. I have load the tflite-trained model into tflite Interpreter using the below code.

AssetFileDescriptor fileDescriptor = this.getAssets().openFd(modelPath); FileInputStream inputStream = new FileInputStream(fileDescriptor.getFileDescriptor()); FileChannel fileChannel = inputStream.getChannel(); long startOffset = fileDescriptor.getStartOffset(); long declaredLength = fileDescriptor.getDeclaredLength(); ByteBuffer model = fileChannel.map(FileChannel.MapMode.READ_ONLY, startOffset, declaredLength); Interpreter.Options options = new Interpreter.Options(); tflite = new Interpreter(model, options);

Please suggest a solution to this problem.

Start with the TensorAudioFormat, TensorAudio.load, and tflite.run calls shown in the issue, then review the Android model-loading code using AssetFileDescriptor and Interpreter. Reproduce the reported buffer-size error and verify that the input and output buffers match the model's expected tensors; the issue is done when inference no longer reports the size mismatch.