Real-time identification of microphone has no result.

[eugeneYz]

I used a microphone to record and converted the PCM data into a stream in WAV format, which was input into the ProcessAsync method,but stuck at this step all along:await foreach (var result in processor.ProcessAsync(waveStream)).
Unable to obtain result, I suspect it is due to a short recording time, and the totaltime of audio data is only 100ms

            {
                nAudioHelper.PcmDataAvailable += async (data) =>
                {
                    try
                    {
                        if (data != null && data.Length > 0)
                        {
                            pcmData = data;
                            MemoryStream outStream = new();
                            MemoryStream memoryStream = new MemoryStream(pcmData);
                            WaveFormat waveFormat = new WaveFormat(16000, 16, 1); 
                            //WaveStream waveStream = new RawSourceWaveStream(pcmData, 0, pcmData.Length, new WaveFormat(16000, 16, 1));
                            WaveStream waveStream = new RawSourceWaveStream(memoryStream, waveFormat);
                            var pcmStream = WaveFormatConversionStream.CreatePcmStream(waveStream);
                            var resampler = new WdlResamplingSampleProvider(pcmStream.ToSampleProvider(), 16000);

                            WaveFileWriter.WriteWavFileToStream(outStream, resampler.ToWaveProvider16());
                            outStream.Seek(0, SeekOrigin.Begin);

                            await foreach (var result in processor.ProcessAsync(waveStream))
                            {
                                string recognizedText = result.Text;
                                Application.Current.Dispatcher.Invoke(() =>
                                {
                                    Value += recognizedText;
                                });
                            }

[sandrohanea]

There are multiple issues with this approach.

1. Currently, we don't support real-time identification reliably as described here: How to handle real-time sound streams #25

2. If you provide a small chunk of audio (let's say half a second) and run ProcessAsync => it will be normal that Whisper won't recognize anything as probably there was no word inside half a second. Also whisper.cpp have some limit on minimum duration of 1second: SIGFPE on certain audio files ggerganov/whisper.cpp#39