Fix: send 'FINISH' text over WebSocket on stop to match server

app.py

@@ -4,13 +4,15 @@ import gradio as gr
 import librosa
 import numpy as np
 # import soundfile as sf
-from transformers import pipeline
 
 TARGET_SAMPLE_RATE = 16_000
 AUDIO_SECONDS_THRESHOLD = 2
-pipe = pipeline("audio-classification", model="MIT/ast-finetuned-audioset-10-10-0.4593")
+#pipe = pipeline("audio-classification", model="MIT/ast-finetuned-audioset-10-10-0.4593")
 prediction = [{"score": 1, "label": "recording..."}]
 
+from server_wrapper import process_chunk_from_bytes
+
+
 
 def normalize_waveform(waveform, datatype=np.float32):  # source datatype: np.int16
     waveform = waveform.astype(dtype=datatype)
@@ -25,7 +27,7 @@ def streaming_recording_fn(stream, new_chunk):
     y = librosa.resample(y, orig_sr=sr, target_sr=TARGET_SAMPLE_RATE)
     if stream is not None:
         if (stream.shape[-1] / TARGET_SAMPLE_RATE) >= AUDIO_SECONDS_THRESHOLD:
-            prediction = pipe(stream)
+            prediction = process_chunk_from_bytes((stream * 32768).astype(np.int16).tobytes())
             file_name = f'./audio/{time.strftime("%Y%m%d_%H%M%S", time.localtime())}.wav'
             # # sf.write(file_name, stream, TARGET_SAMPLE_RATE)
             print(f"SAVE AUDIO: {file_name}")