Update app.py

app.py

@@ -9,7 +9,7 @@ import yaml
 import tensorflow as tf
 from tensorflow_tts.inference import TFAutoModel
 from tensorflow_tts.inference import AutoProcessor
-
+import scipy.signal as sps
 
 
 
@@ -63,10 +63,20 @@ def translate(text):
     
 processor = Wav2Vec2Processor.from_pretrained("Harveenchadha/vakyansh-wav2vec2-hindi-him-4200")
 model = Wav2Vec2ForCTC.from_pretrained("Harveenchadha/vakyansh-wav2vec2-hindi-him-4200")
+
+def read_file(wav):
+    sample_rate, signal = wav                                                                                                                        
+    signal = signal.mean(-1)                                                                                                                              
+    number_of_samples = round(len(signal) * float(16000) / sample_rate)                                                                                   
+    resampled_signal = sps.resample(signal, number_of_samples)
+    return resampled_signal
+
+
 def parse_transcription(wav_file):
-    filename = wav_file.name.split('.')[0]
-    convert(wav_file.name, filename + "16k.wav")
-    speech, _ = sf.read(filename + "16k.wav")
+    #filename = wav_file.name.split('.')[0]
+    #convert(wav_file.name, filename + "16k.wav")
+    #speech, _ = sf.read(filename + "16k.wav")
+    speech = read_file(wav_file)
     input_values = processor(speech, sampling_rate=16_000, return_tensors="pt").input_values
     logits = model(input_values).logits
     predicted_ids = torch.argmax(logits, dim=-1)
@@ -81,7 +91,7 @@ def parse_transcription(wav_file):
 output1 = gr.outputs.Textbox(label="Hindi Output from ASR")
 output2 = gr.outputs.Textbox(label="English Translated Output")
     
-input_ = gr.inputs.Audio(source="microphone", type="file") 
+input_ = gr.inputs.Audio(source="microphone", type="numpy") 
 
 
 output_audio =  gr.outputs.Audio(type="file", label="Output Audio")