Update app.py

app.py

@@ -1,29 +1,29 @@
-import torch
-from transformers import SpeechEncoderDecoder, Wav2Vec2Processor
 import gradio as gr
-import scipy.signal as sps
+import librosa
+from transformers import Wav2Vec2Processor, SpeechEncoderDecoder
+    
+processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-xls-r-300m-en-to-15", use_auth_token="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")
+model = SpeechEncoderDecoder.from_pretrained("facebook/wav2vec2-xls-r-300m-en-to-15", use_auth_token="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")
 
-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 process_audio_file(file):
+    data, sr = librosa.load(file)
+    if sr != 16000:
+        data = librosa.resample(data, sr, 16000)
+    print(data.shape)
+    input_values = processor(data, return_tensors="pt").input_values
+    return input_values
     
-def parse_transcription(wav_file):
-    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)
-    transcription = processor.decode(predicted_ids[0], skip_special_tokens=True)
-    return transcription
+def transcribe(file):
+    input_values = process_audio_file(file)
     
+    sequences = model.generate(input_values, num_beams=1, max_length=30)
     
-processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-xls-r-300m-en-to-15", use_auth_token="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")
-model = SpeechEncoderDecoder.from_pretrained("facebook/wav2vec2-xls-r-300m-en-to-15", use_auth_token="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")
+    transcription = processor.batch_decode(sequences)
+    return transcription[0]
     
-
-#input_ = gr.inputs.Audio(source="microphone", type="file") 
-input_ = gr.inputs.Audio(source="microphone", type="numpy") 
-gr.Interface(parse_transcription, inputs = input_,  outputs="text", 
-             analytics_enabled=False, show_tips=False, enable_queue=True).launch(inline=False);
+iface = gr.Interface(
+    fn=speech_recognize, 
+    inputs=gr.inputs.Audio(source="microphone", type='filepath'),
+    outputs="text",
+)
+iface.launch()