Update app.py

app.py

@@ -1,8 +1,7 @@
 import gradio as gr
 import torch
-import numpy as np
+import torchaudio
 from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
-from scipy.signal import resample
 
 # Load model and processor
 model_id = "facebook/wav2vec2-large-960h-lv60-self"
@@ -10,43 +9,37 @@ processor = Wav2Vec2Processor.from_pretrained(model_id)
 model = Wav2Vec2ForCTC.from_pretrained(model_id)
 
 # Transcription function
-def transcribe(audio_data):
-    if audio_data is None:
+def transcribe(audio_file):
+    if audio_file is None:
         return "⚠️ No audio received."
 
-    audio, sample_rate = audio_data
-
-    # Convert stereo to mono if needed
-    if len(audio.shape) == 2:
-        audio = np.mean(audio, axis=1)
-
-    # Ensure sample_rate is an integer
-    sample_rate = int(sample_rate)
-
-    # Resample to 16000 Hz if needed
+    # Load and convert audio
+    waveform, sample_rate = torchaudio.load(audio_file)
     if sample_rate != 16000:
-        number_of_samples = round(len(audio) * 16000 / sample_rate)
-        audio = resample(audio, number_of_samples)
+        waveform = torchaudio.functional.resample(waveform, orig_freq=sample_rate, new_freq=16000)
+        sample_rate = 16000
 
-    # Normalize audio
-    audio = audio.astype(np.float32)
+    # Only one channel (mono)
+    if waveform.shape[0] > 1:
+        waveform = waveform.mean(dim=0).unsqueeze(0)
+
+    input_values = processor(waveform.squeeze().numpy(), sampling_rate=sample_rate, return_tensors="pt").input_values
 
-    # Process and predict
-    input_values = processor(audio, sampling_rate=16000, return_tensors="pt").input_values
     with torch.no_grad():
         logits = model(input_values).logits
-        predicted_ids = torch.argmax(logits, dim=-1)
-        transcription = processor.batch_decode(predicted_ids)[0]
+
+    predicted_ids = torch.argmax(logits, dim=-1)
+    transcription = processor.batch_decode(predicted_ids)[0]
 
     return transcription.lower()
 
-# Launch UI
+# Gradio UI
 demo = gr.Interface(
     fn=transcribe,
-    inputs=gr.Audio(sources=["microphone"], type="numpy", label="🎤 Speak now"),
+    inputs=gr.Audio(sources=["microphone"], type="filepath", label="🎤 Speak now"),
     outputs=gr.Textbox(label="📝 Transcription"),
     title="Wav2Vec2 Speech Transcription",
-    description="Speak into the microphone and get a transcription using Wav2Vec2 (Hugging Face)."
+    description="Speak into the microphone and get a transcription using Wav2Vec2 (via Hugging Face Transformers)."
 )
 
 demo.launch()