Update app.py

app.py

@@ -6,4 +6,52 @@ import warnings
 
 from transformers import Wav2Vec2Tokenizer, Wav2Vec2ForCTC
 
-warnings.filterwarnings("ignore")
+warnings.filterwarnings("ignore")
+
+#load wav2vec2 tokenizer and model
+# define speech-to-text function
+def asr_transcript(audio_file, language):
+    
+    if language == "English":
+      model_name = "facebook/wav2vec2-large-960h-lv60-self"
+    elif language == "Russian":
+      model_name = "jonatasgrosman/wav2vec2-large-xlsr-53-russian"
+    elif language == "French":
+      model_name = "jonatasgrosman/wav2vec2-large-xlsr-53-french"
+    
+    tokenizer = Wav2Vec2Tokenizer.from_pretrained(model_name)
+    model = Wav2Vec2ForCTC.from_pretrained(model_name)
+
+    transcript = ""
+
+    # Stream over 20 seconds chunks
+    stream = librosa.stream(
+        audio_file.name, block_length=20, frame_length=16000, hop_length=16000
+    )
+
+    for speech in stream:
+        if len(speech.shape) > 1:
+            speech = speech[:, 0] + speech[:, 1]
+
+        input_values = tokenizer(speech, return_tensors="pt").input_values
+        logits = model(input_values).logits
+
+        predicted_ids = torch.argmax(logits, dim=-1)
+        transcription = tokenizer.batch_decode(predicted_ids)[0]
+        transcript += transcription.lower() + " "
+
+    return transcript
+    
+gradio_ui = gr.Interface(
+    fn=asr_transcript,
+    title="Speech-to-Text with HuggingFace+Wav2Vec2",
+    description="Upload an audio clip in Russian, English, or French and let AI do the hard work of transcribing",
+    inputs = [gr.inputs.Audio(label="Upload Audio File", type="file"),
+              gr.inputs.Radio(label="Pick an STT Model - (language)", 
+                              choices=["English",
+                                       "Russian",
+                                       "French"])],
+    outputs=gr.outputs.Textbox(label="Auto-Transcript"),
+)
+
+gradio_ui.launch()