added app.py

app.py

@@ -0,0 +1,50 @@
+from transformers import pipeline
+import gradio as gr
+import os
+
+whisper = pipeline(model='jlvdoorn/whisper-large-v2-atco2-asr-atcosim', use_auth_token=os.environ['HUGGINGFACE_TOKEN'])
+bert_atco_ner = pipeline(model='Jzuluaga/bert-base-ner-atc-en-atco2-1h')
+
+def transcribe(audio_mic, audio_file):
+    if audio_file is not None:
+        return whisper(audio_file)['text']
+    if audio_mic is not None:
+        return whisper(audio_mic)['text']
+    else:
+        return 'There was no audio to transcribe...'
+
+def extractCallSignCommand(transcription):
+    if type(transcription) is str:
+        result = bert_atco_ner(transcription)
+        callsigns = []
+        commands = []
+        values = []
+        for item in result:
+            if 'callsign' in item['entity']:
+                callsigns.append(item['word'])
+            if 'command' in item['entity']:
+                commands.append(item['word'])
+            if 'value' in item['entity']:
+                values.append(item['word'])
+                
+        return 'Callsigns: ' + ', '.join(callsigns) + '\nCommands: ' + ', '.join(commands) + '\nValues: ' + ', '.join(values)
+    else:
+        return 'There was no transcription to extract a callsign or command from...'
+
+def transcribeAndExtract(audio_mic, audio_file, transcribe_only):
+    transcription = transcribe(audio_mic, audio_file)
+    if not transcribe_only:
+        callSignCommandValues = extractCallSignCommand(transcription)
+    else:
+        callSignCommandValues = ''
+    return transcription, callSignCommandValues
+
+iface = gr.Interface(
+        fn=transcribeAndExtract,
+        inputs=[gr.Audio(source='microphone', type='filepath'), gr.Audio(source='upload', type='filepath'), gr.Checkbox(label='Transcribe only', default=False)],
+        outputs=[gr.Text(label='Transcription'), gr.Text(label='Callsigns, commands and values')],
+        title='Whisper Large v2 - ATCO2-ASR-ATCOSIM',
+        description='Whisper Large v2 model fine-tuned on the ATCO2-ASR and ATCOSIM datasets.'
+)
+
+iface.launch()