Using local ASR

app.py

@@ -14,7 +14,7 @@ load_dotenv(find_dotenv())
 from match_info_crawler import get_matches_info
 
 
-USE_LOCAL_ASR_PIPELINE = False
+USE_LOCAL_ASR_PIPELINE = True
 
 
 # used for chat, if provided
@@ -227,26 +227,24 @@ else:
     ASR_API_HEADERS = {"Authorization": f"Bearer {HF_KEY}"}
 
 
-def transcribe(audio_in):
-    # from numpy data:
-    #sr, y = audio_in  # sampling rate and audio data
-    #y2 = y.astype(np.float32)
-    #y2 /= np.max(np.abs(y))
-    #response = ASR_PIPELINE({"sampling_rate": sr, "raw": y})
+def transcribe(audio_file):
+    if USE_LOCAL_ASR_PIPELINE:
+        response = ASR_PIPELINE(audio_file)
+        text = response[0]["text"]
 
-    # using serverless API
-    with open(audio_in, "rb") as f:
-        data = f.read()
-    response = requests.post(ASR_API_URL, headers=ASR_API_HEADERS, data=data)
-    response = response.json()
+    else:
+        # using serverless API
+        with open(audio_file, "rb") as f:
+            data = f.read()
+        response = requests.post(ASR_API_URL, headers=ASR_API_HEADERS, data=data)
+        text = response.json()["text"]
     
-    #print(response)
-    return response['text']
+    return text
 
 
 def transcribe_and_respond(audio_in, system_txtbox, user_msg_txb, *args):
-    user_message = transcribe(audio_in)
-    outputs = respond(system_txtbox, user_message, *args)
+    transcribed_user_msg = transcribe(audio_in)
+    outputs = respond(system_txtbox, transcribed_user_msg, *args)
     return outputs
 
 
@@ -260,7 +258,7 @@ with gr.Blocks() as demo:
     
     user_msg_txb = gr.Textbox(label="Mensagem")
 
-    #audio_in = gr.Audio(label="Mensagem de Áudio", sources=['microphone'], interactive=True, type='filepath')  # TODO: tentar type='numpy'
+    audio_in = gr.Audio(label="Mensagem de Áudio", sources=['microphone'], interactive=True, type='filepath')
     
     submit_btn = gr.Button("Enviar")
     
@@ -289,7 +287,7 @@ with gr.Blocks() as demo:
 
     reset_btn.click(reset_and_apply, inputs=[voice_ddown], outputs=[chatbot_area, audio_out])
 
-    #audio_in.stop_recording( transcribe_and_respond, inputs=[audio_in, system_txtbox, user_msg_txb, chatbot_area, temperature_sldr, voice_ddown], outputs=[user_msg_txb, chatbot_area, audio_out] )
+    audio_in.stop_recording( transcribe_and_respond, inputs=[audio_in, system_txtbox, user_msg_txb, chatbot_area, temperature_sldr, voice_ddown], outputs=[user_msg_txb, chatbot_area, audio_out] )
     submit_btn.click(respond, inputs=[system_txtbox, user_msg_txb, chatbot_area, temperature_sldr, voice_ddown], outputs=[user_msg_txb, chatbot_area, audio_out])    # Click on the button
     user_msg_txb.submit(respond, inputs=[system_txtbox, user_msg_txb, chatbot_area, temperature_sldr, voice_ddown], outputs=[user_msg_txb, chatbot_area, audio_out]) # Press enter to submit - same effect