updated code and implemented a new test: test_workflow

requirements.txt

@@ -1,20 +1,23 @@
-pytorch
-gradio                                  ==5.20.1
-gradio_client                           ==1.7.2
+torch
+gradio
+gradio_client
 llama-parse
-llama-index                             ==0.12.23
-llama-index-cli                         ==0.4.1
-llama-index-core                        ==0.12.23.post2
-llama-index-embeddings-huggingface      ==0.5.2
-llama-index-indices-managed-llama-cloud ==0.6.8
-llama-index-llms-cohere                 ==0.4.0
-llama-index-llms-openai
+llama-index
+llama-index-cli
+llama-index-core
+llama-index-embeddings-huggingface
+llama-index-indices-managed-llama-cloud
+llama-index-llms-cohere
+#llama-index-llms-openai
 llama-index-llms-groq
-llama-index-readers-file                ==0.4.6
-llama-index-readers-llama-parse         ==0.4.0
-llama-index-utils-workflow              ==0.3.0
+llama-index-readers-file
+llama-index-readers-llama-parse
+llama-index-utils-workflow
 #openai-whisper                          ==20240930
-llama-index-readers-whisper
-pydantic                                ==2.10.6
-pydantic_core                           ==2.27.2
+#llama-index-readers-whisper
+pydantic
+pydantic_core
 dotenv
+#faster-whisper
+whisper
+ffmpeg-python

step4.py

@@ -144,12 +144,21 @@ class RAGWorkflow(Workflow):
         # generate one query for each of the fields, and fire them off
         for field in fields:
             question = f"How would you answer this question about the candidate? <field>{field}</field>"
+            # Is there feedback? If so, add it to the query:
+            if hasattr(ev, "feedback"):
+                question += f"""
+                                \nWe previously got feedback about how we answered the questions.
+                                It might not be relevant to this particular field, but here it is:
+                                <feedback>{ev.feedback}</feedback>
+                            """
+                print("\n question : ", question)
+
             ctx.send_event(QueryEvent(
                 field=field,
                 query=question
             ))
 
-        # store the number of fields so we know how many to wait for later
+        # store the number of fields, so we know how many to wait for later
         await ctx.set("total_fields", len(fields))
         print(f"\n DEBUG: total fields from Context : {len(fields)}")
 
@@ -228,19 +237,6 @@ async def main():
         application_form="data/fake_application_form.pdf"
     )
 
-    '''
-    print("DEBUG: Awaiting next event manually...")
-    event = await handler.next_event()
-    print(f"DEBUG: Received event - {event}")
-
-    # Handle the first event if it's InputRequiredEvent
-    if isinstance(event, InputRequiredEvent):
-        print("We've filled in your form! Here are the results:\n")
-        print(event.result)
-        response = input(event.prefix)
-        handler.ctx.send_event(HumanResponseEvent(response=response))
-    '''
-
     print("DEBUG: Starting event stream...")
     async for event in handler.stream_events():
         print(f"DEBUG: Received event type {type(event).__name__}")

test_audio.py

@@ -0,0 +1,52 @@
+from llama_index.readers.whisper import WhisperReader
+from faster_whisper import WhisperModel
+import gradio as gr
+from pathlib import Path
+from dotenv import load_dotenv
+import os
+
+load_dotenv()
+openai_api_key = os.getenv("OPENAI_API_KEY")
+
+transcription_value = ""
+
+
+def transcribe_speech(filepath):
+    if filepath is None:
+        gr.Warning("No audio found, please retry.")
+
+    model = WhisperModel("base", compute_type="float32")
+    segments, _ = model.transcribe(filepath)
+
+    return " ".join(segment.text for segment in segments)
+
+
+def store_transcription(output):
+    global transcription_value
+    transcription_value = output
+    return output
+
+
+mic_transcribe = gr.Interface(
+    fn=lambda x: store_transcription(transcribe_speech(x)),
+    inputs=gr.Audio(sources=["microphone"], type="filepath"),
+    outputs=gr.Textbox(label="Transcription")
+)
+
+
+test_interface = gr.Blocks()
+with test_interface:
+    gr.TabbedInterface(
+        [mic_transcribe],
+        ["Transcribe Microphone"]
+    )
+
+test_interface.launch(
+    share=True,
+    server_port=8000,
+    #prevent_thread_lock=True
+)
+
+print(transcription_value)
+
+#test_interface.close()

test_audio_delay.py

@@ -0,0 +1,40 @@
+import gradio as gr
+import whisper
+import os
+
+
+def transcribe_audio(audio_file):
+
+    if not os.path.exists(audio_file):
+        print(f"Cannot locate file: {audio_file}")
+        return "Error: Audio file not found!"
+    else:
+        print(f"Processing file: {audio_file}")
+
+    model = whisper.load_model("base")
+    result = model.transcribe(audio_file, fp16=False)
+    return result["text"]
+
+
+def main():
+    audio_input = gr.Audio(sources=["microphone"], type="filepath")
+    output_text = gr.Textbox(label="Transcription")
+
+    iface = gr.Interface(fn=transcribe_audio,
+                         inputs=audio_input,
+                         outputs=output_text,
+                         title="Audio Transcription App",
+                         description="Record an audio file and hit the 'Submit' button"
+                         )
+
+    iface.launch(
+        share=True,
+        debug=True,
+        ssr_mode=False,
+        server_port=7860,
+        #prevent_thread_lock=True
+    )
+
+
+if __name__ == '__main__':
+    main()