import os import tempfile import shutil from groq import Groq import gradio as gr from pydub import AudioSegment import re # Initialize the Groq client client = Groq(api_key=os.environ["keko"]) # def process_audio(audio_file): # if audio_file is None: # return None, None, "No audio file provided." # # Convert to MP3 if not already in MP3 format # audio = AudioSegment.from_file(audio_file) # mp3_file = tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") # audio.export(mp3_file.name, format="mp3") # return mp3_file.name, mp3_file.name, "Audio processed successfully." # def transcribe_audio(audio_file): # if audio_file is None: # return "No audio file provided." # # Create a transcription of the audio file # with open(audio_file, "rb") as file: # transcription = client.audio.transcriptions.create( # file=(audio_file, file.read()), # model="whisper-large-v3", # prompt="fix this dictated Text contains gastroenterology terms. remove words such as commas, newline, periods and replace with appropriate punctuations. apply corrections as specified by the user. keep format, minimal necessary changes only", # response_format="json", # temperature=0.0 # ) # return transcription.text # # Create the Gradio interface using Blocks # with gr.Blocks(title="Audio Recorder and Transcriber") as demo: # gr.Markdown("# Audio Recorder and Transcriber") # gr.Markdown("Record audio or upload a file. You can download the audio or transcribe it.") # audio_state = gr.State(None) # with gr.Row(): # audio_input = gr.Audio(sources=["microphone", "upload"], type="filepath", label="Record or Upload Audio") # with gr.Row(): # process_btn = gr.Button("Process Audio") # transcribe_btn = gr.Button("Transcribe Audio") # with gr.Row(): # audio_output = gr.Audio(label="Processed Audio (MP3)", format="mp3") # process_msg = gr.Textbox(label="Process Status") # transcription_output = gr.Textbox(label="Transcription", show_copy_button=True) # def update_audio_state(audio): # return audio if audio else None # audio_input.change( # update_audio_state, # inputs=[audio_input], # outputs=[audio_state] # ) # process_btn.click( # process_audio, # inputs=[audio_state], # outputs=[audio_state, audio_output, process_msg] # ) # transcribe_btn.click( # transcribe_audio, # inputs=[audio_state], # outputs=[transcription_output] # ) # # Launch the interface # demo.launch() ### original: ######################### # def save_audio(audio_file): # if audio_file is None: # return None, "No audio file provided." # # Save the audio file as MP3 # mp3_file = tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") # shutil.copy2(audio_file, mp3_file.name) # return mp3_file.name, "Audio saved successfully." # def transcribe_audio(audio_file): # if audio_file is None: # return "No audio file provided." # # Create a transcription of the audio file # with open(audio_file, "rb") as file: # transcription = client.audio.transcriptions.create( # file=(audio_file, file.read()), # model="whisper-large-v3", # prompt="text may contain medical gastroenterology terms", # response_format="json", # temperature=0.0 # ) # return transcription.text # # Create the Gradio interface using Blocks # with gr.Blocks(title="Audio Recorder and Transcriber") as demo: # gr.Markdown("# Audio Recorder and Transcriber") # gr.Markdown("Record audio or upload a file. You can download the audio or transcribe it.") # with gr.Row(): # audio_input = gr.Audio(sources=["microphone", "upload"], type="filepath", label="Record or Upload Audio") # with gr.Row(): # save_btn = gr.Button("Save Audio") # transcribe_btn = gr.Button("Transcribe Audio") # with gr.Row(): # audio_output = gr.Audio(label="Saved Audio (MP3)", format="mp3") # save_msg = gr.Textbox(label="Save Status") # transcription_output = gr.Textbox(label="Transcription", show_copy_button=True) # save_btn.click( # save_audio, # inputs=[audio_input], # outputs=[audio_output, save_msg] # ) # transcribe_btn.click( # transcribe_audio, # inputs=[audio_input], # outputs=[transcription_output] # ) # # Launch the interface # demo.launch() #### trial send to LLM ######################### import gradio as gr import os from groq import Groq import tempfile import shutil # Initialize the Groq client client = Groq(api_key=os.environ["keko"]) def check_password(password): correct_password = "zoo" # Set your desired password here return password == correct_password def save_audio(audio_file): if audio_file is None: return None, "No audio file provided." mp3_file = tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") shutil.copy2(audio_file, mp3_file.name) return mp3_file.name, "Audio saved successfully." def split_audio(audio_file, chunk_size=25*1024*1024, overlap=10000): audio = AudioSegment.from_file(audio_file) duration = len(audio) chunks = [] start = 0 while start < duration: end = start + chunk_size if end > duration: end = duration chunk = audio[start:end] if len(chunk.raw_data) > chunk_size: end = start + (chunk_size // chunk.frame_width) * chunk.frame_width chunk = audio[start:end] chunks.append(chunk) start = end - overlap return chunks def transcribe_audio(audio_file): if audio_file is None: return "No audio file provided." file_size = os.path.getsize(audio_file) max_size = 25 * 1024 * 1024 # 25 MB in bytes if file_size <= max_size: # If the file is small enough, process it directly with open(audio_file, "rb") as file: transcription = client.audio.transcriptions.create( file=(audio_file, file.read()), model="whisper-large-v3", prompt="fix this dictated Text contains gastroenterology terms. remove words such as commas, newline, periods and replace with appropriate punctuations. apply corrections as specified by the user. keep format, minimal necessary changes only", response_format="json", temperature=0.0 ) return transcription.text else: # If the file is too large, split it and process chunks chunks = split_audio(audio_file) transcriptions = [] for i, chunk in enumerate(chunks): with tempfile.NamedTemporaryFile(suffix=".mp3", delete=False) as temp_file: chunk.export(temp_file.name, format="mp3") with open(temp_file.name, "rb") as file: transcription = client.audio.transcriptions.create( file=(f"chunk_{i}.mp3", file.read()), model="whisper-large-v3", prompt="fix this dictated Text contains gastroenterology terms. ", response_format="json", temperature=0.0 ) transcriptions.append(transcription.text) os.unlink(temp_file.name) return " ".join(transcriptions) ## function to parse text into dictionary of prompts def parse_text_to_dict(text): sections = re.split(r"\n#/\s*", text.strip()) # Split only at "#/" parsed_dict = {} for section in sections: if not section.strip(): # Skip empty sections continue lines = section.split("\n", 1) # Split into title and content title = lines[0].strip().lstrip("#/ ") # Ensure "#/" is removed content = lines[1].strip() if len(lines) > 1 else "" # Remaining text is the value parsed_dict[title] = content return parsed_dict # load the prompts: with open("primpts.txt", "r", encoding="utf-8") as file: contentt = file.read() contentt = parse_text_to_dict(contentt) def generate_text(text, prompt_type): if not text: return "No text provided." prompts = contentt selected_prompt = prompts.get(prompt_type, "You are a helpful assistant. Process the following text.") response = client.chat.completions.create( model="llama-3.3-70b-versatile", messages=[ {"role": "system", "content": selected_prompt}, {"role": "user", "content": text} ], max_tokens=4000, temperature=0.1 ) return response.choices[0].message.content # Define the new function for "Transcribe & Fix" def transcribe_and_generate(audio_file, prompt_type): # Transcribe the audio transcription = transcribe_audio(audio_file) # Generate the text based on the transcription and selected prompt generated_text = generate_text(transcription, prompt_type) return transcription, generated_text # Define a new function that combines save, transcribe, and fix def save_transcribe_fix(audio_file, prompt_type): # Step 1: Save the audio saved_audio, save_status = save_audio(audio_file) if not saved_audio: return None, save_status, None # Return error message if saving fails # Step 2: Transcribe the audio transcription = transcribe_audio(saved_audio) # Step 3: Generate the fixed text fixed_text = generate_text(transcription, prompt_type) return saved_audio, transcription, fixed_text # ------------------------------------------------ # Create the Gradio interface using Blocks # ------------------------------------------------- # Update Gradio UI with a new button with gr.Blocks(title="Audio Recorder, Transcriber, and Text Generator") as demo: gr.Markdown("# Audio Recorder, Transcriber, and Text Generator") gr.Markdown("Enter the correct password to access the application.") with gr.Group() as login_container: password_input = gr.Textbox(type="password", label="Enter Password") login_button = gr.Button("Login") login_message = gr.Markdown() with gr.Column(visible=False) as main_interface: gr.Markdown("Record audio, transcribe it, or enter text manually, and optionally generate text based on the input.") input_toggle = gr.Checkbox(label="Use Manual Text Input Instead of Audio", value=False) # Audio section including transcription output with gr.Column() as audio_section: audio_input = gr.Audio(sources=["microphone", "upload"], type="filepath", label="Record or Upload Audio") with gr.Row(): save_btn = gr.Button("Save Audio") transcribe_btn = gr.Button("Transcribe Audio") transcribe_fix_btn = gr.Button("Transcribe & Fix") save_transcribe_fix_btn = gr.Button("Save, Transcribe & Fix") with gr.Row(): audio_output = gr.Audio(label="Saved Audio (MP3)", format="mp3") save_msg = gr.Textbox(label="Save Status") transcription_output = gr.Textbox(label="Transcription", show_copy_button=True) # Text input section (starts hidden) text_input = gr.Textbox(label="Enter Text Manually", visible=False) # Toggle input sources (audio or text) input_toggle.change( lambda use_text: ( gr.update(visible=not use_text), # Hide audio section if using text input gr.update(visible=use_text), # Show text input when checked gr.update(visible=not use_text) # Hide transcription output when using text input ), inputs=[input_toggle], outputs=[audio_section, text_input, transcription_output] ) with gr.Row(): prompt_type = gr.Dropdown( choices=list(contentt.keys()), # Get dictionary keys as choices label="Select Text Generation Type", value=list(contentt.keys())[0] # Default to the first key ) generate_btn = gr.Button("Generate Text") generated_text_output = gr.Textbox(label="Generated Text", show_copy_button=True) # Actions for audio-related inputs save_btn.click( save_audio, inputs=[audio_input], outputs=[audio_output, save_msg] ) transcribe_btn.click( transcribe_audio, inputs=[audio_input], outputs=[transcription_output] ) transcribe_fix_btn.click( transcribe_and_generate, inputs=[audio_input, prompt_type], outputs=[transcription_output, generated_text_output] ) save_transcribe_fix_btn.click( save_transcribe_fix, inputs=[audio_input, prompt_type], outputs=[audio_output, transcription_output, generated_text_output] ) # Automatically detect if text or audio should be used def determine_input(audio, text, prompt): if text.strip(): # If text is entered, use it return generate_text(text, prompt) elif audio: # Otherwise, check if audio exists and use it return generate_text(transcription_output.value, prompt) else: return "Please provide either text or audio." generate_btn.click( determine_input, inputs=[audio_input, text_input, prompt_type], outputs=[generated_text_output] ) def login(password): if check_password(password): return { login_container: gr.update(visible=False), # Hide login UI main_interface: gr.update(visible=True), # Show main interface login_message: gr.update(value="Login successful.", visible=True) } else: return { login_message: gr.update(value="Incorrect password. Please try again.", visible=True) } login_button.click( login, inputs=[password_input], outputs=[login_container, main_interface, login_message] ) demo.launch(share=True)