Spaces:
Running
Running
| import os | |
| import torch | |
| import gradio as gr | |
| from huggingface_hub import InferenceClient | |
| from model import predict_params, AudioDataset | |
| import torchaudio | |
| # TODO: Que no diga lo de que no hay 1s_normal al predecir | |
| token = os.getenv("HF_TOKEN") | |
| client = InferenceClient("meta-llama/Meta-Llama-3-8B-Instruct", token=token) | |
| device = torch.device("cuda" if torch.cuda.is_available() else "cpu") | |
| model_class, id2label_class = predict_params( | |
| model_path="A-POR-LOS-8000/distilhubert-finetuned-mixed-data", | |
| dataset_path="data/mixed_data", | |
| filter_white_noise=True, | |
| undersample_normal=True | |
| ) | |
| model_mon, id2label_mon = predict_params( | |
| model_path="A-POR-LOS-8000/distilhubert-finetuned-cry-detector", | |
| dataset_path="data/baby_cry_detection", | |
| filter_white_noise=False, | |
| undersample_normal=False | |
| ) | |
| def call(audiopath, model, dataset_path, filter_white_noise, undersample_normal=False): | |
| model.to(device) | |
| model.eval() | |
| audio_dataset = AudioDataset(dataset_path, {}, filter_white_noise, undersample_normal) | |
| processed_audio = audio_dataset.preprocess_audio(audiopath) | |
| inputs = {"input_values": processed_audio.to(device).unsqueeze(0)} | |
| with torch.no_grad(): | |
| outputs = model(**inputs) | |
| logits = outputs.logits | |
| return logits | |
| def predict(audio_path_pred): | |
| with torch.no_grad(): | |
| logits = call(audio_path_pred, model=model_class, dataset_path="data/mixed_data", filter_white_noise=True, undersample_normal=False) | |
| predicted_class_ids_class = torch.argmax(logits, dim=-1).item() | |
| label_class = id2label_class[predicted_class_ids_class] | |
| label_mapping = {0: 'Cansancio/Incomodidad', 1: 'Dolor', 2: 'Hambre', 3: 'Problemas para respirar'} | |
| label_class = label_mapping.get(predicted_class_ids_class, label_class) | |
| return label_class | |
| def predict_stream(audio_path_stream): | |
| with torch.no_grad(): | |
| logits = call(audio_path_stream, model=model_mon, dataset_path="data/baby_cry_detection", filter_white_noise=False, undersample_normal=False) | |
| probabilities = torch.nn.functional.softmax(logits, dim=-1) | |
| crying_probabilities = probabilities[:, 1] | |
| avg_crying_probability = crying_probabilities.mean()*100 | |
| if avg_crying_probability < 15: | |
| label_class = predict(audio_path_stream) | |
| return f"Está llorando por: {label_class}" | |
| else: | |
| return "No está llorando" | |
| def decibelios(audio_path_stream): | |
| waveform, sample_rate = torchaudio.load(audio_path_stream) | |
| rms = torch.sqrt(torch.mean(torch.square(waveform))) | |
| db_level = 20 * torch.log10(rms + 1e-6).item() | |
| min_db = -80 | |
| max_db = 0 | |
| scaled_db_level = (db_level - min_db) / (max_db - min_db) | |
| normalized_db_level = scaled_db_level * 100 | |
| return normalized_db_level | |
| def mostrar_decibelios(audio_path_stream, visual_threshold): | |
| db_level = decibelios(audio_path_stream) | |
| if db_level > visual_threshold: | |
| return f"Prediciendo... Decibelios: {db_level:.2f}" | |
| elif db_level < visual_threshold: | |
| return f"Esperando... Decibelios: {db_level:.2f}" | |
| def predict_stream_decib(audio_path_stream, visual_threshold): | |
| db_level = decibelios(audio_path_stream) | |
| if db_level > visual_threshold: | |
| llorando, probabilidad = predict_stream(audio_path_stream) | |
| return f"{llorando}" | |
| else: | |
| return "" | |
| def chatbot_config(message, history: list[tuple[str, str]]): | |
| system_message = "You are a Chatbot specialized in baby health and care." | |
| max_tokens = 512 | |
| temperature = 0.5 | |
| top_p = 0.95 | |
| messages = [{"role": "system", "content": system_message}] | |
| for val in history: | |
| if val[0]: | |
| messages.append({"role": "user", "content": val[0]}) | |
| if val[1]: | |
| messages.append({"role": "assistant", "content": val[1]}) | |
| messages.append({"role": "user", "content": message}) | |
| response = "" | |
| for message_response in client.chat_completion(messages, max_tokens=max_tokens, stream=True, temperature=temperature, top_p=top_p): | |
| token = message_response.choices[0].delta.content | |
| response += token | |
| yield response | |
| def cambiar_pestaña(): | |
| return gr.update(visible=False), gr.update(visible=True) | |
| my_theme = gr.themes.Soft( | |
| primary_hue="emerald", | |
| secondary_hue="green", | |
| neutral_hue="slate", | |
| text_size="sm", | |
| spacing_size="sm", | |
| font=[gr.themes.GoogleFont('Nunito'), 'ui-sans-serif', 'system-ui', 'sans-serif'], | |
| font_mono=[gr.themes.GoogleFont('Nunito'), 'ui-monospace', 'Consolas', 'monospace'], | |
| ).set( | |
| body_background_fill='*neutral_50', | |
| body_text_color='*neutral_600', | |
| body_text_size='*text_sm', | |
| embed_radius='*radius_md', | |
| shadow_drop='*shadow_spread', | |
| shadow_spread='*button_shadow_active' | |
| ) | |
| with gr.Blocks(theme=my_theme) as demo: | |
| with gr.Column(visible=True) as inicial: | |
| gr.HTML( | |
| """ | |
| <style> | |
| @import url('https://fonts.googleapis.com/css2?family=Lobster&display=swap'); | |
| @import url('https://fonts.googleapis.com/css2?family=Roboto&display=swap'); | |
| h1 { | |
| font-family: 'Lobster', cursive; | |
| font-size: 5em !important; | |
| text-align: center; | |
| margin: 0; | |
| } | |
| .gr-button { | |
| background-color: #4CAF50 !important; | |
| color: white !important; | |
| border: none; | |
| padding: 25px 50px; | |
| text-align: center; | |
| text-decoration: none; | |
| display: inline-block; | |
| font-family: 'Lobster', cursive; | |
| font-size: 2em !important; | |
| margin: 4px 2px; | |
| cursor: pointer; | |
| border-radius: 12px; | |
| } | |
| .gr-button:hover { | |
| background-color: #45a049; | |
| } | |
| h2 { | |
| font-family: 'Lobster', cursive; | |
| font-size: 3em !important; | |
| text-align: center; | |
| margin: 0; | |
| } | |
| p.slogan, h4, p, h3 { | |
| font-family: 'Roboto', sans-serif; | |
| text-align: center; | |
| } | |
| </style> | |
| <h1>Iremia</h1> | |
| <h4 style='text-align: center; font-size: 1.5em'>El mejor aliado para el bienestar de tu bebé</h4> | |
| """ | |
| ) | |
| gr.Markdown( | |
| "<h4 style='text-align: left; font-size: 1.5em;'>¿Qué es Iremia?</h4>" | |
| "<p style='text-align: left'>Iremia es un proyecto llevado a cabo por un grupo de estudiantes interesados en el desarrollo de modelos de inteligencia artificial, enfocados específicamente en casos de uso relevantes para ayudar a cuidar a los más pequeños de la casa.</p>" | |
| "<h4 style='text-align: left; font-size: 1.5em;'>Nuestra misión</h4>" | |
| "<p style='text-align: left'>Sabemos que la paternidad puede suponer un gran desafío. Nuestra misión es brindarles a todos los padres unas herramientas de última tecnología que los ayuden a navegar esos primeros meses de vida tan cruciales en el desarrollo de sus pequeños.</p>" | |
| "<h4 style='text-align: left; font-size: 1.5em;'>¿Qué ofrece Iremia?</h4>" | |
| "<p style='text-align: left'>Chatbot: Pregunta a nuestro asistente que te ayudará con cualquier duda que tengas sobre el cuidado de tu bebé.</p>" | |
| "<p style='text-align: left'>Analizador: Con nuestro modelo de inteligencia artificial somos capaces de predecir por qué tu hijo de menos de 2 años está llorando.</p>" | |
| "<p style='text-align: left'>Monitor: Nuestro monitor no es como otros que hay en el mercado, ya que es capaz de reconocer si un sonido es un llanto del bebé o no; y si está llorando, predice automáticamente la causa. Dándote la tranquilidad de saber siempre qué pasa con tu pequeño, ahorrándote tiempo y horas de sueño.</p>" | |
| ) | |
| boton_inicial = gr.Button("¡Prueba nuestros modelos!") | |
| with gr.Column(visible=False) as chatbot: | |
| gr.Markdown("<h2>Asistente</h2>") | |
| gr.Markdown("<h4 style='text-align: center; font-size: 1.5em'>Pregunta a nuestro asistente cualquier duda que tengas sobre el cuidado de tu bebé</h4>") | |
| gr.ChatInterface( | |
| chatbot_config, | |
| theme=my_theme, | |
| retry_btn=None, | |
| undo_btn=None, | |
| clear_btn="Limpiar 🗑️", | |
| autofocus=True, | |
| fill_height=True, | |
| ) | |
| with gr.Row(): | |
| with gr.Column(): | |
| boton_predictor = gr.Button("Predictor") | |
| with gr.Column(): | |
| boton_monitor = gr.Button("Monitor") | |
| with gr.Column(visible=False) as pag_predictor: | |
| gr.Markdown("<h2>Predictor</h2>") | |
| gr.Markdown("<h4 style='text-align: center; font-size: 1.5em'>Descubre por qué tu bebé está llorando</h4>") | |
| audio_input = gr.Audio( | |
| min_length=1.0, | |
| format="wav", | |
| label="Baby recorder", | |
| type="filepath", | |
| ) | |
| gr.Button("¿Por qué llora?").click( | |
| predict, | |
| inputs=audio_input, | |
| outputs=gr.Textbox(label="Tu bebé llora por:") | |
| ) | |
| gr.Button("Volver").click(cambiar_pestaña, outputs=[pag_predictor, chatbot]) | |
| with gr.Column(visible=False) as pag_monitor: | |
| gr.Markdown("<h2>Monitor</h2>") | |
| gr.Markdown("<h4 style='text-align: center; font-size: 1.5em'>Detecta en tiempo real si tu bebé está llorando y por qué</h4>") | |
| audio_stream = gr.Audio( | |
| format="wav", | |
| label="Baby recorder", | |
| type="filepath", | |
| streaming=True | |
| ) | |
| threshold_db = gr.Slider( | |
| minimum=0, | |
| maximum=100, | |
| step=1, | |
| value=30, | |
| label="Umbral de ruido para activar la predicción:" | |
| ) | |
| audio_stream.stream( | |
| mostrar_decibelios, | |
| inputs=[audio_stream, threshold_db], | |
| outputs=gr.Textbox(value="Esperando...", label="Estado") | |
| ) | |
| audio_stream.stream( | |
| predict_stream_decib, | |
| inputs=[audio_stream, threshold_db], | |
| outputs=gr.Textbox(value="", label="Tu bebé:") | |
| ) | |
| gr.Button("Volver").click(cambiar_pestaña, outputs=[pag_monitor, chatbot]) | |
| boton_inicial.click(cambiar_pestaña, outputs=[inicial, chatbot]) | |
| boton_predictor.click(cambiar_pestaña, outputs=[chatbot, pag_predictor]) | |
| boton_monitor.click(cambiar_pestaña, outputs=[chatbot, pag_monitor]) | |
| demo.launch(share=True) | |