Update app.py

app.py

@@ -3,27 +3,15 @@ from transformers import AutoProcessor, BlipForConditionalGeneration, AutoTokeni
 import librosa
 import numpy as np
 import torch
-
-# from transformers import AutoProcessor, AutoTokenizer, AutoImageProcessor, AutoModelForCausalLM, BlipForConditionalGeneration, Blip2ForConditionalGeneration, VisionEncoderDecoderModel
+import audio_model
 import open_clip
 
 #CONSTANTS
-speaker_embeddings = {
-    "BDL": "spkemb/cmu_us_bdl_arctic-wav-arctic_a0009.npy",
-    "CLB": "spkemb/cmu_us_clb_arctic-wav-arctic_a0144.npy",
-    "RMS": "spkemb/cmu_us_rms_arctic-wav-arctic_b0353.npy",
-    "SLT": "spkemb/cmu_us_slt_arctic-wav-arctic_a0508.npy",
-}
+
 # Carga el modelo de clasificación de imagen a texto
 blip_processor_large = AutoProcessor.from_pretrained("Salesforce/blip-image-captioning-large")
 blip_model_large = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-large")
 
-# # Carga el modelo de clasificación de tetxo a audio speech
-checkpoint = "microsoft/speecht5_tts"
-processor = SpeechT5Processor.from_pretrained(checkpoint)
-model = SpeechT5ForTextToSpeech.from_pretrained(checkpoint)
-vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")
-
 device = "cuda" if torch.cuda.is_available() else "cpu"
 blip_model_large.to(device)
 
@@ -52,48 +40,13 @@ def generate_caption_coca(model, transform, image):
     return open_clip.decode(generated[0].detach()).split("<end_of_text>")[0].replace("<start_of_text>", "")
 
 
-def generate_captions(image):
+def generate_captions_speech(image):
 
     caption_blip_large = generate_caption(blip_processor_large, blip_model_large, image)
     print('generate_captions>>>'+caption_blip_large)
     return caption_blip_large,text_to_speech(caption_blip_large,"Surprise Me!")
     
 #####END IMAGE MODEL TO TEXT
-
-### TEXT TO AUDIO SPEECH  MODEL 2    
-# Define la función que convierte texto en voz
-def text_to_speech(text,speaker):
-    # Genera el audio utilizando el modelo 
-    if len(text.strip()) == 0:
-       return (16000, np.zeros(0).astype(np.int16))
-    inputs = processor(text=text, return_tensors="pt")
-
-    # limit input length
-    input_ids = inputs["input_ids"]
-    input_ids = input_ids[..., :model.config.max_text_positions]
-
-    if speaker == "Surprise Me!":
-        # load one of the provided speaker embeddings at random
-        idx = np.random.randint(len(speaker_embeddings))
-        key = list(speaker_embeddings.keys())[idx]
-        speaker_embedding = np.load(speaker_embeddings[key])
-
-        # randomly shuffle the elements
-        np.random.shuffle(speaker_embedding)
-
-        # randomly flip half the values
-        x = (np.random.rand(512) >= 0.5) * 1.0
-        x[x == 0] = -1.0
-        speaker_embedding *= x
-
-        #speaker_embedding = np.random.rand(512).astype(np.float32) * 0.3 - 0.15
-    speaker_embedding = torch.tensor(speaker_embedding).unsqueeze(0)
-
-    speech = model.generate_speech(input_ids, speaker_embedding, vocoder=vocoder)
-
-    speech = (speech.numpy() * 32767).astype(np.int16)
-    return (16000, speech)
-### END TEXT TO AUDIO SPEECH  MODEL 2 
     
 # Define la interfaz de usuario utilizando Gradio entradas y salidas
 inputsImg = [
@@ -106,7 +59,7 @@ title = "Clasificación de imagen a texto y conversión de texto a voz"
 description = "Carga una imagen y obtén una descripción de texto de lo que contiene la imagen, así como un archivo de audio de la trasncripcion de la imagen en audio descrito."
 examples = []
 
-interface = gr.Interface(fn=generate_captions, 
+interface = gr.Interface(fn=generate_captions_speech, 
                          inputs=inputsImg,
                          outputs=outputs,
                          examples=examples,