app.py

import gradio as gr
import numpy as np
import torch
from transformers import pipeline, VitsModel, AutoTokenizer, AutoTokenizer
from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor

device = "cuda:0" if torch.cuda.is_available() else "cpu"


translation_models = {
    "en": "Helsinki-NLP/opus-mt-en-es",  # Inglés a Español
    "fr": "Helsinki-NLP/opus-mt-fr-es",  # Francés a Español
    "de": "Helsinki-NLP/opus-mt-de-es",  # Alemán a Español
    "it": "Helsinki-NLP/opus-mt-it-es",  # Italiano a Español
    "pt": "Helsinki-NLP/opus-mt-pt-es",  # Portugués a Español
    "nl": "Helsinki-NLP/opus-mt-nl-es",  # Neerlandés (Holandés) a Español
    "fi": "Helsinki-NLP/opus-mt-fi-es",  # Finés a Español
    "sv": "Helsinki-NLP/opus-mt-sv-es",  # Sueco a Español
    "da": "Helsinki-NLP/opus-mt-da-es",  # Danés a Español
    "no": "Helsinki-NLP/opus-mt-no-es",  # Noruego a Español
    "ru": "Helsinki-NLP/opus-mt-ru-es",  # Ruso a Español
    "pl": "Helsinki-NLP/opus-mt-pl-es",  # Polaco a Español
    "cs": "Helsinki-NLP/opus-mt-cs-es",  # Checo a Español
    "tr": "Helsinki-NLP/opus-mt-tr-es",  # Turco a Español
    "zh": "Helsinki-NLP/opus-mt-zh-es",  # Chino a Español
    "ja": "Helsinki-NLP/opus-mt-ja-es",  # Japonés a Español
    "ar": "Helsinki-NLP/opus-mt-ar-es",  # Árabe a Español
    "ro": "Helsinki-NLP/opus-mt-ro-es",  # Rumano a Español
    "el": "Helsinki-NLP/opus-mt-el-es",  # Griego a Español
    "bg": "Helsinki-NLP/opus-mt-bg-es",  # Búlgaro a Español
    "uk": "Helsinki-NLP/opus-mt-uk-es",  # Ucraniano a Español
    "he": "Helsinki-NLP/opus-mt-he-es",  # Hebreo a Español
    "lt": "Helsinki-NLP/opus-mt-lt-es",  # Lituano a Español
    "et": "Helsinki-NLP/opus-mt-et-es",  # Estonio a Español
    "hr": "Helsinki-NLP/opus-mt-hr-es",  # Croata a Español
    "hu": "Helsinki-NLP/opus-mt-hu-es",  # Húngaro a Español
    "lv": "Helsinki-NLP/opus-mt-lv-es",  # Letón a Español
    "sl": "Helsinki-NLP/opus-mt-sl-es",  # Esloveno a Español
    "sk": "Helsinki-NLP/opus-mt-sk-es",  # Eslovaco a Español
    "sr": "Helsinki-NLP/opus-mt-sr-es",  # Serbio a Español
    "fa": "Helsinki-NLP/opus-mt-fa-es",  # Persa a Español
}

asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)

#vist_model = VitsModel.from_pretrained("facebook/mms-tts-spa")
#vist_tokenizer = AutoTokenizer.from_pretrained("facebook/mms-tts-spa")


model = SpeechT5ForTextToSpeech.from_pretrained(
    "juangtzi/speecht5_finetuned_voxpopuli_es"
)
checkpoint = "microsoft/speecht5_tts"
processor = SpeechT5Processor.from_pretrained(checkpoint)
vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")

speaker_embeddings2 = np.load('speaker_embeddings.npy')
speaker_embeddings2 = torch.tensor(speaker_embeddings2)
print(speaker_embeddings2)
lang_detector = pipeline("text-classification", model="papluca/xlm-roberta-base-language-detection")

def language_detector(text):
    resultado = lang_detector(text)
    idioma_detectado = resultado[0]['label']
    print(idioma_detectado)
    return idioma_detectado

def translate(audio):
    transcribe = asr_pipe(audio, max_new_tokens=256)

    codigo_idioma = language_detector(transcribe['text'])

    if codigo_idioma in translation_models:
        translator = pipeline("translation", model=translation_models[codigo_idioma])
        traduccion = translator(transcribe['text'])
    else:
        transcribe = transcribe['text']
        print(f"No hay un modelo de traducción disponible para el idioma detectado {codigo_idioma}")
        return transcribe
    
    return traduccion

def synthesise(text):
    if isinstance(text, list):
        text = text[0]['translation_text']
    else:
        text = text
    print(text)
    inputs = processor(text, return_tensors="pt")
    output = model.generate_speech(inputs["input_ids"], speaker_embeddings2, vocoder=vocoder)
    return output

def speech_to_speech_translation(audio):
    translated_text = translate(audio)
    synthesised_speech = synthesise(translated_text)
    audio_data = synthesised_speech.cpu().numpy()
    audio_data = np.squeeze(audio_data)
    audio_data = audio_data / np.max(np.abs(audio_data))
    sample_rate = 16000
    return (sample_rate, audio_data)

title = "Cascaded STST"
description = """
Demo for cascaded speech-to-speech translation (STST), mapping from source speech in any language to target speech in Spanish.

![Cascaded STST](https://huggingface.co/datasets/huggingface-course/audio-course-images/resolve/main/s2st_cascaded.png "Diagram of cascaded speech to speech translation")
"""

demo = gr.Blocks()

mic_translate = gr.Interface(
    fn=speech_to_speech_translation,
    inputs=gr.Audio(sources="microphone", type="filepath"),
    outputs=gr.Audio(label="Generated Speech", type="numpy"),
    title=title,
    description=description,
)

file_translate = gr.Interface(
    fn=speech_to_speech_translation,
    inputs=gr.Audio(sources="upload", type="filepath"),
    outputs=gr.Audio(label="Generated Speech", type="numpy"),
    examples=[["./example.wav"]],
    title=title,
    description=description,
)

with demo:
    gr.TabbedInterface([mic_translate, file_translate], ["Microphone", "Audio File"])

demo.launch()