app.py

import gradio as gr
import numpy as np
import torch
from datasets import load_dataset, Audio
from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline, WhisperFeatureExtractor, WhisperTokenizer, WhisperProcessor, AutomaticSpeechRecognitionPipeline, WhisperForConditionalGeneration
from dataclasses import dataclass
import re




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

# load speech translation checkpoint
asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-large-v3", device=device)

# load text-to-speech checkpoint and speaker embeddings
processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")

model = SpeechT5ForTextToSpeech.from_pretrained("Daniel981215/speecht5_tts_finetuned_voxpopuli_es").to(device)
vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)

embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)


replacements = {'á': 'a', 'é': 'e', 'í': 'i', 'ó': 'o', 'ú': 'u', '¿': '', '?': '', '1': 'uno', '2':'dos','3':'tres', '4':'cuatro', '5':'cinco', '6': 'seis', '7':'siete', '8':'ocho', '9':'nueve', '0':'cero'}

def normalize_replace_string(input_string, replacements):
    normalized_string = re.sub(r'\s+', ' ', input_string).strip().lower()
    
    for old, new in replacements.items():
        normalized_string = normalized_string.replace(old, new)
    
    return normalized_string


def translate(audio):
    outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "es"})
    output_txt = normalize_replace_string(outputs["text"], replacements)
    
    return output_txt


def synthesise(text):
    inputs = processor(text=text, return_tensors="pt")
    speech = model.generate_speech(inputs["input_ids"].to(device), speaker_embeddings.to(device), vocoder=vocoder)
    return speech.cpu()


def speech_to_speech_translation(audio):
    translated_text = translate(audio)
    synthesised_speech = synthesise(translated_text)
    synthesised_speech = (synthesised_speech.numpy() * 32767).astype(np.int16)
    return 16000, synthesised_speech


title = "Cascaded STST"
description = """
speech-to-speech translation (STST)
"""

demo = gr.Blocks()

mic_translate = gr.Interface(
    fn=speech_to_speech_translation,
    inputs=gr.Audio(source="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(source="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()