app.py

import gradio as gr
import librosa
import numpy as np
import torch
import os
import torch
from speechbrain.pretrained import EncoderClassifier
from scipy.io import wavfile
from IPython.display import Audio
from transformers import AutoProcessor, AutoModelForTextToSpectrogram, SpeechT5HifiGan

processor = AutoProcessor.from_pretrained("Prasada/speecht5_tts_voxpopuli_nl")
model = AutoModelForTextToSpectrogram.from_pretrained("Prasada/speecht5_tts_voxpopuli_nl")
vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")


spk_model_name = "speechbrain/spkrec-xvect-voxceleb"

device = "cuda" if torch.cuda.is_available() else "cpu"
speaker_model = EncoderClassifier.from_hparams(
    source=spk_model_name,
    run_opts={"device": device},
    savedir=os.path.join("/tmp", spk_model_name))

def create_speaker_embedding(waveform):
    with torch.no_grad():
        speaker_embeddings = speaker_model.encode_batch(torch.tensor(waveform))
        speaker_embeddings = torch.nn.functional.normalize(speaker_embeddings, dim=2)
        speaker_embeddings = speaker_embeddings.squeeze().cpu().numpy()
    return speaker_embeddings

def prepare_data(temp_text, temp_audio):
    rate, audio_data = temp_audio
    example = processor(
        text=temp_text,
        audio_target=audio_data,
        sampling_rate=16000,
        return_attention_mask=False,)
    example["speaker_embeddings"] = create_speaker_embedding(audio_data)
    example_embeddings = torch.tensor(example["speaker_embeddings"]).unsqueeze(0)
    return example_embeddings


def predict(temp_text, temp_audio, text):
    text = text
    embeddings=prepare_data(temp_text, temp_audio)
    inputs = processor(text=text, return_tensors="pt")
    spectrogram = model.generate_speech(inputs["input_ids"], embeddings)

    with torch.no_grad():
        speech = vocoder(spectrogram)
 
    return speech.numpy()



gr.Interface(
    fn=predict,
    inputs=[
        gr.Text(label="Template Text"),
        gr.Audio(label="Template Speech", type="numpy"),
        gr.Text(label="Input Text"),
    ],
    outputs=[
        gr.Audio(label="Generated Speech", type="numpy"),
    ],

).launch()