app.py

import gradio as gr
import time
import torch
import scipy.io.wavfile
from espnet2.bin.tts_inference import Text2Speech
from espnet2.utils.types import str_or_none

tagen = 'espnet/english_male_ryanspeech_tacotron'
vocoder_tagen = "parallel_wavegan/ljspeech_melgan.v1.long"

text2speechen = Text2Speech.from_pretrained(
    model_tag=str_or_none(tagen),
    vocoder_tag=str_or_none(vocoder_tagen),
    device="cpu",
)

def inference(text, gender):
    with torch.no_grad():
        if gender == "male":
            wav = text2speechen(text)["wav"]
            scipy.io.wavfile.write("out.wav", text2speechen.fs, wav.view(-1).cpu().numpy())
    return "out.wav"


title = "RyanSpeech TTS"
description = "Gradio demo for RyanSpeech: First high quality speech dataset in the domain of conversation. (the female voice will be added in future).You get much better outputs when you use our <a href='https://www.kaggle.com/datasets/roholazandie/conformer-fastspeech2-ryanspeech'>pre-trained vocoder</a>. To use it, simply input a text, or click one of the examples to load. Please <a href='https://www.isca-speech.org/archive/interspeech_2021/zandie21_interspeech.html'>cite</a> our work"
article = "<p style='text-align: center'>" "<a href='https://arxiv.org/abs/2106.08468' target='_blank'>" "RyanSpeech-TTS</a> | <a href='http://mohammadmahoor.com/ryanspeech/' target='_blank'>Website</a> | <a href='https://www.kaggle.com/datasets/roholazandie/ryanspeech' target='_blank'>Download Dataset</a> | <a href='https://github.com/roholazandie/ryan-tts'>Github</a></p>"

examples = [['When he reached the suburbs, the light of homes was shining through curtains of all colors', "male"],
            ['I am a fully autonomous social robot. I can talk, listen, express, understand, and remember. My programming lets me have a conversation with just about anyone.', "male"],
            ['When in the very midst of our victory, here comes an order to halt.', "male"]]

gr.Interface(
    inference,
    [gr.inputs.Textbox(label="input text", lines=10),
     gr.inputs.Radio(choices=["male", "female"], type="value", default="male", label="Gender")],
    gr.outputs.Audio(type="file", label="Output"),
    title=title,
    description=description,
    article=article,
    enable_queue=True,
    examples=examples
).launch(debug=True)