app.py

import gradio as gr


from transformers import pipeline 

from datasets import load_dataset
import soundfile as sf
import torch
import os
import io 
import base64
import numpy as np
from pydub import AudioSegment

os.environ['TRANSFORMERS_CACHE'] = '.cache'

print ("----- setting up pipeline -----")


synthesiser = pipeline("text-to-speech", "microsoft/speecht5_tts")

print ("----- setting up dataset -----")


embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
speaker_embedding = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
# You can replace this embedding with your own as well.

print ("----- synthetizing audio -----")

#speech = synthesiser("Hello, my dog is cooler than you!", forward_params={"speaker_embeddings": speaker_embedding})

#sf.write("speech.wav", speech["audio"], samplerate=speech["sampling_rate"])

def greet(name):
    return "Hello " + name + "!!"


def synthesise_audio(text, forward_params=None):
    if len(text) > 100:
        raise ValueError("Error: El texto es demasiado largo. Por favor, limita tu entrada a 100 caracteres.")

    speech = synthesiser(text, forward_params={"speaker_embeddings": speaker_embedding})
    
    
    sf.write("speech.wav", speech["audio"], samplerate=speech["sampling_rate"])
    return "speech.wav"

    # sf.write("speech.wav", speech["audio"], samplerate=speech["sampling_rate"])
    # return "speech.wav"
    
    # Convert numpy array to audio
    #with io.BytesIO() as f:
    #    sf.write(f, speech["audio"], samplerate=speech["sampling_rate"], format='wav')
    #    audio = f.getvalue()

    # Convert numpy array to audio
    #audio = np.int16(speech["audio"] * 32767)
    #audio_segment = AudioSegment(audio, sample_width=2, frame_rate=speech["sampling_rate"], channels=1)

    # Convert numpy array to list
    #audio = speech["audio"]

    #return speech["audio"]

    #return audio
    
    # Ensure audio is a numpy array
    #if isinstance(speech["audio"], int):
    #    audio = np.array([speech["audio"]])
    #else:
    #    audio = speech["audio"]


    # Create an in-memory buffer to store the audio data
    #print("Creating in-memory buffer")
    #audio_buffer = io.BytesIO()

    # Write the audio data to the in-memory buffer
    #print("Writing audio data to in-memory buffer")
    #sf.write(audio_buffer, speech["audio"], samplerate=speech["sampling_rate"], format="WAV")

    # Move the buffer cursor to the beginning of the buffer
    #audio_buffer.seek(0)

    # Read the audio data from the in-memory buffer into a numpy array
    #print("Reading audio data from in-memory buffer")
    #audio, sr = sf.read(audio_buffer)
    #print("Audio data read from in-memory buffer, returning audio data and sample rate")
    # Ensure audio is a numpy array before returning
    #audio = np.array(audio)

    #return audio, sr     
  
    


#demo = gr.Interface(fn=greet, inputs="text", outputs="text", description="----- TTS Testing -----")

input_text = gr.Textbox(lines=10, label="Type the text you want to convert to speech:")
#examples = gr.Examples(inputs=[["Feliz cumpleaños Nuria"]], outputs=[[]])

demo = gr.Interface(fn=synthesise_audio,
                    inputs=input_text,
                    outputs="audio",
                    #outputs = gr.Audio(type="numpy"),
                    description="----- manuai Text To Speech generator test -----",
                    allow_flagging = False)

demo.launch(debug = True)