import tempfile
from typing import Optional
from TTS.config import load_config
import gradio as gr
import numpy as np
import os
from TTS.utils.manage import ModelManager
from TTS.utils.synthesizer import Synthesizer
from espeak_phonemizer import Phonemizer
from engine import Piper
from festival import festival_synthesize

MAX_TXT_LEN = 325

fonemitzador = Phonemizer("ca")

def carrega_bsc():
    model_path = os.getcwd() + "/models/bsc/best_model.pth"
    config_path = os.getcwd() + "/models/bsc/config.json"
    speakers_file_path = os.getcwd() + "/models/bsc/speakers.pth"
    vocoder_path = None
    vocoder_config_path = None

    synthesizer = Synthesizer(
        model_path, config_path, speakers_file_path, None, vocoder_path, vocoder_config_path,
    )

    return synthesizer

def carrega_collectivat():
    model_path = os.getcwd() + "/models/collectivat/fast-speech_best_model.pth"
    config_path = os.getcwd() + "/models/collectivat/fast-speech_config.json"
    vocoder_path = os.getcwd() + "/models/collectivat/ljspeech--hifigan_v2_model_file.pth"
    vocoder_config_path = os.getcwd() + "/models/collectivat/ljspeech--hifigan_v2_config.json"
    synthesizer = Synthesizer(
        model_path, config_path, None, None, vocoder_path, vocoder_config_path
    )

    return synthesizer

def carrega_piper():
    return Piper(os.getcwd() + "/models/piper/ca-upc_ona-x-low.onnx")


model_bsc = carrega_bsc()
SPEAKERS = model_bsc.tts_model.speaker_manager.speaker_names

model_collectivat = carrega_collectivat()

model_piper = carrega_piper()

def tts(text, festival_voice, speaker_idx):
    if len(text) > MAX_TXT_LEN:
        text = text[:MAX_TXT_LEN]
        print(f"Input text was cutoff since it went over the {MAX_TXT_LEN} character limit.")
    print(text)

    # synthesize
    wav_bsc = model_bsc.tts(text, speaker_idx)
    wav_coll = model_collectivat.tts(text)
    wav_piper = model_piper.synthesize(text)

    #return (model_bsc.tts_config.audio["sample_rate"], wav_bsc), (22000, wav_coll), (16000, wav_piper)

    # return output
    fp_bsc = ""
    with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as fp:
        model_bsc.save_wav(wav_bsc, fp)
        fp_bsc = fp.name

    fp_coll = ""
    with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as fp:
        model_collectivat.save_wav(wav_coll, fp)
        fp_coll = fp.name

    fp_piper = ""
    with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as fp:
        fp.write(wav_piper)
        fp_piper = fp.name

    fonemes = fonemitzador.phonemize(text, keep_clause_breakers=True)

    fp_festival = festival_synthesize(text, festival_voice)

    return fonemes, fp_festival, fp_bsc, fp_coll, fp_piper


description="""
Amb aquesta aplicació podeu sintetitzar text a veu amb els últims models neuronals lliures pel català.

1. Model multi-parlant VITS entrenat pel BSC (Projecte Aina)
    https://huggingface.co/projecte-aina/tts-ca-coqui-vits-multispeaker

2. Model Fastspeech entrenat per Col·lectivat
    https://github.com/CollectivaT-dev/TTS-API

3. Model VITS entrenat per Piper/Home Assistant
    https://github.com/rhasspy/piper

Els dós últims models han estat entrenats amb la veu d'Ona de FestCAT. El primer model ha estat entrenat amb totes les veus de FestCAT, els talls de Common Voice 8 i un altre corpus pel que conté moltes veus de qualitat variable. La veu d'Ona està seleccionada per defecte per la comparativa però podeu provar les altres.

Aquesta aplicació fa servir l'últim estat de l'espeak millorat per Carme Armentano del BSC
https://github.com/projecte-aina/espeak-ng

NOTA: El model de col·lectivat treballa amb grafemes pel que no fa servir espeak com a fonemitzador.
"""
article= ""

iface = gr.Interface(
    fn=tts,
    inputs=[
        gr.Textbox(
            label="Text",
            value="L'Èlia i l'Alí a l'aula.  L'oli i l'ou.  Lulú olorava la lila.",
        ),
        gr.Dropdown(label="Parlant del motor Festival", choices=["ona", "pau"], value="ona"),
        gr.Dropdown(label="Parlant del model VITS multi-parlant del BSC", choices=SPEAKERS, value="ona")
    ],
    outputs=[
        gr.Markdown(label="Fonemes"),
        gr.Audio(label="Festival",type="filepath"),
        gr.Audio(label="BSC VITS",type="filepath"),
        gr.Audio(label="Collectivat Fastspeech",type="filepath"),
        gr.Audio(label="Piper VITS",type="filepath")
    ],
    title="Comparativa de síntesi lliure en català️",
    description=description,
    article=article,
    allow_flagging="never",
    layout="vertical",
    live=False
)
iface.launch(server_name="0.0.0.0", server_port=7860)