local/app.vctk.py

"""
TODO:
    + [x] Load Configuration
    + [ ] Checking
    + [ ] Better saving directory
"""
import numpy as np
from pathlib import Path
import jiwer
import pdb
import torch.nn as nn
import torch
import torchaudio
from transformers import pipeline
from time import process_time, time
from pathlib import Path
# local import
import sys
from espnet2.bin.tts_inference import Text2Speech
# pdb.set_trace()
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu") 

sys.path.append("src")

import gradio as gr

# ASR part

audio_files = [str(x) for x in  sorted(Path("/home/kevingeng/Disk2/laronix/laronix_automos/data/20230103_video").glob("**/*wav"))]
# audio_files = [str(x) for x in  sorted(Path("./data/Patient_sil_trim_16k_normed_5_snr_40/Rainbow").glob("**/*wav"))]
# transcriber = pipeline("automatic-speech-recognition", model="KevinGeng/PAL_John_128_train_dev_test_seed_1")
transcriber = pipeline("automatic-speech-recognition", model="KevinGeng/PAL_John_128_p326_300_train_dev_test_seed_1")
# 【Female】kan-bayashi ljspeech parallel wavegan
# tts_model = Text2Speech.from_pretrained("espnet/kan-bayashi_ljspeech_vits")
# 【Male】fastspeech2-en-200_speaker-cv4, hifigan vocoder
# pdb.set_trace()
from fairseq.checkpoint_utils import load_model_ensemble_and_task_from_hf_hub
from fairseq.models.text_to_speech.hub_interface import TTSHubInterface

#@title English multi-speaker pretrained model { run: "auto" }
lang = 'English'
tag = 'kan-bayashi/libritts_xvector_vits' 
# tag = "kan-bayashi/vctk_multi_spk_vits"
# vits needs no 
vocoder_tag = "parallel_wavegan/vctk_parallel_wavegan.v1.long" #@param ["none", "parallel_wavegan/vctk_parallel_wavegan.v1.long", "parallel_wavegan/vctk_multi_band_melgan.v2", "parallel_wavegan/vctk_style_melgan.v1", "parallel_wavegan/vctk_hifigan.v1", "parallel_wavegan/libritts_parallel_wavegan.v1.long", "parallel_wavegan/libritts_multi_band_melgan.v2", "parallel_wavegan/libritts_hifigan.v1", "parallel_wavegan/libritts_style_melgan.v1"] {type:"string"}
from espnet2.bin.tts_inference import Text2Speech
from espnet2.utils.types import str_or_none

text2speech = Text2Speech.from_pretrained(
    model_tag=str_or_none(tag),
    vocoder_tag=str_or_none(vocoder_tag),
    device="cuda",
    use_att_constraint=False,
    backward_window=1,
    forward_window=3,
    speed_control_alpha=1.0,
)


import glob
import os
import numpy as np
import kaldiio

# Get model directory path
from espnet_model_zoo.downloader import ModelDownloader
d = ModelDownloader()
model_dir = os.path.dirname(d.download_and_unpack(tag)["train_config"])

# Speaker x-vector selection

xvector_ark = [p for p in glob.glob(f"{model_dir}/../../dump/**/spk_xvector.ark", recursive=True) if "tr" in p][0]
xvectors = {k: v for k, v in kaldiio.load_ark(xvector_ark)}
spks = list(xvectors.keys())

male_spks = {"M1": "2300_131720", "M2": "1320_122612", "M3": "1188_133604", "M4": "61_70970"}
female_spks = {"F1": "2961_961", "F2": "8463_287645", "F3": "121_121726"}
spks = dict(male_spks, **female_spks)
spk_names = sorted(spks.keys())

def ASRTTS(audio_file, spk_name, ref_text=""):
    spk = spks[spk_name]
    spembs = xvectors[spk]
    if ref_text == "":
        reg_text = transcriber(audio_file)['text']
    else:
        reg_text = ref_text
    
    speech, sr = torchaudio.load(audio_file, channels_first=True) # Mono channel
    wav_tensor_spembs = text2speech(text=reg_text, speech=speech, spembs=spembs)["wav"]
    wav_numpy = wav_tensor_spembs.unsqueeze(1).to("cpu")
    sample_rate = 22050
    save_id = "./wav/" + Path(audio_file).stem + "_" + spk_name +"_spkembs.wav"
    torchaudio.save(save_id, src=wav_tensor_spembs.unsqueeze(0).to("cpu"), sample_rate=22050)

    return save_id, reg_text

def ref_reg_callback(audio_file, spk_name, ref_text):
    reg_text = ref_text
    return audio_file, spk_name, reg_text
    
reference_textbox = gr.Textbox(
    value="",
    placeholder="Input reference here",
    label="Reference",
)

recognization_textbox = gr.Textbox(
    value="",
    placeholder="Output recognization here",
    label="recognization_textbox",
)

speaker_option = gr.Radio(choices=spk_names, label="Speaker")

input_audio = gr.Audio(
            source="microphone",
            type="filepath",
            label="Audio_to_Evaluate"
)
output_audio = gr.Audio(
        source="upload",
        file="filepath",
        label="Synthesized Audio"
)
examples = [["./samples/001.wav",'M1', ""],
            ["./samples/002.wav",'M2', ""],
            ["./samples/003.wav",'F1', ""],
            ["./samples/004.wav",'F2', ""]]

# ASRTTS(*examples[0])
iface = gr.Interface(
    fn = ASRTTS, 
    inputs = [
        input_audio,
        speaker_option,
        reference_textbox,
    ],
    outputs = [
        output_audio,
        recognization_textbox
    ],
    examples = examples
)
iface.input_callback = ref_reg_callback
iface.launch(share=False)