File size: 3,580 Bytes
090156b 41f1fc2 090156b 41f1fc2 090156b 3d08f21 090156b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 |
import os
from typing import Text
import gradio as gr
import soundfile as sf
from transformers import pipeline
import numpy as np
import torch
import re
from speechbrain.pretrained import EncoderClassifier
def create_speaker_embedding(speaker_model, waveform: np.ndarray) -> np.ndarray:
with torch.no_grad():
speaker_embeddings = speaker_model.encode_batch(torch.tensor(waveform))
speaker_embeddings = torch.nn.functional.normalize(speaker_embeddings, dim=2)
if device.type != 'cuda':
speaker_embeddings = speaker_embeddings.squeeze().numpy()
else:
speaker_embeddings = speaker_embeddings.squeeze().cpu().numpy()
speaker_embeddings = torch.tensor(speaker_embeddings, dtype=dtype).unsqueeze(0).to(device)
return speaker_embeddings
def remove_special_characters_s(text: Text) -> Text:
chars_to_remove_regex = '[\=\´\–\“\”\…\=]'
# remove special characters
text = re.sub(chars_to_remove_regex, '', text)
text = re.sub("‘", "'", text)
text = re.sub("’", "'", text)
text = re.sub("´", "'", text)
text = text.lower()
return text
def dutch_to_english(text: Text) -> Text:
replacements = [
("à", "a"),
("ç", "c"),
("è", "e"),
("ë", "e"),
("í", "i"),
("ï", "i"),
("ö", "o"),
("ü", "u"),
('&', "en"),
('á','a'),
('ä','a'),
('î','i'),
('ó','o'),
('ö','o'),
('ú','u'),
('û','u'),
('ă','a'),
('ć','c'),
('đ','d'),
('š','s'),
('ţ','t'),
('j', 'y'),
('k', 'k'),
('ci', 'si'),
('ce', 'se'),
('ca', 'ka'),
('co', 'ko'),
('cu', 'ku'),
(' sch', ' sg'),
('sch ', 's '),
('ch', 'g'),
('eeuw', 'eaw'),
('ee', 'ea'),
('aai','ay'),
('oei', 'ooy'),
('ooi', 'oay'),
('ieuw', 'eew'),
('ie', 'ee'),
('oo', 'oa'),
('oe', 'oo'),
('ei', '\\i\\'),
('ij', 'i'),
('\\i\\', 'i')
]
for src, dst in replacements:
text = text.replace(src, dst)
return text
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
if torch.cuda.is_available():
dtype = torch.bfloat16 if torch.cuda.get_device_capability()[0] == 8 else torch.float16
else:
dtype = torch.float32
spk_model_name = "speechbrain/spkrec-xvect-voxceleb"
speaker_model = EncoderClassifier.from_hparams(
source=spk_model_name,
run_opts={"device": device},
savedir=os.path.join("/tmp", spk_model_name)
)
waveform, samplerate = sf.read("files/speaker.wav")
speaker_embeddings = create_speaker_embedding(speaker_model, waveform)
transcriber = pipeline("text-to-speech", model="Oysiyl/speecht5_tts_common_voice_nl")
def transcribe(text: Text) -> tuple((int, np.ndarray)):
text = remove_special_characters_s(text)
text = dutch_to_english(text)
out = transcriber(text, forward_params={"speaker_embeddings": speaker_embeddings})
audio, sr = out["audio"], out["sampling_rate"]
return sr, audio
demo = gr.Interface(
transcribe,
gr.Textbox(),
outputs="audio",
title="Text to Speech for Dutch language demo",
description="Click on the example below or type text!",
examples=[["Goedenavond, ik kom uit Oekraïne!"],
["Hallo allemaal, ik praat nederlands. Groetjes aan iedereen!"]],
cache_examples=True
)
demo.launch() |