|
|
import gradio as gr |
|
|
import torch |
|
|
import torchaudio |
|
|
import re |
|
|
import os |
|
|
from transformers import SpeechT5Processor, SpeechT5ForTextToSpeech, SpeechT5HifiGan |
|
|
from speechbrain.pretrained import EncoderClassifier |
|
|
|
|
|
device = "cuda" if torch.cuda.is_available() else "cpu" |
|
|
|
|
|
|
|
|
processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts") |
|
|
model = SpeechT5ForTextToSpeech.from_pretrained("Somalitts/8aad").to(device) |
|
|
vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device) |
|
|
|
|
|
speaker_model = EncoderClassifier.from_hparams( |
|
|
source="speechbrain/spkrec-xvect-voxceleb", |
|
|
run_opts={"device": device}, |
|
|
savedir="./spk_model" |
|
|
) |
|
|
|
|
|
|
|
|
EMB_PATH = "speaker_embedding.pt" |
|
|
if os.path.exists(EMB_PATH): |
|
|
speaker_embedding = torch.load(EMB_PATH).to(device) |
|
|
else: |
|
|
audio, sr = torchaudio.load("1.wav") |
|
|
audio = torchaudio.functional.resample(audio, sr, 16000).mean(dim=0).unsqueeze(0).to(device) |
|
|
with torch.no_grad(): |
|
|
emb = speaker_model.encode_batch(audio) |
|
|
emb = torch.nn.functional.normalize(emb, dim=2).squeeze() |
|
|
torch.save(emb.cpu(), EMB_PATH) |
|
|
speaker_embedding = emb |
|
|
|
|
|
|
|
|
number_words = { |
|
|
0: "eber", 1: "koow", 2: "labo", 3: "seddex", 4: "afar", 5: "shan", |
|
|
6: "lix", 7: "todobo", 8: "sideed", 9: "sagaal", 10: "toban", |
|
|
11: "toban iyo koow", 12: "toban iyo labo", 13: "toban iyo seddex", |
|
|
14: "toban iyo afar", 15: "toban iyo shan", 16: "toban iyo lix", |
|
|
17: "toban iyo todobo", 18: "toban iyo sideed", 19: "toban iyo sagaal", |
|
|
20: "labaatan", 30: "sodon", 40: "afartan", 50: "konton", |
|
|
60: "lixdan", 70: "todobaatan", 80: "sideetan", 90: "sagaashan", |
|
|
100: "boqol", 1000: "kun", |
|
|
} |
|
|
|
|
|
def number_to_words(number): |
|
|
if number < 20: |
|
|
return number_words[number] |
|
|
elif number < 100: |
|
|
tens, unit = divmod(number, 10) |
|
|
return number_words[tens * 10] + (" " + number_words[unit] if unit else "") |
|
|
elif number < 1000: |
|
|
hundreds, remainder = divmod(number, 100) |
|
|
return (number_words[hundreds] + " boqol" if hundreds > 1 else "BOQOL") + (" " + number_to_words(remainder) if remainder else "") |
|
|
elif number < 1000000: |
|
|
thousands, remainder = divmod(number, 1000) |
|
|
return (number_to_words(thousands) + " kun" if thousands > 1 else "KUN") + (" " + number_to_words(remainder) if remainder else "") |
|
|
elif number < 1000000000: |
|
|
millions, remainder = divmod(number, 1000000) |
|
|
return number_to_words(millions) + " malyan" + (" " + number_to_words(remainder) if remainder else "") |
|
|
elif number < 1000000000000: |
|
|
billions, remainder = divmod(number, 1000000000) |
|
|
return number_to_words(billions) + " milyaar" + (" " + number_to_words(remainder) if remainder else "") |
|
|
else: |
|
|
return str(number) |
|
|
|
|
|
def replace_numbers_with_words(text): |
|
|
def replace(match): |
|
|
number = int(match.group()) |
|
|
return number_to_words(number) |
|
|
return re.sub(r'\b\d+\b', replace, text) |
|
|
|
|
|
def normalize_text(text): |
|
|
text = text.lower() |
|
|
text = replace_numbers_with_words(text) |
|
|
text = re.sub(r'[^\w\s]', '', text) |
|
|
return text |
|
|
|
|
|
|
|
|
def text_to_speech(text): |
|
|
text = normalize_text(text) |
|
|
inputs = processor(text=text, return_tensors="pt").to(device) |
|
|
with torch.no_grad(): |
|
|
speech = model.generate_speech(inputs["input_ids"], speaker_embedding.unsqueeze(0), vocoder=vocoder) |
|
|
return (16000, speech.cpu().numpy()) |
|
|
|
|
|
|
|
|
iface = gr.Interface( |
|
|
fn=text_to_speech, |
|
|
inputs=gr.Textbox(label="Geli qoraalka af-soomaali"), |
|
|
outputs=gr.Audio(label="Codka la abuuray", type="numpy"), |
|
|
title="Somali TTS", |
|
|
description="TTS Soomaaliyeed oo la adeegsaday cod gaar ah (11.wav)" |
|
|
) |
|
|
|
|
|
iface.launch() |
|
|
|
