File size: 2,700 Bytes
499588c de1d53c e715e06 faa0b89 bd639ee e715e06 bd639ee e715e06 ddd7573 86df8f3 e715e06 017c55e ddd7573 e715e06 bd639ee e715e06 956d8c0 e715e06 956d8c0 e715e06 |
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 |
import os
import spaces
import gradio as gr
import librosa
import numpy as np
import torch
from speechbrain.inference import EncoderClassifier
from transformers import pipeline
synthesiser = pipeline("text-to-speech", "techiaith/microsoft_speecht5_finetuned_bu_tts_cy_en")
speaker_embeddings = {
"GGP": "spkemb/speaker0.npy",
"BGP": "spkemb/speaker1.npy",
"BDP": "spkemb/speaker2.npy",
}
spk_model_name = "speechbrain/spkrec-xvect-voxceleb"
device = "cuda" if torch.cuda.is_available() else "cpu"
print(f">>>>> DEVICE {device}")
speaker_model = EncoderClassifier.from_hparams(
source=spk_model_name,
run_opts={"device": device},
savedir=os.path.join("/tmp", spk_model_name),
)
def create_speaker_embedding(waveform):
with torch.no_grad():
se = speaker_model.encode_batch(torch.tensor(waveform))
se = torch.nn.functional.normalize(se, dim=2)
se = se.squeeze().cpu().numpy()
return se
@spaces.GPU
def predict(text, speaker):
if len(text.strip()) == 0:
return (16000, np.zeros(0).astype(np.int16))
speaker_embedding = np.load(speaker_embeddings[speaker[:3]])
speaker_embedding = prepare_dataset(speaker_embedding)
speaker_embedding = torch.tensor(speaker_embedding).unsqueeze(0)
speech = synthesiser(text, forward_params={"speaker_embeddings": speaker_embedding})
speech = (speech.numpy() * 32767).astype(np.int16)
return (16000, speech)
title = "Techiaith Finetune Microsoft/SpeechT5: Speech Synthesis"
description = """
Lleisiau TTS microsoft_speech_T5_finetune_bu_tts_cy_en
"""
examples = [
["Rhyfeddod neu ffenomenon optegol a meteorolegol yw enfys, pan fydd sbectrwm o olau yn ymddangos yn yr awyr pan fo'r haul yn disgleirio ar ddiferion o leithder yn atmosffer y ddaear.", "GGP (gwryw-gogledd-pro)"],
["Rhyfeddod neu ffenomenon optegol a meteorolegol yw enfys, pan fydd sbectrwm o olau yn ymddangos yn yr awyr pan fo'r haul yn disgleirio ar ddiferion o leithder yn atmosffer y ddaear.", "BGP (benyw-gogledd-pro)"],
["Rhyfeddod neu ffenomenon optegol a meteorolegol yw enfys, pan fydd sbectrwm o olau yn ymddangos yn yr awyr pan fo'r haul yn disgleirio ar ddiferion o leithder yn atmosffer y ddaear.", "BDP (benyw-de-pro)"],
]
gr.Interface(
fn=predict,
inputs=[
gr.Text(label="Input Text"),
gr.Radio(label="Speaker", choices=[
"GGP (gwryw-gogledd-pro)",
"BGP (benyw-gogledd-pro)",
"BDP (benyw-de-pro)",
],
value="GGP (gwryw-gogledd-pro)"),
],
outputs=[
gr.Audio(label="Generated Speech", type="numpy"),
],
title=title,
description=description,
examples=examples,
).launch()
|