Spaces:
Sleeping
Sleeping
File size: 3,120 Bytes
03e487a |
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 |
import gradio as gr
import librosa
import numpy as np
import torch
from transformers import SpeechT5Processor, SpeechT5ForTextToSpeech, SpeechT5HifiGan
checkpoint = "microsoft/speecht5_tts"
processor = SpeechT5Processor.from_pretrained(checkpoint)
model = SpeechT5ForTextToSpeech.from_pretrained(checkpoint)
vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")
speaker_embeddings = {
"BDL": "spkemb/cmu_us_bdl_arctic-wav-arctic_a0009.npy",
"CLB": "spkemb/cmu_us_clb_arctic-wav-arctic_a0144.npy",
"KSP": "spkemb/cmu_us_ksp_arctic-wav-arctic_b0087.npy",
"RMS": "spkemb/cmu_us_rms_arctic-wav-arctic_b0353.npy",
"SLT": "spkemb/cmu_us_slt_arctic-wav-arctic_a0508.npy",
}
def predict(text, speaker):
if len(text.strip()) == 0:
return (16000, np.zeros(0).astype(np.int16))
inputs = processor(text=text, return_tensors="pt")
# limit input length
input_ids = inputs["input_ids"]
input_ids = input_ids[..., :model.config.max_text_positions]
if speaker == "Surprise Me!":
# load one of the provided speaker embeddings at random
idx = np.random.randint(len(speaker_embeddings))
key = list(speaker_embeddings.keys())[idx]
speaker_embedding = np.load(speaker_embeddings[key])
# randomly shuffle the elements
np.random.shuffle(speaker_embedding)
# randomly flip half the values
x = (np.random.rand(512) >= 0.5) * 1.0
x[x == 0] = -1.0
speaker_embedding *= x
#speaker_embedding = np.random.rand(512).astype(np.float32) * 0.3 - 0.15
else:
speaker_embedding = np.load(speaker_embeddings[speaker[:3]])
speaker_embedding = torch.tensor(speaker_embedding).unsqueeze(0)
speech = model.generate_speech(input_ids, speaker_embedding, vocoder=vocoder)
speech = (speech.numpy() * 32767).astype(np.int16)
return (16000, speech)
examples = [
["It is not in the stars to hold our destiny but in ourselves.", "BDL (male)"],
["The octopus and Oliver went to the opera in October.", "CLB (female)"],
["She sells seashells by the seashore. I saw a kitten eating chicken in the kitchen.", "RMS (male)"],
["Brisk brave brigadiers brandished broad bright blades, blunderbusses, and bludgeons—balancing them badly.", "SLT (female)"],
["A synonym for cinnamon is a cinnamon synonym.", "BDL (male)"],
["How much wood would a woodchuck chuck if a woodchuck could chuck wood? He would chuck, he would, as much as he could, and chuck as much wood as a woodchuck would if a woodchuck could chuck wood.", "CLB (female)"],
]
gr.Interface(
fn=predict,
inputs=[
gr.Text(label="Input Text"),
gr.Radio(label="Speaker", choices=[
"BDL (male)",
"CLB (female)",
"KSP (male)",
"RMS (male)",
"SLT (female)",
"Surprise Me!"
],
value="BDL (male)"),
],
outputs=[
gr.Audio(label="Generated Speech", type="numpy"),
],
title=title,
description=description,
article=article,
examples=examples,
).launch() |