Spaces:
Running
on
T4
Running
on
T4
File size: 3,928 Bytes
0c98d02 85d7512 a95f5b7 0c98d02 84a8081 a95f5b7 84a8081 85d7512 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 |
import gradio as gr
"""
Audio processing tools to convert between spectrogram images and waveforms.
"""
import io
import typing as T
import numpy as np
from PIL import Image
import pydub
from scipy.io import wavfile
import torch
import torchaudio
from diffusers import StableDiffusionPipeline
model_id = "riffusion/riffusion-model-v1"
pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16)
pipe = pipe.to("cuda")
def get_spectro(prompt):
image = pipe(prompt).images[0]
return image
def wav_bytes_from_spectrogram_image(image: Image.Image) -> T.Tuple[io.BytesIO, float]:
"""
Reconstruct a WAV audio clip from a spectrogram image. Also returns the duration in seconds.
"""
max_volume = 50
power_for_image = 0.25
Sxx = spectrogram_from_image(image, max_volume=max_volume, power_for_image=power_for_image)
sample_rate = 44100 # [Hz]
clip_duration_ms = 5000 # [ms]
bins_per_image = 512
n_mels = 512
# FFT parameters
window_duration_ms = 100 # [ms]
padded_duration_ms = 400 # [ms]
step_size_ms = 10 # [ms]
# Derived parameters
num_samples = int(image.width / float(bins_per_image) * clip_duration_ms) * sample_rate
n_fft = int(padded_duration_ms / 1000.0 * sample_rate)
hop_length = int(step_size_ms / 1000.0 * sample_rate)
win_length = int(window_duration_ms / 1000.0 * sample_rate)
samples = waveform_from_spectrogram(
Sxx=Sxx,
n_fft=n_fft,
hop_length=hop_length,
win_length=win_length,
num_samples=num_samples,
sample_rate=sample_rate,
mel_scale=True,
n_mels=n_mels,
max_mel_iters=200,
num_griffin_lim_iters=32,
)
wav_bytes = io.BytesIO()
wavfile.write(wav_bytes, sample_rate, samples.astype(np.int16))
wav_bytes.seek(0)
duration_s = float(len(samples)) / sample_rate
return wav_bytes
def spectrogram_from_image(
image: Image.Image, max_volume: float = 50, power_for_image: float = 0.25
) -> np.ndarray:
"""
Compute a spectrogram magnitude array from a spectrogram image.
TODO(hayk): Add image_from_spectrogram and call this out as the reverse.
"""
# Convert to a numpy array of floats
data = np.array(image).astype(np.float32)
# Flip Y take a single channel
data = data[::-1, :, 0]
# Invert
data = 255 - data
# Rescale to max volume
data = data * max_volume / 255
# Reverse the power curve
data = np.power(data, 1 / power_for_image)
return data
def waveform_from_spectrogram(
Sxx: np.ndarray,
n_fft: int,
hop_length: int,
win_length: int,
num_samples: int,
sample_rate: int,
mel_scale: bool = True,
n_mels: int = 512,
max_mel_iters: int = 200,
num_griffin_lim_iters: int = 32,
device: str = "cuda:0",
) -> np.ndarray:
"""
Reconstruct a waveform from a spectrogram.
This is an approximate inverse of spectrogram_from_waveform, using the Griffin-Lim algorithm
to approximate the phase.
"""
Sxx_torch = torch.from_numpy(Sxx).to(device)
# TODO(hayk): Make this a class that caches the two things
if mel_scale:
mel_inv_scaler = torchaudio.transforms.InverseMelScale(
n_mels=n_mels,
sample_rate=sample_rate,
f_min=0,
f_max=10000,
n_stft=n_fft // 2 + 1,
norm=None,
mel_scale="htk",
max_iter=max_mel_iters,
).to(device)
Sxx_torch = mel_inv_scaler(Sxx_torch)
griffin_lim = torchaudio.transforms.GriffinLim(
n_fft=n_fft,
win_length=win_length,
hop_length=hop_length,
power=1.0,
n_iter=num_griffin_lim_iters,
).to(device)
waveform = griffin_lim(Sxx_torch).cpu().numpy()
return waveform
gr.Interface(fn=get_spectro, inputs=[gr.Textbox()], outputs=[gr.Image()]).launch() |