Commit ·
c131f40
1
Parent(s): 1ae750e
Upload files
Browse files- inference.py +67 -0
- requirements.txt +18 -0
inference.py
ADDED
|
@@ -0,0 +1,67 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# @title Imports
|
| 2 |
+
from diffusers import DiffusionPipeline
|
| 3 |
+
from riffusion.spectrogram_image_converter import SpectrogramImageConverter
|
| 4 |
+
from riffusion.spectrogram_params import SpectrogramParams
|
| 5 |
+
from io import BytesIO
|
| 6 |
+
|
| 7 |
+
|
| 8 |
+
# @title Define a `predict` function
|
| 9 |
+
|
| 10 |
+
params = SpectrogramParams()
|
| 11 |
+
converter = SpectrogramImageConverter(params)
|
| 12 |
+
|
| 13 |
+
|
| 14 |
+
def preprocess_function(text):
|
| 15 |
+
with open(text, "r", encoding="utf-8") as f:
|
| 16 |
+
data = f.read()
|
| 17 |
+
print(data)
|
| 18 |
+
# pass the textand the target tanguage to be translated separated by a ";" semicolon
|
| 19 |
+
# data = text_path.read().decode("utf-8")
|
| 20 |
+
prompt = data.split(";")[0]
|
| 21 |
+
negative_prompt = data.split(";")[1].strip()
|
| 22 |
+
print(negative_prompt.strip())
|
| 23 |
+
print(data)
|
| 24 |
+
return (prompt, negative_prompt)
|
| 25 |
+
|
| 26 |
+
|
| 27 |
+
def predict_function(params, pipe):
|
| 28 |
+
prompt, negative_prompt = params
|
| 29 |
+
spec = pipe(
|
| 30 |
+
prompt,
|
| 31 |
+
negative_prompt=negative_prompt,
|
| 32 |
+
width=768,
|
| 33 |
+
).images[0]
|
| 34 |
+
|
| 35 |
+
wav = converter.audio_from_spectrogram_image(image=spec)
|
| 36 |
+
wav.export("output.wav", format="wav")
|
| 37 |
+
return ("output.wav", spec)
|
| 38 |
+
|
| 39 |
+
|
| 40 |
+
def model_load_function(model_path):
|
| 41 |
+
pipe = DiffusionPipeline.from_pretrained(model_path)
|
| 42 |
+
pipe = pipe.to("cuda")
|
| 43 |
+
return pipe
|
| 44 |
+
|
| 45 |
+
|
| 46 |
+
def postprocess_function(audio_file, content_type=None):
|
| 47 |
+
audio = open(audio_file, "rb")
|
| 48 |
+
audio = audio.read()
|
| 49 |
+
print(type(audio))
|
| 50 |
+
audio_bytes = BytesIO(audio)
|
| 51 |
+
response = dict()
|
| 52 |
+
audio_bytes.seek(0)
|
| 53 |
+
response["output"] = {"data": audio_bytes, "ext": "wav"}
|
| 54 |
+
return response
|
| 55 |
+
|
| 56 |
+
|
| 57 |
+
## Test the script
|
| 58 |
+
"""
|
| 59 |
+
if __name__ == '__main__':
|
| 60 |
+
text = ""
|
| 61 |
+
data = preprocess_function(text)
|
| 62 |
+
model_path = "./model_files"
|
| 63 |
+
path = model_load_function(model_path)
|
| 64 |
+
predictions = predict_function(data,path)
|
| 65 |
+
out = postprocess_function(audio_file)
|
| 66 |
+
print(out)
|
| 67 |
+
"""
|
requirements.txt
ADDED
|
@@ -0,0 +1,18 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
accelerate
|
| 2 |
+
argh
|
| 3 |
+
dacite
|
| 4 |
+
demucs
|
| 5 |
+
diffusers
|
| 6 |
+
numpy
|
| 7 |
+
pillow
|
| 8 |
+
plotly
|
| 9 |
+
pydub
|
| 10 |
+
pysoundfile
|
| 11 |
+
scipy
|
| 12 |
+
soundfile
|
| 13 |
+
sox
|
| 14 |
+
torch
|
| 15 |
+
torchaudio
|
| 16 |
+
torchvision
|
| 17 |
+
transformers
|
| 18 |
+
riffusion
|