Spaces:
Runtime error
Runtime error
clean up UI
Browse files
app.py
CHANGED
@@ -18,7 +18,6 @@ from audiocraft.data.audio import audio_write
|
|
18 |
from share_btn import community_icon_html, loading_icon_html, share_js, css
|
19 |
|
20 |
MODEL = None
|
21 |
-
IS_SHARED_SPACE = "radames/MusicGen-Continuation" in os.environ.get("SPACE_ID", "")
|
22 |
|
23 |
|
24 |
def load_model(version):
|
@@ -30,7 +29,6 @@ def predict(
|
|
30 |
text,
|
31 |
melody_input,
|
32 |
duration=30,
|
33 |
-
continuation=False,
|
34 |
continuation_start=0,
|
35 |
continuation_end=30,
|
36 |
topk=250,
|
@@ -42,10 +40,13 @@ def predict(
|
|
42 |
topk = int(topk)
|
43 |
if MODEL is None:
|
44 |
MODEL = load_model("melody")
|
|
|
|
|
|
|
45 |
|
46 |
if duration > MODEL.lm.cfg.dataset.segment_duration:
|
47 |
raise gr.Error("MusicGen currently supports durations of up to 30 seconds!")
|
48 |
-
if
|
49 |
raise gr.Error("The end time must be greater than the start time!")
|
50 |
MODEL.set_generation_params(
|
51 |
use_sampling=True,
|
@@ -61,34 +62,19 @@ def predict(
|
|
61 |
# sr, melody = melody_input[0], torch.from_numpy(melody_input[1]).to(MODEL.device).float().t().unsqueeze(0)
|
62 |
if melody.dim() == 2:
|
63 |
melody = melody[None]
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
)
|
78 |
-
else:
|
79 |
-
print("\nGenerating with melody\n")
|
80 |
-
melody_wavform = melody[
|
81 |
-
..., : int(sr * MODEL.lm.cfg.dataset.segment_duration)
|
82 |
-
]
|
83 |
-
output = MODEL.generate_with_chroma(
|
84 |
-
descriptions=[text],
|
85 |
-
melody_wavs=melody_wavform,
|
86 |
-
melody_sample_rate=sr,
|
87 |
-
progress=True,
|
88 |
-
)
|
89 |
-
else:
|
90 |
-
print("\nGenerating without melody\n")
|
91 |
-
output = MODEL.generate(descriptions=[text], progress=False)
|
92 |
|
93 |
output = output.detach().cpu().float()[0]
|
94 |
with NamedTemporaryFile("wb", suffix=".wav", delete=False) as file:
|
@@ -150,20 +136,19 @@ def ui(**kwargs):
|
|
150 |
gr.Markdown(
|
151 |
"""
|
152 |
# MusicGen Continuation
|
153 |
-
This
|
154 |
presented at: ["Simple and Controllable Music Generation"](https://huggingface.co/papers/2306.05284)
|
|
|
|
|
155 |
"""
|
156 |
)
|
157 |
-
|
158 |
-
|
159 |
-
"""
|
160 |
-
⚠ This Space doesn't work in this shared UI ⚠
|
161 |
-
|
162 |
<a href="https://huggingface.co/spaces/radames/MusicGen-Continuation?duplicate=true" style="display: inline-block;margin-top: .5em;margin-right: .25em;" target="_blank">
|
163 |
<img style="margin-bottom: 0em;display: inline;margin-top: -.25em;" src="https://bit.ly/3gLdBN6" alt="Duplicate Space"></a>
|
164 |
-
to use it privately
|
165 |
"""
|
166 |
-
|
167 |
with gr.Row():
|
168 |
with gr.Column():
|
169 |
with gr.Row():
|
@@ -177,7 +162,8 @@ def ui(**kwargs):
|
|
177 |
radio = gr.Radio(
|
178 |
["file", "mic"],
|
179 |
value="file",
|
180 |
-
label="Melody Condition
|
|
|
181 |
)
|
182 |
melody = gr.Audio(
|
183 |
source="upload",
|
@@ -188,46 +174,38 @@ def ui(**kwargs):
|
|
188 |
)
|
189 |
with gr.Row():
|
190 |
submit = gr.Button("Submit")
|
191 |
-
# with gr.Row():
|
192 |
-
# model = gr.Radio(
|
193 |
-
# ["melody", "medium", "small", "large"],
|
194 |
-
# label="Model",
|
195 |
-
# value="melody",
|
196 |
-
# interactive=True,
|
197 |
-
# )
|
198 |
with gr.Row():
|
199 |
duration = gr.Slider(
|
200 |
minimum=1,
|
201 |
maximum=30,
|
202 |
value=10,
|
203 |
-
label="Total Duration",
|
204 |
interactive=True,
|
205 |
)
|
206 |
-
with gr.
|
207 |
-
|
208 |
-
|
209 |
-
|
210 |
-
|
211 |
-
|
212 |
-
|
213 |
-
|
214 |
-
|
215 |
-
|
216 |
-
|
217 |
-
|
218 |
-
|
219 |
-
|
220 |
-
|
221 |
-
|
222 |
-
|
223 |
-
|
224 |
-
|
225 |
-
|
226 |
-
|
227 |
-
|
228 |
-
|
229 |
-
|
230 |
-
)
|
231 |
with gr.Accordion(label="Advanced Settings", open=False):
|
232 |
with gr.Row():
|
233 |
topk = gr.Number(label="Top-k", value=250, interactive=True)
|
@@ -276,7 +254,6 @@ def ui(**kwargs):
|
|
276 |
text,
|
277 |
melody,
|
278 |
duration,
|
279 |
-
continuation,
|
280 |
continuation_start,
|
281 |
continuation_end,
|
282 |
topk,
|
|
|
18 |
from share_btn import community_icon_html, loading_icon_html, share_js, css
|
19 |
|
20 |
MODEL = None
|
|
|
21 |
|
22 |
|
23 |
def load_model(version):
|
|
|
29 |
text,
|
30 |
melody_input,
|
31 |
duration=30,
|
|
|
32 |
continuation_start=0,
|
33 |
continuation_end=30,
|
34 |
topk=250,
|
|
|
40 |
topk = int(topk)
|
41 |
if MODEL is None:
|
42 |
MODEL = load_model("melody")
|
43 |
+
|
44 |
+
if melody_input is None:
|
45 |
+
raise gr.Error("Please upload a melody to continue!")
|
46 |
|
47 |
if duration > MODEL.lm.cfg.dataset.segment_duration:
|
48 |
raise gr.Error("MusicGen currently supports durations of up to 30 seconds!")
|
49 |
+
if continuation_end < continuation_start:
|
50 |
raise gr.Error("The end time must be greater than the start time!")
|
51 |
MODEL.set_generation_params(
|
52 |
use_sampling=True,
|
|
|
62 |
# sr, melody = melody_input[0], torch.from_numpy(melody_input[1]).to(MODEL.device).float().t().unsqueeze(0)
|
63 |
if melody.dim() == 2:
|
64 |
melody = melody[None]
|
65 |
+
print("\nGenerating continuation\n")
|
66 |
+
melody_wavform = melody[
|
67 |
+
..., int(sr * continuation_start) : int(sr * continuation_end)
|
68 |
+
]
|
69 |
+
melody_duration = melody_wavform.shape[-1] / sr
|
70 |
+
if duration + melody_duration > MODEL.lm.cfg.dataset.segment_duration:
|
71 |
+
raise gr.Error("Duration + continuation duration must be <= 30 seconds")
|
72 |
+
output = MODEL.generate_continuation(
|
73 |
+
prompt=melody_wavform,
|
74 |
+
prompt_sample_rate=sr,
|
75 |
+
descriptions=[text],
|
76 |
+
progress=True,
|
77 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
78 |
|
79 |
output = output.detach().cpu().float()[0]
|
80 |
with NamedTemporaryFile("wb", suffix=".wav", delete=False) as file:
|
|
|
136 |
gr.Markdown(
|
137 |
"""
|
138 |
# MusicGen Continuation
|
139 |
+
This a [MusicGen](https://github.com/facebookresearch/audiocraft), a simple and controllable model for music generation
|
140 |
presented at: ["Simple and Controllable Music Generation"](https://huggingface.co/papers/2306.05284)
|
141 |
+
|
142 |
+
This Spaces only does melody continuation, you can try other features [here](https://huggingface.co/spaces/facebook/MusicGen)
|
143 |
"""
|
144 |
)
|
145 |
+
gr.Markdown(
|
146 |
+
"""
|
|
|
|
|
|
|
147 |
<a href="https://huggingface.co/spaces/radames/MusicGen-Continuation?duplicate=true" style="display: inline-block;margin-top: .5em;margin-right: .25em;" target="_blank">
|
148 |
<img style="margin-bottom: 0em;display: inline;margin-top: -.25em;" src="https://bit.ly/3gLdBN6" alt="Duplicate Space"></a>
|
149 |
+
to use it privately
|
150 |
"""
|
151 |
+
)
|
152 |
with gr.Row():
|
153 |
with gr.Column():
|
154 |
with gr.Row():
|
|
|
162 |
radio = gr.Radio(
|
163 |
["file", "mic"],
|
164 |
value="file",
|
165 |
+
label="Melody Inital Condition File or Mic",
|
166 |
+
info="Make sure the audio is no longer than total generation duration which is max 30 seconds, you can trim the audio in the next section",
|
167 |
)
|
168 |
melody = gr.Audio(
|
169 |
source="upload",
|
|
|
174 |
)
|
175 |
with gr.Row():
|
176 |
submit = gr.Button("Submit")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
177 |
with gr.Row():
|
178 |
duration = gr.Slider(
|
179 |
minimum=1,
|
180 |
maximum=30,
|
181 |
value=10,
|
182 |
+
label="Total Generation Duration",
|
183 |
interactive=True,
|
184 |
)
|
185 |
+
with gr.Accordion(label="Input Melody Trimming (optional)", open=False):
|
186 |
+
with gr.Row():
|
187 |
+
continuation_start = gr.Slider(
|
188 |
+
minimum=0,
|
189 |
+
maximum=30,
|
190 |
+
step=0.01,
|
191 |
+
value=0,
|
192 |
+
label="melody cut start",
|
193 |
+
interactive=True,
|
194 |
+
)
|
195 |
+
continuation_end = gr.Slider(
|
196 |
+
minimum=0,
|
197 |
+
maximum=30,
|
198 |
+
step=0.01,
|
199 |
+
value=0,
|
200 |
+
label="melody cut end",
|
201 |
+
interactive=True,
|
202 |
+
)
|
203 |
+
cut_btn = gr.Button("Cut Melody").style(full_width=False)
|
204 |
+
with gr.Row():
|
205 |
+
preview_cut = gr.Audio(
|
206 |
+
type="numpy",
|
207 |
+
label="Cut Preview",
|
208 |
+
)
|
|
|
209 |
with gr.Accordion(label="Advanced Settings", open=False):
|
210 |
with gr.Row():
|
211 |
topk = gr.Number(label="Top-k", value=250, interactive=True)
|
|
|
254 |
text,
|
255 |
melody,
|
256 |
duration,
|
|
|
257 |
continuation_start,
|
258 |
continuation_end,
|
259 |
topk,
|