Spaces:
Sleeping
Sleeping
Xu Xuenan
commited on
Commit
·
676ec69
1
Parent(s):
5152717
Transformers MusicGen
Browse files- app.py +2 -10
- configs/mm_story_agent.yaml +1 -2
- mm_story_agent/modality_agents/music_agent.py +13 -9
app.py
CHANGED
@@ -97,7 +97,7 @@ def write_story_fn(story_topic, main_role, scene,
|
|
97 |
|
98 |
def modality_assets_generation_fn(
|
99 |
height, width, image_seed, sound_guidance_scale, sound_seed,
|
100 |
-
n_candidate_per_text,
|
101 |
config,
|
102 |
story_data):
|
103 |
deep_update(config, {
|
@@ -117,11 +117,6 @@ def modality_assets_generation_fn(
|
|
117 |
"n_candidate_per_text": n_candidate_per_text
|
118 |
}
|
119 |
},
|
120 |
-
"music_generation": {
|
121 |
-
"call_cfg": {
|
122 |
-
"duration": music_duration
|
123 |
-
}
|
124 |
-
}
|
125 |
})
|
126 |
story_gen_agent = MMStoryAgent()
|
127 |
images = story_gen_agent.generate_modality_assets(config, story_data)
|
@@ -180,9 +175,6 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
180 |
sound_seed = gr.Number(label="Sound Seed", value=default_sound_config["call_cfg"]['seed'])
|
181 |
n_candidate_per_text = gr.Slider(label="Number of Candidates per Text", minimum=0, maximum=5, step=1, value=default_sound_config["call_cfg"]['n_candidate_per_text'])
|
182 |
|
183 |
-
with gr.Accordion("Detailed Music Configuration (Optional)", open=False):
|
184 |
-
music_duration = gr.Number(label="Music Duration", min_width=30.0, maximum=120.0, value=default_music_config["call_cfg"]["duration"])
|
185 |
-
|
186 |
with gr.Accordion("Detailed Slideshow Effect (Optional)", open=False):
|
187 |
fade_duration = gr.Slider(label="Fade Duration", minimum=0.1, maximum=1.5, step=0.1, value=default_slideshow_effect['fade_duration'])
|
188 |
slide_duration = gr.Slider(label="Slide Duration", minimum=0.1, maximum=1.0, step=0.1, value=default_slideshow_effect['slide_duration'])
|
@@ -244,7 +236,7 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
244 |
).then(
|
245 |
fn=modality_assets_generation_fn,
|
246 |
inputs=[height, width, image_seed, sound_guidance_scale, sound_seed,
|
247 |
-
n_candidate_per_text,
|
248 |
config,
|
249 |
story_data],
|
250 |
outputs=[image_gallery]
|
|
|
97 |
|
98 |
def modality_assets_generation_fn(
|
99 |
height, width, image_seed, sound_guidance_scale, sound_seed,
|
100 |
+
n_candidate_per_text,
|
101 |
config,
|
102 |
story_data):
|
103 |
deep_update(config, {
|
|
|
117 |
"n_candidate_per_text": n_candidate_per_text
|
118 |
}
|
119 |
},
|
|
|
|
|
|
|
|
|
|
|
120 |
})
|
121 |
story_gen_agent = MMStoryAgent()
|
122 |
images = story_gen_agent.generate_modality_assets(config, story_data)
|
|
|
175 |
sound_seed = gr.Number(label="Sound Seed", value=default_sound_config["call_cfg"]['seed'])
|
176 |
n_candidate_per_text = gr.Slider(label="Number of Candidates per Text", minimum=0, maximum=5, step=1, value=default_sound_config["call_cfg"]['n_candidate_per_text'])
|
177 |
|
|
|
|
|
|
|
178 |
with gr.Accordion("Detailed Slideshow Effect (Optional)", open=False):
|
179 |
fade_duration = gr.Slider(label="Fade Duration", minimum=0.1, maximum=1.5, step=0.1, value=default_slideshow_effect['fade_duration'])
|
180 |
slide_duration = gr.Slider(label="Slide Duration", minimum=0.1, maximum=1.0, step=0.1, value=default_slideshow_effect['slide_duration'])
|
|
|
236 |
).then(
|
237 |
fn=modality_assets_generation_fn,
|
238 |
inputs=[height, width, image_seed, sound_guidance_scale, sound_seed,
|
239 |
+
n_candidate_per_text,
|
240 |
config,
|
241 |
story_data],
|
242 |
outputs=[image_gallery]
|
configs/mm_story_agent.yaml
CHANGED
@@ -56,8 +56,7 @@ music_generation:
|
|
56 |
revise_cfg:
|
57 |
num_turns: 3
|
58 |
obj_cfg: {}
|
59 |
-
call_cfg:
|
60 |
-
duration: 60.0
|
61 |
|
62 |
slideshow_effect:
|
63 |
fade_duration: 0.8
|
|
|
56 |
revise_cfg:
|
57 |
num_turns: 3
|
58 |
obj_cfg: {}
|
59 |
+
call_cfg: {}
|
|
|
60 |
|
61 |
slideshow_effect:
|
62 |
fade_duration: 0.8
|
mm_story_agent/modality_agents/music_agent.py
CHANGED
@@ -2,9 +2,9 @@ from pathlib import Path
|
|
2 |
import json
|
3 |
from typing import List, Union
|
4 |
|
|
|
5 |
import torchaudio
|
6 |
-
from
|
7 |
-
from audiocraft.data.audio import audio_write
|
8 |
|
9 |
from mm_story_agent.modality_agents.llm import QwenAgent
|
10 |
from mm_story_agent.prompts_en import story_to_music_reviser_system, story_to_music_reviewer_system
|
@@ -17,19 +17,23 @@ class MusicGenSynthesizer:
|
|
17 |
device: str = 'cuda',
|
18 |
sample_rate: int = 16000,
|
19 |
) -> None:
|
20 |
-
self.
|
|
|
|
|
21 |
self.sample_rate = sample_rate
|
22 |
|
23 |
def call(self,
|
24 |
prompt: Union[str, List[str]],
|
25 |
save_path: Union[str, Path],
|
26 |
-
duration: float = 60.0,
|
27 |
):
|
28 |
-
self.
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
|
|
|
|
|
|
33 |
|
34 |
|
35 |
class MusicGenAgent:
|
|
|
2 |
import json
|
3 |
from typing import List, Union
|
4 |
|
5 |
+
import soundfile as sf
|
6 |
import torchaudio
|
7 |
+
from transformers import AutoProcessor, MusicgenForConditionalGeneration
|
|
|
8 |
|
9 |
from mm_story_agent.modality_agents.llm import QwenAgent
|
10 |
from mm_story_agent.prompts_en import story_to_music_reviser_system, story_to_music_reviewer_system
|
|
|
17 |
device: str = 'cuda',
|
18 |
sample_rate: int = 16000,
|
19 |
) -> None:
|
20 |
+
self.device = device
|
21 |
+
self.processor = AutoProcessor.from_pretrained(model_name)
|
22 |
+
self.model = MusicgenForConditionalGeneration.from_pretrained(model_name).to(device)
|
23 |
self.sample_rate = sample_rate
|
24 |
|
25 |
def call(self,
|
26 |
prompt: Union[str, List[str]],
|
27 |
save_path: Union[str, Path],
|
|
|
28 |
):
|
29 |
+
inputs = self.processor(
|
30 |
+
text=[prompt],
|
31 |
+
padding=True,
|
32 |
+
return_tensors="pt",
|
33 |
+
).to(self.device)
|
34 |
+
wav = self.model.generate(**inputs, max_new_tokens=1536)[0, 0].cpu()
|
35 |
+
wav = torchaudio.functional.resample(wav, self.model.config.audio_encoder.sampling_rate, self.sample_rate)
|
36 |
+
sf.write(save_path, wav.numpy(), self.sample_rate)
|
37 |
|
38 |
|
39 |
class MusicGenAgent:
|