Spaces:
Runtime error
Runtime error
Use accordion
Browse files
app.py
CHANGED
|
@@ -47,12 +47,12 @@ class Tango:
|
|
| 47 |
self.scheduler = DDPMScheduler.from_pretrained(main_config["scheduler_name"], subfolder = "scheduler")
|
| 48 |
|
| 49 |
def chunks(self, lst, n):
|
| 50 |
-
|
| 51 |
for i in range(0, len(lst), n):
|
| 52 |
yield lst[i:i + n]
|
| 53 |
|
| 54 |
def generate(self, prompt, steps = 100, guidance = 3, samples = 1, disable_progress = True):
|
| 55 |
-
|
| 56 |
with torch.no_grad():
|
| 57 |
latents = self.model.inference([prompt], self.scheduler, steps, guidance, samples, disable_progress = disable_progress)
|
| 58 |
mel = self.vae.decode_first_stage(latents)
|
|
@@ -60,7 +60,7 @@ class Tango:
|
|
| 60 |
return wave[0]
|
| 61 |
|
| 62 |
def generate_for_batch(self, prompts, steps = 200, guidance = 3, samples = 1, batch_size = 8, disable_progress = True):
|
| 63 |
-
|
| 64 |
outputs = []
|
| 65 |
for k in tqdm(range(0, len(prompts), batch_size)):
|
| 66 |
batch = prompts[k: k + batch_size]
|
|
@@ -80,7 +80,19 @@ tango.vae.to(device_type)
|
|
| 80 |
tango.stft.to(device_type)
|
| 81 |
tango.model.to(device_type)
|
| 82 |
|
| 83 |
-
def
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 84 |
output_wave = tango.generate(prompt, steps, guidance)
|
| 85 |
return gr.make_waveform((16000, output_wave))
|
| 86 |
|
|
@@ -106,14 +118,19 @@ with gr.Blocks() as interface:
|
|
| 106 |
"""
|
| 107 |
)
|
| 108 |
input_text = gr.Textbox(label = "Prompt", value = "Snort of a horse", lines = 2, autofocus = True)
|
| 109 |
-
|
| 110 |
-
|
|
|
|
| 111 |
|
| 112 |
submit = gr.Button("Generate 🚀", variant = "primary")
|
| 113 |
|
| 114 |
output_audio = gr.Audio(label = "Generated Audio")
|
| 115 |
|
| 116 |
-
submit.click(fn =
|
|
|
|
|
|
|
|
|
|
|
|
|
| 117 |
input_text,
|
| 118 |
denoising_steps,
|
| 119 |
guidance_scale
|
|
@@ -122,7 +139,7 @@ with gr.Blocks() as interface:
|
|
| 122 |
], scroll_to_output = True)
|
| 123 |
|
| 124 |
gr.Examples(
|
| 125 |
-
fn =
|
| 126 |
inputs = [
|
| 127 |
input_text,
|
| 128 |
denoising_steps,
|
|
|
|
| 47 |
self.scheduler = DDPMScheduler.from_pretrained(main_config["scheduler_name"], subfolder = "scheduler")
|
| 48 |
|
| 49 |
def chunks(self, lst, n):
|
| 50 |
+
# Yield successive n-sized chunks from a list
|
| 51 |
for i in range(0, len(lst), n):
|
| 52 |
yield lst[i:i + n]
|
| 53 |
|
| 54 |
def generate(self, prompt, steps = 100, guidance = 3, samples = 1, disable_progress = True):
|
| 55 |
+
# Generate audio for a single prompt string
|
| 56 |
with torch.no_grad():
|
| 57 |
latents = self.model.inference([prompt], self.scheduler, steps, guidance, samples, disable_progress = disable_progress)
|
| 58 |
mel = self.vae.decode_first_stage(latents)
|
|
|
|
| 60 |
return wave[0]
|
| 61 |
|
| 62 |
def generate_for_batch(self, prompts, steps = 200, guidance = 3, samples = 1, batch_size = 8, disable_progress = True):
|
| 63 |
+
# Generate audio for a list of prompt strings
|
| 64 |
outputs = []
|
| 65 |
for k in tqdm(range(0, len(prompts), batch_size)):
|
| 66 |
batch = prompts[k: k + batch_size]
|
|
|
|
| 80 |
tango.stft.to(device_type)
|
| 81 |
tango.model.to(device_type)
|
| 82 |
|
| 83 |
+
def check(
|
| 84 |
+
prompt,
|
| 85 |
+
steps,
|
| 86 |
+
guidance
|
| 87 |
+
):
|
| 88 |
+
if prompt is None or prompt == "":
|
| 89 |
+
raise gr.Error("Please provide a prompt input.")
|
| 90 |
+
|
| 91 |
+
def text2audio(
|
| 92 |
+
prompt,
|
| 93 |
+
steps,
|
| 94 |
+
guidance
|
| 95 |
+
):
|
| 96 |
output_wave = tango.generate(prompt, steps, guidance)
|
| 97 |
return gr.make_waveform((16000, output_wave))
|
| 98 |
|
|
|
|
| 118 |
"""
|
| 119 |
)
|
| 120 |
input_text = gr.Textbox(label = "Prompt", value = "Snort of a horse", lines = 2, autofocus = True)
|
| 121 |
+
with gr.Accordion("Advanced options", open = False):
|
| 122 |
+
denoising_steps = gr.Slider(label = "Steps", info = "lower=faster & variant, higher=audio quality & similar", minimum = 100, maximum = 200, value = 100, step = 1, interactive = True)
|
| 123 |
+
guidance_scale = gr.Slider(label = "Guidance Scale", info = "lower=audio quality, higher=follow the prompt", minimum = 1, maximum = 10, value = 3, step = 0.1, interactive = True)
|
| 124 |
|
| 125 |
submit = gr.Button("Generate 🚀", variant = "primary")
|
| 126 |
|
| 127 |
output_audio = gr.Audio(label = "Generated Audio")
|
| 128 |
|
| 129 |
+
submit.click(fn = check, inputs = [
|
| 130 |
+
input_text,
|
| 131 |
+
denoising_steps,
|
| 132 |
+
guidance_scale
|
| 133 |
+
], outputs = [], queue = False, show_progress = False).success(fn = text2audio, inputs = [
|
| 134 |
input_text,
|
| 135 |
denoising_steps,
|
| 136 |
guidance_scale
|
|
|
|
| 139 |
], scroll_to_output = True)
|
| 140 |
|
| 141 |
gr.Examples(
|
| 142 |
+
fn = text2audio,
|
| 143 |
inputs = [
|
| 144 |
input_text,
|
| 145 |
denoising_steps,
|