Handle seed

app.py

@@ -2,7 +2,7 @@ import gradio as gr
 import json
 import torch
 import time
-import wavio
+import random
 
 from tqdm import tqdm
 from huggingface_hub import snapshot_download
@@ -11,6 +11,8 @@ from audioldm.audio.stft import TacotronSTFT
 from audioldm.variational_autoencoder import AutoencoderKL
 from pydub import AudioSegment
 
+max_64_bit_int = 2**63 - 1
+
 # Automatic device detection
 if torch.cuda.is_available():
     device_type = "cuda"
@@ -82,17 +84,21 @@ tango.vae.to(device_type)
 tango.stft.to(device_type)
 tango.model.to(device_type)
 
+def update_seed(is_randomize_seed, seed):
+    if is_randomize_seed:
+        return random.randint(0, max_64_bit_int)
+    return seed
+
 def check(
     prompt,
-    output_format,
     output_number,
     steps,
-    guidance
+    guidance,
+    is_randomize_seed,
+    seed
 ):
     if prompt is None or prompt == "":
         raise gr.Error("Please provide a prompt input.")
-    if not output_format in ["wav", "mp3"]:
-        raise gr.Error("Please choose an allowed output format (.wav or .mp3).")
     if not output_number in [1, 2, 3]:
         raise gr.Error("Please ask for 1, 2 or 3 output files.")
 
@@ -100,45 +106,31 @@ def update_output(output_format, output_number):
     return [
         gr.update(format = output_format),
         gr.update(format = output_format, visible = (2 <= output_number)),
-        gr.update(format = output_format, visible = (output_number == 3))
+        gr.update(format = output_format, visible = (output_number == 3)),
+        gr.update(visible = False)
     ]
 
 def text2audio(
     prompt,
-    output_format,
     output_number,
     steps,
-    guidance
+    guidance,
+    is_randomize_seed,
+    seed
 ):
     start = time.time()
-    output_wave = tango.generate(prompt, steps, guidance, output_number)
 
-    output_filename_1 = "tmp1.wav"
-    wavio.write(output_filename_1, output_wave[0], rate = 16000, sampwidth = 2)
+    if seed is None:
+        seed = random.randint(0, max_64_bit_int)
 
-    if (output_format == "mp3"):
-        AudioSegment.from_wav("tmp1.wav").export("tmp1.mp3", format = "mp3")
-        output_filename_1 = "tmp1.mp3"
+    random.seed(seed)
+    torch.manual_seed(seed)
 
-    if (2 <= output_number):
-        output_filename_2 = "tmp2.wav"
-        wavio.write(output_filename_2, output_wave[1], rate = 16000, sampwidth = 2)
-
-        if (output_format == "mp3"):
-            AudioSegment.from_wav("tmp2.wav").export("tmp2.mp3", format = "mp3")
-            output_filename_2 = "tmp2.mp3"
-    else:
-        output_filename_2 = None
-
-    if (output_number == 3):
-        output_filename_3 = "tmp3.wav"
-        wavio.write(output_filename_3, output_wave[2], rate = 16000, sampwidth = 2)
+    output_wave = tango.generate(prompt, steps, guidance, output_number)
 
-        if (output_format == "mp3"):
-            AudioSegment.from_wav("tmp3.wav").export("tmp3.mp3", format = "mp3")
-            output_filename_3 = "tmp3.mp3"
-    else:
-        output_filename_3 = None
+    output_wave_1 = gr.make_waveform((16000, output_wave[0]))
+    output_wave_2 = gr.make_waveform((16000, output_wave[1])) if (2 <= output_number) else None
+    output_wave_3 = gr.make_waveform((16000, output_wave[2])) if (output_number == 3) else None
 
     end = time.time()
     secondes = int(end - start)
@@ -147,10 +139,10 @@ def text2audio(
     hours = minutes // 60
     minutes = minutes - (hours * 60)
     return [
-        output_filename_1,
-        output_filename_2,
-        output_filename_3,
-        "Start again to get a different result. The output have been generated in " + ((str(hours) + " h, ") if hours != 0 else "") + ((str(minutes) + " min, ") if hours != 0 or minutes != 0 else "") + str(secondes) + " sec."
+        output_wave_1,
+        output_wave_2,
+        output_wave_3,
+        gr.update(visible = True, value = "Start again to get a different result. The output have been generated in " + ((str(hours) + " h, ") if hours != 0 else "") + ((str(minutes) + " min, ") if hours != 0 or minutes != 0 else "") + str(secondes) + " sec.")
     ]
 
 # Gradio interface
@@ -168,45 +160,55 @@ with gr.Blocks() as interface:
         <li>If you need to generate <b>music</b>, I recommend to use <i>MusicGen</i>,</li>
         </ul>
         <br/>
-        🐌 Slow process... ~2 hours. Your computer must <b><u>not</u></b> enter into standby mode.<br/>You can duplicate this space on a free account, it works on CPU.<br/>
-        <a href='https://huggingface.co/spaces/Fabrice-TIERCELIN/Text-to-Audio?duplicate=true'><img src='https://img.shields.io/badge/-Duplicate%20Space-blue?labelColor=white&style=flat&logo=data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAABAAAAAQCAYAAAAf8/9hAAAAAXNSR0IArs4c6QAAAP5JREFUOE+lk7FqAkEURY+ltunEgFXS2sZGIbXfEPdLlnxJyDdYB62sbbUKpLbVNhyYFzbrrA74YJlh9r079973psed0cvUD4A+4HoCjsA85X0Dfn/RBLBgBDxnQPfAEJgBY+A9gALA4tcbamSzS4xq4FOQAJgCDwV2CPKV8tZAJcAjMMkUe1vX+U+SMhfAJEHasQIWmXNN3abzDwHUrgcRGmYcgKe0bxrblHEB4E/pndMazNpSZGcsZdBlYJcEL9Afo75molJyM2FxmPgmgPqlWNLGfwZGG6UiyEvLzHYDmoPkDDiNm9JR9uboiONcBXrpY1qmgs21x1QwyZcpvxt9NS09PlsPAAAAAElFTkSuQmCC&logoWidth=14'></a>
+        🐌 Slow process... ~5 min. Your computer must <b><u>not</u></b> enter into standby mode.<br/>You can duplicate this space on a free account, it works on CPU.<br/>
+        <a href='https://huggingface.co/spaces/Fabrice-TIERCELIN/Text-to-Audio?duplicate=true&hidden=public&hidden=public'><img src='https://img.shields.io/badge/-Duplicate%20Space-blue?labelColor=white&style=flat&logo=data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAABAAAAAQCAYAAAAf8/9hAAAAAXNSR0IArs4c6QAAAP5JREFUOE+lk7FqAkEURY+ltunEgFXS2sZGIbXfEPdLlnxJyDdYB62sbbUKpLbVNhyYFzbrrA74YJlh9r079973psed0cvUD4A+4HoCjsA85X0Dfn/RBLBgBDxnQPfAEJgBY+A9gALA4tcbamSzS4xq4FOQAJgCDwV2CPKV8tZAJcAjMMkUe1vX+U+SMhfAJEHasQIWmXNN3abzDwHUrgcRGmYcgKe0bxrblHEB4E/pndMazNpSZGcsZdBlYJcEL9Afo75molJyM2FxmPgmgPqlWNLGfwZGG6UiyEvLzHYDmoPkDDiNm9JR9uboiONcBXrpY1qmgs21x1QwyZcpvxt9NS09PlsPAAAAAElFTkSuQmCC&logoWidth=14'></a>
         <br/>
         ⚖️ You can use, modify and share the generated sounds but not for commercial uses.
         """
     )
     input_text = gr.Textbox(label = "Prompt", value = "Snort of a horse", lines = 2, autofocus = True)
-    output_format = gr.Radio(label = "Output format", info = "The file you can dowload", choices = ["mp3", "wav"], value = "wav")
     with gr.Accordion("Advanced options", open = False):
+        output_format = gr.Radio(label = "Output format", info = "The file you can dowload", choices = ["mp3", "wav"], value = "wav")
         output_number = gr.Slider(label = "Number of generations", info = "1, 2 or 3 output files", minimum = 1, maximum = 3, value = 3, step = 1, interactive = True)
-        denoising_steps = gr.Slider(label = "Steps", info = "lower=faster & variant, higher=audio quality & similar", minimum = 100, maximum = 200, value = 100, step = 1, interactive = True)
+        denoising_steps = gr.Slider(label = "Steps", info = "lower=faster & variant, higher=audio quality & similar", minimum = 10, maximum = 200, value = 10, step = 1, interactive = True)
         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)
+        randomize_seed = gr.Checkbox(label = "\U0001F3B2 Randomize seed", value = True, info = "If checked, result is always different")
+        seed = gr.Slider(minimum = 0, maximum = max_64_bit_int, step = 1, randomize = True, label = "Seed")
 
     submit = gr.Button("🚀 Generate", variant = "primary")
 
-    output_audio_1 = gr.Audio(label = "Generated Audio #1/3", format = "wav", type="filepath", autoplay = True)
-    output_audio_2 = gr.Audio(label = "Generated Audio #2/3", format = "wav", type="filepath")
-    output_audio_3 = gr.Audio(label = "Generated Audio #3/3", format = "wav", type="filepath")
+    output_audio_1 = gr.Audio(label = "Generated Audio #1/3", format = "wav", type="numpy", autoplay = True)
+    output_audio_2 = gr.Audio(label = "Generated Audio #2/3", format = "wav", type="numpy")
+    output_audio_3 = gr.Audio(label = "Generated Audio #3/3", format = "wav", type="numpy")
     information = gr.Label(label = "Information")
 
-    submit.click(fn = check, inputs = [
+    submit.click(fn = update_seed, inputs = [
+        randomize_seed,
+        seed
+    ], outputs = [
+        seed
+    ], queue = False, show_progress = False).then(fn = check, inputs = [
         input_text,
-        output_format,
         output_number,
         denoising_steps,
-        guidance_scale
+        guidance_scale,
+        randomize_seed,
+        seed
     ], outputs = [], queue = False, show_progress = False).success(fn = update_output, inputs = [
         output_format,
         output_number
     ], outputs = [
         output_audio_1,
         output_audio_2,
-        output_audio_3
+        output_audio_3,
+        information
     ], queue = False, show_progress = False).success(fn = text2audio, inputs = [
         input_text,
-        output_format,
         output_number,
         denoising_steps,
-        guidance_scale
+        guidance_scale,
+        randomize_seed,
+        seed
     ], outputs = [
         output_audio_1,
         output_audio_2,
@@ -218,10 +220,11 @@ with gr.Blocks() as interface:
         fn = text2audio,
 	    inputs = [
             input_text,
-            output_format,
             output_number,
             denoising_steps,
-            guidance_scale
+            guidance_scale,
+            randomize_seed,
+            seed
         ],
 	    outputs = [
             output_audio_1,
@@ -230,11 +233,11 @@ with gr.Blocks() as interface:
             information
         ],
         examples = [
-                ["A hammer is hitting a wooden surface", "mp3", 3, 100, 3],
-                ["Peaceful and calming ambient music with singing bowl and other instruments.", "wav", 3, 100, 3],
-                ["A man is speaking in a small room.", "mp3", 2, 100, 3],
-                ["A female is speaking followed by footstep sound", "mp3", 1, 100, 3],
-                ["Wooden table tapping sound followed by water pouring sound.", "mp3", 3, 200, 3],
+                ["A hammer is hitting a wooden surface", 3, 100, 3, False, 123],
+                ["Peaceful and calming ambient music with singing bowl and other instruments.", 3, 100, 3, False, 123],
+                ["A man is speaking in a small room.", 2, 100, 3, False, 123],
+                ["A female is speaking followed by footstep sound", 1, 100, 3, False, 123],
+                ["Wooden table tapping sound followed by water pouring sound.", 3, 200, 3, False, 123],
             ],
         cache_examples = "lazy",
     )