Option to set a custom random seed

.gitignore

@@ -13,4 +13,8 @@ __pycache__/
 /docs/harald_24000.wav
 /docs/harald.wav
 /models
-/bark/assets/prompts/MeMyselfAndI.npz
+.venv/*
+build/*
+Outputs/*
+suno_bark.egg-info/*
+/bark/assets/prompts/MeMyselfAndI.npz

bark/__init__.py

@@ -1,2 +1,2 @@
-from .api import generate_audio, text_to_semantic, semantic_to_waveform, save_as_prompt
+from .api import generate_audio, text_to_semantic, semantic_to_waveform, save_as_prompt, set_seed
 from .generation import SAMPLE_RATE, preload_models

bark/api.py

@@ -2,6 +2,10 @@
 
 import numpy as np
 
+import torch
+import random
+import os
+
 from .generation import codec_decode, generate_coarse, generate_fine, generate_text_semantic
 
 
@@ -123,3 +127,44 @@ def generate_audio(
     else:
         audio_arr = out
     return audio_arr
+
+def set_seed(seed: int = 0):
+    """Set the seed
+
+    seed = 0         Generate a random seed
+    seed = -1        Disable deterministic algorithms
+    0 < seed < 2**32 Set the seed
+
+    Args:
+        seed: integer to use as seed
+
+    Returns:
+        integer used as seed
+    """
+
+    original_seed = seed
+
+    # See for more informations: https://pytorch.org/docs/stable/notes/randomness.html
+    if seed == -1:
+        # Disable deterministic
+        torch.backends.cudnn.deterministic = False
+        torch.backends.cudnn.benchmark = True
+    else:
+        # Enable deterministic
+        torch.backends.cudnn.deterministic = True
+        torch.backends.cudnn.benchmark = False
+
+    if seed <= 0:
+        # Generate random seed
+        # Use default_rng() because it is independent of np.random.seed()
+        seed = np.random.default_rng().integers(1, 2**32 - 1)
+
+    assert(0 < seed and seed < 2**32)
+
+    np.random.seed(seed)
+    random.seed(seed)
+    torch.manual_seed(seed)
+    torch.cuda.manual_seed_all(seed)
+    os.environ["PYTHONHASHSEED"] = str(seed)
+
+    return original_seed if original_seed != 0 else seed

webui.py

@@ -10,7 +10,7 @@
 from xml.sax import saxutils
 #import nltk
 
-from bark import SAMPLE_RATE, generate_audio
+from bark import SAMPLE_RATE, generate_audio, set_seed
 from bark.clonevoice import clone_voice
 from bark.generation import SAMPLE_RATE, preload_models, codec_decode, generate_coarse, generate_fine, generate_text_semantic
 from scipy.io.wavfile import write as write_wav
@@ -54,7 +54,7 @@ def generate_with_settings(text_prompt, semantic_temp=0.7, semantic_top_k=50, se
         return full_generation, codec_decode(x_fine_gen)
     return codec_decode(x_fine_gen)
 
-def generate_text_to_speech(text, selected_speaker, text_temp, waveform_temp, quick_generation, complete_settings, progress=gr.Progress(track_tqdm=True)):
+def generate_text_to_speech(text, selected_speaker, text_temp, waveform_temp, quick_generation, complete_settings, random_seed_number, progress=gr.Progress(track_tqdm=True)):
     if text == None or len(text) < 1:
         raise gr.Error('No text entered!')
 
@@ -80,6 +80,8 @@ def generate_text_to_speech(text, selected_speaker, text_temp, waveform_temp, qu
     use_coarse_history_prompt = "Use coarse history" in complete_settings
     use_fine_history_prompt = "Use fine history" in complete_settings
     use_last_generation_as_history = "Use last generation as history" in complete_settings
+    random_seed = int(random_seed_number)
+
     progress(0, desc="Generating")
 
     silenceshort = np.zeros(int(0.25 * SAMPLE_RATE), dtype=np.float32)  # quarter second of silence
@@ -91,6 +93,8 @@ def generate_text_to_speech(text, selected_speaker, text_temp, waveform_temp, qu
         list_speak = create_clips_from_ssml(text)
         prev_speaker = None
         for i, clip in tqdm(enumerate(list_speak), total=len(list_speak)):
+            # set seed for consistent generation
+            set_seed(random_seed)
             selected_speaker = clip[0]
             # Add pause break between speakers
             if i > 0 and selected_speaker != prev_speaker:
@@ -109,6 +113,8 @@ def generate_text_to_speech(text, selected_speaker, text_temp, waveform_temp, qu
     else:
         texts = split_and_recombine_text(text)
         for i, text in tqdm(enumerate(texts), total=len(texts)):
+            # set seed for consistent generation
+            set_seed(random_seed)
             print(f"\nGenerating Text ({i+1}/{len(texts)}) -> {selected_speaker}:`{text}`")
             if quick_generation == True:
                 audio_array = generate_audio(text, selected_speaker, text_temp, waveform_temp)
@@ -150,6 +156,10 @@ def generate_text_to_speech(text, selected_speaker, text_temp, waveform_temp, qu
             if text[-1] in "!?.\n" and i > 1:
                 all_parts += [silenceshort.copy()]
 
+            #all_parts += [audio_array, silencelong.copy()]
+    # reset seed
+    set_seed(-1)
+
     # save & play audio
     result = create_filename(OUTPUTFOLDER, "final",".wav")
     save_wav(np.concatenate(all_parts), result)
@@ -298,9 +308,12 @@ def convert_text_to_ssml(text, selected_speaker):
             with gr.Column():
                 quick_gen_checkbox = gr.Checkbox(label="Quick Generation", value=True)
             with gr.Column():
-                settings_checkboxes = ["Use semantic history", "Use coarse history", "Use fine history", "Use last generation as history"]
-                complete_settings = gr.CheckboxGroup(choices=settings_checkboxes, value=settings_checkboxes, label="Detailed Generation Settings", type="value", interactive=True, visible=False)
-                quick_gen_checkbox.change(fn=on_quick_gen_changed, inputs=quick_gen_checkbox, outputs=complete_settings)
+                with gr.Row():
+                    settings_checkboxes = ["Use semantic history", "Use coarse history", "Use fine history", "Use last generation as history"]
+                    complete_settings = gr.CheckboxGroup(choices=settings_checkboxes, value=settings_checkboxes, label="Detailed Generation Settings", type="value", interactive=True, visible=False)
+                    random_seed_number = gr.inputs.Number(label="Random Seed", default=-1)
+                    #random_seed_settings = gr.NumberGroup([random_seed_number], label="Random Seed Settings", interactive=True, visible=False)
+                    quick_gen_checkbox.change(fn=on_quick_gen_changed, inputs=quick_gen_checkbox, outputs=[complete_settings])
 
         with gr.Row():
             with gr.Column():
@@ -321,7 +334,7 @@ def convert_text_to_ssml(text, selected_speaker):
         dummy = gr.Text(label="Progress")
 
         convert_to_ssml_button.click(convert_text_to_ssml, inputs=[input_text, speaker],outputs=input_text)
-        tts_create_button.click(generate_text_to_speech, inputs=[input_text, speaker, text_temp, waveform_temp, quick_gen_checkbox, complete_settings],outputs=output_audio)
+        tts_create_button.click(generate_text_to_speech, inputs=[input_text, speaker, text_temp, waveform_temp, quick_gen_checkbox, complete_settings, random_seed_number],outputs=output_audio)
         # Javascript hack to display modal confirmation dialog
         js = "(x) => confirm('Are you sure? This will remove all files from output folder')"
         button_delete_files.click(None, None, hidden_checkbox, _js=js)