add audiogen

README.md

@@ -1,4 +1,4 @@
-# TTS Generation WebUI (Bark, MusicGen, Tortoise, RVC, Vocos, Demucs)
+# TTS Generation WebUI (Bark, MusicGen + AudioGen, Tortoise, RVC, Vocos, Demucs)
 
 ## One click installers
 
@@ -34,6 +34,9 @@ Google Colab demo: [![Open In Colab](https://colab.research.google.com/assets/co
 https://rsxdalv.github.io/bark-speaker-directory/
 
 ## Changelog
+Aug 3:
+* Add AudioGen https://github.com/rsxdalv/tts-generation-webui/pull/105
+
 Aug 2:
 * Fix Model locations not showing after restart
 
@@ -260,3 +263,4 @@ Most notably:
 - Bark: CC BY-NC 4.0 (MIT According to repo, but CC BY-NC 4.0 according to HuggingFace)
 - Tortoise: *Unknown* (Apache-2.0 according to repo, but no license file in HuggingFace)
 - MusicGen: CC BY-NC 4.0
+- AudioGen: CC BY-NC 4.0

requirements_audiocraft.txt

@@ -1,2 +1,2 @@
 # audiocraft==0.0.1 # MIT License
-audiocraft @ git+https://[email protected]/facebookresearch/audiocraft@d874966#egg=audiocraft # MIT License
+audiocraft @ git+https://[email protected]/facebookresearch/audiocraft@c5157b5bf14bf83449c17ea1eeb66c19fb4bc7f0#egg=audiocraft # MIT License

src/musicgen/musicgen_tab.py

@@ -1,6 +1,7 @@
 import torch
 import gradio as gr
 from audiocraft.models.musicgen import MusicGen
+from audiocraft.models.audiogen import AudioGen
 from typing import Optional, Tuple, TypedDict
 import numpy as np
 import os
@@ -112,6 +113,8 @@ def save_generation(
 
 
 def load_model(version):
+    if version == "facebook/audiogen-medium":
+        return AudioGen.get_pretrained(version)
     print("Loading model", version)
     return MusicGen.get_pretrained(version)
 
@@ -197,9 +200,8 @@ def generate(params: MusicGenGeneration, melody_in: Optional[Tuple[int, np.ndarr
     ]
 
 
-
 def generation_tab_musicgen():
-    with gr.Tab("MusicGen") as tab:
+    with gr.Tab("MusicGen + AudioGen") as tab:
         musicgen_atom = gr.JSON(
             # visible=True,
             visible=False,
@@ -223,7 +225,7 @@ def generation_tab_musicgen():
                     label="Prompt", lines=3, placeholder="Enter text here..."
                 )
                 model = gr.Radio(
-                    ["melody", "medium", "small", "large"],
+                    ["melody", "medium", "small", "large", "facebook/audiogen-medium"],
                     label="Model",
                     value="melody",
                 )