adapt the version number to the release

app.py

@@ -2,7 +2,7 @@ import os
 
 import spaces
 
-os.system("git clone --branch v2.5 https://github.com/DigitalPhonetics/IMS-Toucan.git toucan_codebase")
+os.system("git clone --branch v3.0 https://github.com/DigitalPhonetics/IMS-Toucan.git toucan_codebase")
 os.system("mv toucan_codebase/* .")
 
 from run_model_downloader import download_models
@@ -39,7 +39,6 @@ class ControllableInterface(torch.nn.Module):
              language,
              accent,
              voice_seed,
-             prosody_creativity,
              duration_scaling_factor,
              pause_duration_scaling_factor,
              pitch_variance_scale,
@@ -114,7 +113,6 @@ class ControllableInterface(torch.nn.Module):
                                   energy_variance_scale=energy_variance_scale,
                                   pause_duration_scaling_factor=pause_duration_scaling_factor,
                                   return_plot_as_filepath=True,
-                                  prosody_creativity=prosody_creativity,
                                   loudness_in_db=loudness_in_db)
         return sr, wav, fig
 
@@ -132,7 +130,6 @@ controllable_ui = ControllableInterface(available_artificial_voices=available_ar
 def read(prompt,
          language,
          voice_seed,
-         prosody_creativity,
          duration_scaling_factor,
          pitch_variance_scale,
          energy_variance_scale,
@@ -147,7 +144,6 @@ def read(prompt,
                                             language.split(" ")[-1].split("(")[1].split(")")[0],
                                             language.split(" ")[-1].split("(")[1].split(")")[0],
                                             voice_seed,
-                                            prosody_creativity,
                                             duration_scaling_factor,
                                             1.,
                                             pitch_variance_scale,
@@ -177,7 +173,6 @@ iface = gr.Interface(fn=read,
                              gr.Slider(minimum=0, maximum=available_artificial_voices, step=1,
                                        value=279,
                                        label="Random Seed for the artificial Voice"),
-                             gr.Slider(minimum=0.0, maximum=0.8, step=0.1, value=0.7, label="Prosody Creativity"),
                              gr.Slider(minimum=0.7, maximum=1.3, step=0.1, value=1.0, label="Duration Scale"),
                              gr.Slider(minimum=0.5, maximum=1.5, step=0.1, value=1.0, label="Pitch Variance Scale"),
                              gr.Slider(minimum=0.5, maximum=1.5, step=0.1, value=1.0, label="Energy Variance Scale"),

app_future.py

@@ -0,0 +1,193 @@
+import os
+
+import spaces
+
+os.system("git clone --branch v3.1 https://github.com/DigitalPhonetics/IMS-Toucan.git toucan_codebase")
+os.system("mv toucan_codebase/* .")
+
+from run_model_downloader import download_models
+
+download_models()
+
+import gradio as gr
+import torch.cuda
+from Preprocessing.multilinguality.SimilaritySolver import load_json_from_path
+from Utility.utils import float2pcm
+
+import os
+
+import torch
+
+from Architectures.ControllabilityGAN.GAN import GanWrapper
+from InferenceInterfaces.ToucanTTSInterface import ToucanTTSInterface
+from Utility.storage_config import MODELS_DIR
+
+
+class ControllableInterface(torch.nn.Module):
+
+    def __init__(self, available_artificial_voices=1000):
+        super().__init__()
+        self.model = ToucanTTSInterface(device="cpu", tts_model_path="Meta")
+        self.wgan = GanWrapper(os.path.join(MODELS_DIR, "Embedding", "embedding_gan.pt"), device="cpu")
+        self.generated_speaker_embeds = list()
+        self.available_artificial_voices = available_artificial_voices
+        self.current_language = ""
+        self.current_accent = ""
+
+    def read(self,
+             prompt,
+             language,
+             accent,
+             voice_seed,
+             prosody_creativity,
+             duration_scaling_factor,
+             pause_duration_scaling_factor,
+             pitch_variance_scale,
+             energy_variance_scale,
+             emb_slider_1,
+             emb_slider_2,
+             emb_slider_3,
+             emb_slider_4,
+             emb_slider_5,
+             emb_slider_6,
+             loudness_in_db
+             ):
+        if self.current_language != language:
+            self.model.set_phonemizer_language(language)
+            self.current_language = language
+        if self.current_accent != accent:
+            self.model.set_accent_language(accent)
+            self.current_accent = accent
+
+        self.wgan.set_latent(voice_seed)
+        controllability_vector = torch.tensor([emb_slider_1,
+                                               emb_slider_2,
+                                               emb_slider_3,
+                                               emb_slider_4,
+                                               emb_slider_5,
+                                               emb_slider_6], dtype=torch.float32)
+        embedding = self.wgan.modify_embed(controllability_vector)
+        self.model.set_utterance_embedding(embedding=embedding)
+
+        phones = self.model.text2phone.get_phone_string(prompt)
+        if len(phones) > 1800:
+            if language == "deu":
+                prompt = "Deine Eingabe war zu lang. Bitte versuche es entweder mit einem kürzeren Text oder teile ihn in mehrere Teile auf."
+            elif language == "ell":
+                prompt = "Η εισήγησή σας ήταν πολύ μεγάλη. Παρακαλώ δοκιμάστε είτε ένα μικρότερο κείμενο είτε χωρίστε το σε διάφορα μέρη."
+            elif language == "spa":
+                prompt = "Su entrada es demasiado larga. Por favor, intente un texto más corto o divídalo en varias partes."
+            elif language == "fin":
+                prompt = "Vastauksesi oli liian pitkä. Kokeile joko lyhyempää tekstiä tai jaa se useampaan osaan."
+            elif language == "rus":
+                prompt = "Ваш текст слишком длинный. Пожалуйста, попробуйте либо сократить текст, либо разделить его на несколько частей."
+            elif language == "hun":
+                prompt = "Túl hosszú volt a bevitele. Kérjük, próbáljon meg rövidebb szöveget írni, vagy ossza több részre."
+            elif language == "nld":
+                prompt = "Uw input was te lang. Probeer een kortere tekst of splits het in verschillende delen."
+            elif language == "fra":
+                prompt = "Votre saisie était trop longue. Veuillez essayer un texte plus court ou le diviser en plusieurs parties."
+            elif language == 'pol':
+                prompt = "Twój wpis był zbyt długi. Spróbuj skrócić tekst lub podzielić go na kilka części."
+            elif language == 'por':
+                prompt = "O seu contributo foi demasiado longo. Por favor, tente um texto mais curto ou divida-o em várias partes."
+            elif language == 'ita':
+                prompt = "Il tuo input era troppo lungo. Per favore, prova un testo più corto o dividilo in più parti."
+            elif language == 'cmn':
+                prompt = "你的输入太长了。请尝试使用较短的文本或将其拆分为多个部分。"
+            elif language == 'vie':
+                prompt = "Đầu vào của bạn quá dài. Vui lòng thử một văn bản ngắn hơn hoặc chia nó thành nhiều phần."
+            else:
+                prompt = "Your input was too long. Please try either a shorter text or split it into several parts."
+                if self.current_language != "eng":
+                    self.model.set_phonemizer_language("eng")
+                    self.current_language = "eng"
+                if self.current_accent != "eng":
+                    self.model.set_accent_language("eng")
+                    self.current_accent = "eng"
+
+        print(prompt)
+        wav, sr, fig = self.model(prompt,
+                                  input_is_phones=False,
+                                  duration_scaling_factor=duration_scaling_factor,
+                                  pitch_variance_scale=pitch_variance_scale,
+                                  energy_variance_scale=energy_variance_scale,
+                                  pause_duration_scaling_factor=pause_duration_scaling_factor,
+                                  return_plot_as_filepath=True,
+                                  prosody_creativity=prosody_creativity,
+                                  loudness_in_db=loudness_in_db)
+        return sr, wav, fig
+
+
+title = "Controllable Text-to-Speech for over 7000 Languages"
+article = "Check out the IMS Toucan TTS Toolkit at https://github.com/DigitalPhonetics/IMS-Toucan"
+available_artificial_voices = 1000
+path_to_iso_list = "Preprocessing/multilinguality/iso_to_fullname.json"
+iso_to_name = load_json_from_path(path_to_iso_list)
+text_selection = [f"{iso_to_name[iso_code]} Text ({iso_code})" for iso_code in iso_to_name]
+controllable_ui = ControllableInterface(available_artificial_voices=available_artificial_voices)
+
+
+@spaces.GPU
+def read(prompt,
+         language,
+         voice_seed,
+         prosody_creativity,
+         duration_scaling_factor,
+         pitch_variance_scale,
+         energy_variance_scale,
+         emb1,
+         emb2
+         ):
+    if torch.cuda.is_available():
+        controllable_ui.to("cuda")
+        controllable_ui.device = "cuda"
+    try:
+        sr, wav, fig = controllable_ui.read(prompt,
+                                            language.split(" ")[-1].split("(")[1].split(")")[0],
+                                            language.split(" ")[-1].split("(")[1].split(")")[0],
+                                            voice_seed,
+                                            prosody_creativity,
+                                            duration_scaling_factor,
+                                            1.,
+                                            pitch_variance_scale,
+                                            energy_variance_scale,
+                                            emb1,
+                                            emb2,
+                                            0.,
+                                            0.,
+                                            0.,
+                                            0.,
+                                            -24.)
+    finally:
+        controllable_ui.to("cpu")
+        controllable_ui.device = "cpu"
+    return (sr, float2pcm(wav)), fig
+
+
+iface = gr.Interface(fn=read,
+                     inputs=[gr.Textbox(lines=2,
+                                        placeholder="write what you want the synthesis to read here...",
+                                        value="The woods are lovely, dark and deep, but I have promises to keep, and miles to go, before I sleep.",
+                                        label="Text input"),
+                             gr.Dropdown(text_selection,
+                                         type="value",
+                                         value='English Text (eng)',
+                                         label="Select the Language of the Text (type on your keyboard to find it quickly)"),
+                             gr.Slider(minimum=0, maximum=available_artificial_voices, step=1,
+                                       value=279,
+                                       label="Random Seed for the artificial Voice"),
+                             gr.Slider(minimum=0.0, maximum=0.8, step=0.1, value=0.7, label="Prosody Creativity"),
+                             gr.Slider(minimum=0.7, maximum=1.3, step=0.1, value=1.0, label="Duration Scale"),
+                             gr.Slider(minimum=0.5, maximum=1.5, step=0.1, value=1.0, label="Pitch Variance Scale"),
+                             gr.Slider(minimum=0.5, maximum=1.5, step=0.1, value=1.0, label="Energy Variance Scale"),
+                             gr.Slider(minimum=-10.0, maximum=10.0, step=0.1, value=0.0, label="Femininity / Masculinity"),
+                             gr.Slider(minimum=-10.0, maximum=10.0, step=0.1, value=0.0, label="Voice Depth")
+                             ],
+                     outputs=[gr.Audio(type="numpy", label="Speech"),
+                              gr.Image(label="Visualization")],
+                     title=title,
+                     theme="default",
+                     allow_flagging="never",
+                     article=article)
+iface.launch()