feat(webui): support external spk_emb (#467)

Author: fumiama

examples/web/funcs.py

@@ -21,6 +21,10 @@
 custom_path: Optional[str] = None
 
 has_interrupted = False
+is_in_generate = False
+
+seed_min = 1
+seed_max = 4294967295
 
 # 音色选项：用于预置合适的音色
 voices = {
@@ -38,13 +42,18 @@
 
 
 def generate_seed():
-    return gr.update(value=random.randint(1, 100000000))
+    return gr.update(value=random.randint(seed_min, seed_max))
 
 
 # 返回选择音色对应的seed
 def on_voice_change(vocie_selection):
     return voices.get(vocie_selection)["seed"]
 
+def on_audio_seed_change(audio_seed_input):
+    with TorchSeedContext(audio_seed_input):
+        rand_spk = chat.sample_random_speaker()
+    return rand_spk
+
 
 def load_chat(cust_path: Optional[str], coef: Optional[str]) -> bool:
     if cust_path == None:
@@ -79,6 +88,12 @@ def load_chat(cust_path: Optional[str], coef: Optional[str]) -> bool:
 
 
 def reload_chat(coef: Optional[str]) -> str:
+    global is_in_generate
+
+    if is_in_generate:
+        gr.Warning("Cannot reload when generating!")
+        return coef
+
     chat.unload()
     gr.Info("Model unloaded.")
     if len(coef) != 230:
@@ -119,37 +134,33 @@ def refine_text(
 
     return text[0] if isinstance(text, list) else text
 
-def generate_audio(text, temperature, top_P, top_K, audio_seed_input, stream):
+def generate_audio(text, temperature, top_P, top_K, spk_emb_text: str, stream):
     global chat, has_interrupted
 
-    if not text or has_interrupted:
+    if not text or has_interrupted or not spk_emb_text.startswith("蘁淰"):
         return None
 
-    with TorchSeedContext(audio_seed_input):
-        rand_spk = chat.sample_random_speaker()
-
     params_infer_code = ChatTTS.Chat.InferCodeParams(
-        spk_emb=rand_spk,
+        spk_emb=spk_emb_text,
         temperature=temperature,
         top_P=top_P,
         top_K=top_K,
     )
 
-    with TorchSeedContext(audio_seed_input):
-        wav = chat.infer(
-            text,
-            skip_refine_text=True,
-            params_infer_code=params_infer_code,
-            stream=stream,
-        )
-        if stream:
-            for gen in wav:
-                audio = gen[0]
-                if audio is not None and len(audio) > 0:
-                    yield wav_arr_to_mp3_view(audio[0]).tobytes()
-                    del audio
-        else:
-            yield wav_arr_to_mp3_view(np.array(wav[0]).flatten()).tobytes()
+    wav = chat.infer(
+        text,
+        skip_refine_text=True,
+        params_infer_code=params_infer_code,
+        stream=stream,
+    )
+    if stream:
+        for gen in wav:
+            audio = gen[0]
+            if audio is not None and len(audio) > 0:
+                yield wav_arr_to_mp3_view(audio[0]).tobytes()
+                del audio
+    else:
+        yield wav_arr_to_mp3_view(np.array(wav[0]).flatten()).tobytes()
 
 
 def interrupt_generate():
@@ -159,17 +170,20 @@ def interrupt_generate():
     chat.interrupt()
 
 def set_buttons_before_generate(generate_button, interrupt_button):
-    global has_interrupted
+    global has_interrupted, is_in_generate
 
     has_interrupted = False
+    is_in_generate = True
 
     return _set_generate_buttons(
         generate_button,
         interrupt_button,
     )
 
 def set_buttons_after_generate(generate_button, interrupt_button, audio_output):
-    global has_interrupted
+    global has_interrupted, is_in_generate
+
+    is_in_generate = False
 
     return _set_generate_buttons(
         generate_button,

examples/web/webui.py

@@ -55,17 +55,31 @@ def main():
             voice_selection = gr.Dropdown(
                 label="Timbre", choices=voices.keys(), value="Default"
             )
-            audio_seed_input = gr.Number(value=2, label="Audio Seed", interactive=True)
+            audio_seed_input = gr.Number(
+                value=2, label="Audio Seed", interactive=True,
+                minimum=seed_min, maximum=seed_max,
+            )
             generate_audio_seed = gr.Button("\U0001F3B2")
-            text_seed_input = gr.Number(value=42, label="Text Seed")
+            text_seed_input = gr.Number(
+                value=42, label="Text Seed", interactive=True,
+                minimum=seed_min, maximum=seed_max,
+            )
             generate_text_seed = gr.Button("\U0001F3B2")
 
         with gr.Row():
+            spk_emb_text = gr.Textbox(
+                label="Speaker Embedding",
+                max_lines=3,
+                show_copy_button=True,
+                interactive=True,
+                scale=2,
+            )
             dvae_coef_text = gr.Textbox(
                 label="DVAE Coefficient",
                 max_lines=3,
                 show_copy_button=True,
-                scale=4,
+                interactive=True,
+                scale=2,
             )
             reload_chat_button = gr.Button("Reload", scale=1)
 
@@ -88,9 +102,11 @@ def main():
             fn=on_voice_change, inputs=voice_selection, outputs=audio_seed_input
         )
 
-        generate_audio_seed.click(generate_seed, inputs=[], outputs=audio_seed_input)
+        generate_audio_seed.click(generate_seed, outputs=audio_seed_input)
+
+        generate_text_seed.click(generate_seed, outputs=text_seed_input)
 
-        generate_text_seed.click(generate_seed, inputs=[], outputs=text_seed_input)
+        audio_seed_input.change(on_audio_seed_change, inputs=audio_seed_input, outputs=spk_emb_text)
 
         reload_chat_button.click(
             reload_chat, inputs=dvae_coef_text, outputs=dvae_coef_text
@@ -123,7 +139,7 @@ def make_audio(autoplay, stream):
                     temperature_slider,
                     top_p_slider,
                     top_k_slider,
-                    audio_seed_input,
+                    spk_emb_text,
                     stream_mode_checkbox,
                 ],
                 outputs=audio_output,
@@ -168,6 +184,7 @@ def make_audio(autoplay, stream):
         logger.error("Models load failed.")
         sys.exit(1)
 
+    spk_emb_text.value = on_audio_seed_change(audio_seed_input.value)
     dvae_coef_text.value = chat.coef
 
     demo.launch(