feat(webui): impl. zero shot infer

ChatTTS/core.py

@@ -297,7 +297,7 @@ def _load(
             logger=self.logger,
         ).eval()
         assert gpt_ckpt_path, "gpt_ckpt_path should not be None"
-        gpt.load_state_dict(torch.load(gpt_ckpt_path, weights_only=True, mmap=True))
+        gpt.from_pretrained(gpt_ckpt_path)
         gpt.prepare(compile=compile and "cuda" in str(device))
         self.gpt = gpt
         spk_stat_path = os.path.join(os.path.dirname(gpt_ckpt_path), "spk_stat.pt")

examples/web/funcs.py

@@ -5,7 +5,7 @@
 
 import gradio as gr
 
-from tools.audio import float_to_int16, has_ffmpeg_installed
+from tools.audio import float_to_int16, has_ffmpeg_installed, load_audio
 from tools.logger import get_logger
 
 logger = get_logger(" WebUI ")
@@ -114,6 +114,15 @@ def reload_chat(coef: Optional[str]) -> str:
     return chat.coef
 
 
+def on_upload_sample_audio(sample_audio_input: Optional[str]) -> str:
+    if sample_audio_input is None:
+        return ""
+    sample_audio = load_audio(sample_audio_input, 24000)
+    spk_smp = chat.sample_audio_speaker(sample_audio)
+    del sample_audio
+    return spk_smp
+
+
 def _set_generate_buttons(generate_button, interrupt_button, is_reset=False):
     return gr.update(
         value=generate_button, visible=is_reset, interactive=is_reset
@@ -142,7 +151,7 @@ def refine_text(
 
 
 def generate_audio(
-    text, temperature, top_P, top_K, spk_emb_text: str, stream, audio_seed_input
+    text, temperature, top_P, top_K, spk_emb_text: str, stream, audio_seed_input, sample_text_input, sample_audio_code_input,
 ):
     global chat, has_interrupted
 
@@ -156,6 +165,11 @@ def generate_audio(
         top_K=top_K,
     )
 
+    if sample_text_input and sample_audio_code_input:
+        params_infer_code.txt_smp = sample_text_input
+        params_infer_code.spk_smp = sample_audio_code_input
+        params_infer_code.spk_emb = None
+
     with TorchSeedContext(audio_seed_input):
         wav = chat.infer(
             text,

examples/web/webui.py

@@ -21,13 +21,43 @@ def main():
         gr.Markdown("- **GitHub Repo**: https://github.com/2noise/ChatTTS")
         gr.Markdown("- **HuggingFace Repo**: https://huggingface.co/2Noise/ChatTTS")
 
-        text_input = gr.Textbox(
-            label="Input Text",
-            lines=4,
-            placeholder="Please Input Text...",
-            value=ex[0][0],
-            interactive=True,
-        )
+        with gr.Row():
+            with gr.Column(scale=2):
+                text_input = gr.Textbox(
+                    label="Input Text",
+                    lines=4,
+                    max_lines=4,
+                    placeholder="Please Input Text...",
+                    value=ex[0][0],
+                    interactive=True,
+                )
+                sample_text_input = gr.Textbox(
+                    label="Sample Text",
+                    lines=4,
+                    max_lines=4,
+                    placeholder="If Sample Audio and Sample Text are available, the Speaker Embedding will be disabled.",
+                    interactive=True,
+                )
+            with gr.Column():
+                with gr.Tab(label="Sample Audio"):
+                    sample_audio_input = gr.Audio(
+                        value=None,
+                        type="filepath",
+                        interactive=True,
+                        show_label=False,
+                        waveform_options=gr.WaveformOptions(
+                            sample_rate=24000,
+                        ),
+                        scale=1,
+                    )
+                with gr.Tab(label="Sample Audio Code"):
+                    sample_audio_code_input = gr.Textbox(
+                        lines=12,
+                        max_lines=12,
+                        show_label=False,
+                        placeholder="Paste the Code copied before after uploading Sample Audio.",
+                        interactive=True,
+                    )
 
         with gr.Row():
             refine_text_checkbox = gr.Checkbox(
@@ -126,6 +156,11 @@ def main():
             show_copy_button=True,
         )
 
+        sample_audio_input.change(
+            fn=on_upload_sample_audio,
+            inputs=sample_audio_input, outputs=sample_audio_code_input,
+        ).then(fn=lambda: gr.Info("Sampled Audio Code generated at another Tab."))
+
         # 使用Gradio的回调功能来更新数值输入框
         voice_selection.change(
             fn=on_voice_change, inputs=voice_selection, outputs=audio_seed_input
@@ -181,6 +216,8 @@ def make_audio(autoplay, stream):
                     spk_emb_text,
                     stream_mode_checkbox,
                     audio_seed_input,
+                    sample_text_input,
+                    sample_audio_code_input,
                 ],
                 outputs=audio_output,
             ).then(