Update app.py

app.py

@@ -20,36 +20,40 @@ language_marks = {
     "English": "[EN]",
     "Mix": "",
 }
-lang = ['日本語', '简体中文', 'English', 'Mix']
 
+limitation = os.getenv("SYSTEM") == "spaces"  # limit text and audio length in huggingface spaces
 
 def get_text(text, hps, is_symbol):
-    text_norm = text_to_sequence(
-        text, hps.symbols, [] if is_symbol else hps.data.text_cleaners)
+    text_norm = text_to_sequence(text, hps.symbols, [] if is_symbol else hps.data.text_cleaners)
     if hps.data.add_blank:
         text_norm = commons.intersperse(text_norm, 0)
     text_norm = LongTensor(text_norm)
     return text_norm
 
-
 def create_tts_fn(model, hps, speaker_ids):
-    def tts_fn(text, speaker, language, speed):
+    def tts_fn(text, speaker, language, ns, nsw, speed, is_symbol):
+        if limitation:
+            text_len = len(re.sub("\[([A-Z]{2})\]", "", text))
+            max_len = 150
+            if is_symbol:
+                max_len *= 3
+            if text_len > max_len:
+                return "Error: Text is too long", None
         if language is not None:
             text = language_marks[language] + text + language_marks[language]
         speaker_id = speaker_ids[speaker]
-        stn_tst = get_text(text, hps, False)
+        stn_tst = get_text(text, hps, is_symbol)
         with no_grad():
             x_tst = stn_tst.unsqueeze(0).to(device)
             x_tst_lengths = LongTensor([stn_tst.size(0)]).to(device)
             sid = LongTensor([speaker_id]).to(device)
-            audio = model.infer(x_tst, x_tst_lengths, sid=sid, noise_scale=.667, noise_scale_w=0.8,
+            audio = model.infer(x_tst, x_tst_lengths, sid=sid, noise_scale=ns, noise_scale_w=nsw,
                                 length_scale=1.0 / speed)[0][0, 0].data.cpu().float().numpy()
         del stn_tst, x_tst, x_tst_lengths, sid
         return "Success", (hps.data.sampling_rate, audio)
 
     return tts_fn
 
-
 def create_vc_fn(model, hps, speaker_ids):
     def vc_fn(original_speaker, target_speaker, record_audio, upload_audio):
         input_audio = record_audio if record_audio is not None else upload_audio
@@ -63,8 +67,7 @@ def create_vc_fn(model, hps, speaker_ids):
         if len(audio.shape) > 1:
             audio = librosa.to_mono(audio.transpose(1, 0))
         if sampling_rate != hps.data.sampling_rate:
-            audio = librosa.resample(
-                audio, orig_sr=sampling_rate, target_sr=hps.data.sampling_rate)
+            audio = librosa.resample(audio, orig_sr=sampling_rate, target_sr=hps.data.sampling_rate)
         with no_grad():
             y = torch.FloatTensor(audio)
             y = y / max(-y.min(), y.max()) / 0.99
@@ -83,32 +86,70 @@ def create_vc_fn(model, hps, speaker_ids):
 
     return vc_fn
 
+def get_text(text, hps, is_symbol):
+    text_norm = text_to_sequence(text, hps.symbols, [] if is_symbol else hps.data.text_cleaners)
+    if hps.data.add_blank:
+        text_norm = commons.intersperse(text_norm, 0)
+    text_norm = LongTensor(text_norm)
+    return text_norm
+
+
+def create_to_symbol_fn(hps):
+    def to_symbol_fn(is_symbol_input, input_text, temp_text):
+        return (_clean_text(input_text, hps.data.text_cleaners), input_text) if is_symbol_input \
+            else (temp_text, temp_text)
+
+    return to_symbol_fn
 
+models_info = [
+    {
+        "languages": ['日本語', '简体中文', 'English', 'Mix'],
+        "description": """
+
+                       这个模型包含公主连结Re:Dive的161名角色，能合成中日英三语。
+
+
+                       """,
+        "model_path": "./OUTPUT_MODEL/G_9700.pth",
+        "config_path": "./configs/modified_finetune_speaker.json",
+        "examples": [['大切な人の誕生日を祝えるって、すごく幸せなことなんですよね。', '佩可莉姆', '日本語', 1, False],
+                     ['その…この制服、どうですか？', '栞', '日本語', 1, False],
+                     ['你们全都给我让开！敢挡路的家伙，我统统斩了！', '矛依未', '简体中文', 1, False],
+                     ['Can you tell me how much the shirt is?', '美咲', 'English', 1, False],
+                     ['[EN]Excuse me?[EN][JA]お帰りなさい，お兄様！[JA]', '咲恋（夏日）', 'Mix', 1, False]],
+    }
+]
+
+models_tts = []
+models_vc = []
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
-    parser.add_argument("--model_dir", default="./models/G_9700.pth",
-                        help="directory to your fine-tuned model")
-    parser.add_argument("--config_dir", default="./configs/modified_finetune_speaker.json",
-                        help="directory to your model config file")
-    parser.add_argument("--share", action="store_true", default=False,
-                        help="make link public (used in colab)")
-
+    parser.add_argument("--share", action="store_true", default=False, help="share gradio app")
     args = parser.parse_args()
-    hps = utils.get_hparams_from_file(args.config_dir)
-
-    net_g = SynthesizerTrn(
-        len(hps.symbols),
-        hps.data.filter_length // 2 + 1,
-        hps.train.segment_size // hps.data.hop_length,
-        n_speakers=hps.data.n_speakers,
-        **hps.model).to(device)
-    _ = net_g.eval()
-
-    _ = utils.load_checkpoint(args.model_dir, net_g, None)
-    speaker_ids = hps.speakers
-    speakers = list(hps.speakers.keys())
-    tts_fn = create_tts_fn(net_g, hps, speaker_ids)
-    vc_fn = create_vc_fn(net_g, hps, speaker_ids)
+    for info in models_info:
+        lang = info['languages']
+        examples = info['examples']
+        config_path = info['config_path']
+        model_path = info['model_path']
+        description = info['description']
+        hps = utils.get_hparams_from_file(config_path)
+
+        net_g = SynthesizerTrn(
+            len(hps.symbols),
+            hps.data.filter_length // 2 + 1,
+            hps.train.segment_size // hps.data.hop_length,
+            n_speakers=hps.data.n_speakers,
+            **hps.model).to(device)
+        _ = net_g.eval()
+
+        _ = utils.load_checkpoint(model_path, net_g, None)
+        speaker_ids = hps.speakers
+        speakers = list(hps.speakers.keys())
+        models_tts.append((description, speakers, lang, examples,
+                            hps.symbols, create_tts_fn(net_g, hps, speaker_ids),
+                            create_to_symbol_fn(hps)))
+        models_vc.append((description, speakers, create_vc_fn(net_g, hps, speaker_ids)))
+
     app = gr.Blocks()
     with app:
         gr.Markdown(
@@ -119,46 +160,88 @@ if __name__ == "__main__":
             "[![Duplicate this Space](https://huggingface.co/datasets/huggingface/badges/raw/main/duplicate-this-space-sm-dark.svg)](https://huggingface.co/spaces/FrankZxShen/vits-fast-finetuning-pcr?duplicate=true)\n\n"
             "[![Finetune your own model](https://badgen.net/badge/icon/github?icon=github&label=Finetune%20your%20own%20model)](https://github.com/Plachtaa/VITS-fast-fine-tuning)"
         )
-        with gr.Tab("Text-to-Speech"):
-            with gr.Row():
-                with gr.Column():
-                    textbox = gr.TextArea(label="Text",
-                                          placeholder="Type your sentence here",
-                                          value="新たなキャラを解放できるようになったようですね。", elem_id=f"tts-input")
-                    # select character
-                    char_dropdown = gr.Dropdown(
-                        choices=speakers, value=speakers[0], label='character')
-                    language_dropdown = gr.Dropdown(
-                        choices=lang, value=lang[0], label='language')
-                    duration_slider = gr.Slider(minimum=0.1, maximum=5, value=1, step=0.1,
-                                                label='速度 Speed')
-                with gr.Column():
-                    text_output = gr.Textbox(label="Message")
-                    audio_output = gr.Audio(
-                        label="Output Audio", elem_id="tts-audio")
-                    btn = gr.Button("Generate!")
-                    btn.click(tts_fn,
-                              inputs=[textbox, char_dropdown,
-                                      language_dropdown, duration_slider, ],
-                              outputs=[text_output, audio_output])
-        with gr.Tab("Voice Conversion"):
-            gr.Markdown("""
-                            录制或上传声音，并选择要转换的音色。
-            """)
-            with gr.Column():
-                record_audio = gr.Audio(
-                    label="record your voice", source="microphone")
-                upload_audio = gr.Audio(
-                    label="or upload audio here", source="upload")
-                source_speaker = gr.Dropdown(
-                    choices=speakers, value=speakers[0], label="source speaker")
-                target_speaker = gr.Dropdown(
-                    choices=speakers, value=speakers[0], label="target speaker")
-            with gr.Column():
-                message_box = gr.Textbox(label="Message")
-                converted_audio = gr.Audio(label='converted audio')
-            btn = gr.Button("Convert!")
-            btn.click(vc_fn, inputs=[source_speaker, target_speaker, record_audio, upload_audio],
-                      outputs=[message_box, converted_audio])
-    webbrowser.open("http://127.0.0.1:7899")
-    app.queue(concurrency_count=1, api_open=False).launch(share=args.share)
+        gr.Markdown("# TTS&Voice Conversion for Princess Connect! Re:Dive\n\n"
+                    )
+        with gr.Tabs():
+            with gr.Tab("TTS"):
+                for i, (description, speakers, lang, example, symbols, tts_fn, to_symbol_fn) in enumerate(
+                        models_tts):
+                    gr.Markdown(description)
+                    with gr.Row():
+                        with gr.Column():
+                            textbox = gr.TextArea(label="Text",
+                                                    placeholder="Type your sentence here (Maximum 150 words)",
+                                                    value="新たなキャラを解放できるようになったようですね。", elem_id=f"tts-input")
+                            with gr.Accordion(label="Phoneme Input", open=False):
+                                temp_text_var = gr.Variable()
+                                symbol_input = gr.Checkbox(value=False, label="Symbol input")
+                                symbol_list = gr.Dataset(label="Symbol list", components=[textbox],
+                                                            samples=[[x] for x in symbols],
+                                                            elem_id=f"symbol-list")
+                                symbol_list_json = gr.Json(value=symbols, visible=False)
+                            symbol_input.change(to_symbol_fn,
+                                                [symbol_input, textbox, temp_text_var],
+                                                [textbox, temp_text_var])
+                            symbol_list.click(None, [symbol_list, symbol_list_json], textbox,
+                                                _js=f"""
+                            (i, symbols, text) => {{
+                                let root = document.querySelector("body > gradio-app");
+                                if (root.shadowRoot != null)
+                                    root = root.shadowRoot;
+                                let text_input = root.querySelector("#tts-input").querySelector("textarea");
+                                let startPos = text_input.selectionStart;
+                                let endPos = text_input.selectionEnd;
+                                let oldTxt = text_input.value;
+                                let result = oldTxt.substring(0, startPos) + symbols[i] + oldTxt.substring(endPos);
+                                text_input.value = result;
+                                let x = window.scrollX, y = window.scrollY;
+                                text_input.focus();
+                                text_input.selectionStart = startPos + symbols[i].length;
+                                text_input.selectionEnd = startPos + symbols[i].length;
+                                text_input.blur();
+                                window.scrollTo(x, y);
+                                text = text_input.value;
+                                return text;
+                            }}""")
+                            # select character
+                            char_dropdown = gr.Dropdown(choices=speakers, value=speakers[0], label='character')
+                            language_dropdown = gr.Dropdown(choices=lang, value=lang[0], label='language')
+                            ns = gr.Slider(label="noise_scale", minimum=0.1, maximum=1.0, step=0.1, value=0.6, interactive=True)
+                            nsw = gr.Slider(label="noise_scale_w", minimum=0.1, maximum=1.0, step=0.1, value=0.668, interactive=True)
+                            duration_slider = gr.Slider(minimum=0.1, maximum=5, value=1, step=0.1,
+                                                        label='速度 Speed')
+                        with gr.Column():
+                            text_output = gr.Textbox(label="Message")
+                            audio_output = gr.Audio(label="Output Audio", elem_id="tts-audio")
+                            btn = gr.Button("Generate!")
+                            btn.click(tts_fn,
+                                        inputs=[textbox, char_dropdown, language_dropdown, ns, nsw, duration_slider,
+                                                symbol_input],
+                                        outputs=[text_output, audio_output])
+                    gr.Examples(
+                        examples=example,
+                        inputs=[textbox, char_dropdown, language_dropdown,
+                                duration_slider, symbol_input],
+                        outputs=[text_output, audio_output],
+                        fn=tts_fn
+                    )
+            with gr.Tab("Voice Conversion"):
+                for i, (description, speakers, vc_fn) in enumerate(
+                        models_vc):
+                    gr.Markdown("""
+                                    录制或上传声音，并选择要转换的音色。
+                    """)
+                    with gr.Column():
+                        record_audio = gr.Audio(label="record your voice", source="microphone")
+                        upload_audio = gr.Audio(label="or upload audio here", source="upload")
+                        source_speaker = gr.Dropdown(choices=speakers, value=speakers[0], label="source speaker")
+                        target_speaker = gr.Dropdown(choices=speakers, value=speakers[0], label="target speaker")
+                    with gr.Column():
+                        message_box = gr.Textbox(label="Message")
+                        converted_audio = gr.Audio(label='converted audio')
+                    btn = gr.Button("Convert!")
+                    btn.click(vc_fn, inputs=[source_speaker, target_speaker, record_audio, upload_audio],
+                            outputs=[message_box, converted_audio])
+
+    app.queue(concurrency_count=3).launch(show_api=False, share=args.share)
+