voice-cloner-studio 2.0.0__tar.gz

voice_cloner_studio-2.0.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Mubashir
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

voice_cloner_studio-2.0.0/PKG-INFO

@@ -0,0 +1,92 @@
+Metadata-Version: 2.4
+Name: voice-cloner-studio
+Version: 2.0.0
+Summary: Natural voice cloning in 23 languages (incl. Hindi) with a browser GUI, CLI and Python API - powered by Chatterbox Multilingual.
+Author: Mubashir
+License-Expression: MIT
+Keywords: voice cloning,tts,text-to-speech,chatterbox,hindi,gradio
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Operating System :: OS Independent
+Classifier: Topic :: Multimedia :: Sound/Audio :: Speech
+Classifier: Environment :: Web Environment
+Requires-Python: <3.14,>=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+License-File: src/voice_cloner_studio/assets/LICENSE-resemblyzer.txt
+Requires-Dist: chatterbox-tts>=0.1.7
+Requires-Dist: soundfile>=0.12
+Requires-Dist: numpy
+Requires-Dist: librosa
+Requires-Dist: pyloudnorm
+Requires-Dist: gradio
+Provides-Extra: denoise
+Requires-Dist: noisereduce>=3.0; extra == "denoise"
+Provides-Extra: test
+Requires-Dist: pytest>=8; extra == "test"
+Requires-Dist: gradio_client; extra == "test"
+Dynamic: license-file
+
+# 🎙️ Voice Cloner Studio
+
+Natural voice cloning in **23 languages, including Hindi**. You get a browser GUI, a one-line command and a simple Python API.
+It is powered by [Chatterbox Multilingual](https://github.com/resemble-ai/chatterbox) from Resemble AI, an open model (MIT licence) whose naturalness is in the same league as commercial services.
+
+```bash
+pip install voice-cloner-studio
+voice-cloner-studio            # opens the GUI at http://127.0.0.1:7860
+```
+
+## Python, in 4 lines
+
+```python
+from voice_cloner_studio import VoiceCloner
+
+vc = VoiceCloner()
+vc.clone("my_voice", "my_voice.wav")                                  # once - saved for next time
+vc.speak("Hello everyone! Welcome to my channel.", voice="my_voice")  # -> dict with "path", "similarity"
+vc.speak("नमस्ते दोस्तों!", voice="my_voice", language="Hindi", style="Expressive / energetic")
+```
+
+## Command line
+
+```bash
+voice-clone --ref my_voice.wav --voice me --text "Hello there!" --out hello.wav
+voice-clone --voice me --language Hindi --text "नमस्ते, आप कैसे हैं?"
+voice-clone --voice me --text-file story.txt --style "Calm narrator" --takes 3
+```
+
+## Features
+
+- **Clone** a voice from a clip or a microphone recording. The app cleans the clip, normalizes loudness and shows a quality report.
+- **Choose a style:** Natural, Expressive, Calm narrator, Fast speaker, or Accent fix for speaking another language. You can also fine-tune emotion, pace and variation.
+- **Best of N takes:** the app keeps the take that sounds most like the voice. A built-in speaker-verification model scores how close each result is.
+- **Built-in default voice:** you can generate speech without cloning anything.
+- **Long texts** are split into sentences automatically. A blank line adds a paragraph pause.
+- **Voice library:** clone once and reuse forever.
+- **Clear errors:** the GUI shows the real reason, and full details go to `error.log`.
+
+Voices and outputs are saved in `VoiceClonerStudio` in your home folder. Change this with `--data-dir`.
+The model (~2 GB) downloads once, on first use.
+
+## Best results
+
+| | Do this |
+|---|---|
+| Reference | **One clean 10-20 s clip.** The model listens to the first ~10 s |
+| Recording | Quiet room, no music or echo, normal volume |
+| Delivery | Speak in the tone you want, because the clone copies it |
+| Other language | Use the *Other-language accent fix* style |
+| Bad take | Set Takes to 3, or change the seed |
+
+An NVIDIA GPU makes generation real-time. On a laptop CPU, expect about 3-10 s of waiting per second of speech.
+
+## Responsible use
+
+Clone only your own voice or voices you have permission to use. Chatterbox adds an inaudible watermark ([Perth](https://github.com/resemble-ai/perth)) to generated audio, so it can be identified as AI-made.
+
+## Licence
+
+The code is under the MIT licence. Chatterbox is MIT, and the bundled Resemblyzer speaker encoder is Apache-2.0.

voice_cloner_studio-2.0.0/README.md

@@ -0,0 +1,61 @@
+# 🎙️ Voice Cloner Studio
+
+Natural voice cloning in **23 languages, including Hindi**. You get a browser GUI, a one-line command and a simple Python API.
+It is powered by [Chatterbox Multilingual](https://github.com/resemble-ai/chatterbox) from Resemble AI, an open model (MIT licence) whose naturalness is in the same league as commercial services.
+
+```bash
+pip install voice-cloner-studio
+voice-cloner-studio            # opens the GUI at http://127.0.0.1:7860
+```
+
+## Python, in 4 lines
+
+```python
+from voice_cloner_studio import VoiceCloner
+
+vc = VoiceCloner()
+vc.clone("my_voice", "my_voice.wav")                                  # once - saved for next time
+vc.speak("Hello everyone! Welcome to my channel.", voice="my_voice")  # -> dict with "path", "similarity"
+vc.speak("नमस्ते दोस्तों!", voice="my_voice", language="Hindi", style="Expressive / energetic")
+```
+
+## Command line
+
+```bash
+voice-clone --ref my_voice.wav --voice me --text "Hello there!" --out hello.wav
+voice-clone --voice me --language Hindi --text "नमस्ते, आप कैसे हैं?"
+voice-clone --voice me --text-file story.txt --style "Calm narrator" --takes 3
+```
+
+## Features
+
+- **Clone** a voice from a clip or a microphone recording. The app cleans the clip, normalizes loudness and shows a quality report.
+- **Choose a style:** Natural, Expressive, Calm narrator, Fast speaker, or Accent fix for speaking another language. You can also fine-tune emotion, pace and variation.
+- **Best of N takes:** the app keeps the take that sounds most like the voice. A built-in speaker-verification model scores how close each result is.
+- **Built-in default voice:** you can generate speech without cloning anything.
+- **Long texts** are split into sentences automatically. A blank line adds a paragraph pause.
+- **Voice library:** clone once and reuse forever.
+- **Clear errors:** the GUI shows the real reason, and full details go to `error.log`.
+
+Voices and outputs are saved in `VoiceClonerStudio` in your home folder. Change this with `--data-dir`.
+The model (~2 GB) downloads once, on first use.
+
+## Best results
+
+| | Do this |
+|---|---|
+| Reference | **One clean 10-20 s clip.** The model listens to the first ~10 s |
+| Recording | Quiet room, no music or echo, normal volume |
+| Delivery | Speak in the tone you want, because the clone copies it |
+| Other language | Use the *Other-language accent fix* style |
+| Bad take | Set Takes to 3, or change the seed |
+
+An NVIDIA GPU makes generation real-time. On a laptop CPU, expect about 3-10 s of waiting per second of speech.
+
+## Responsible use
+
+Clone only your own voice or voices you have permission to use. Chatterbox adds an inaudible watermark ([Perth](https://github.com/resemble-ai/perth)) to generated audio, so it can be identified as AI-made.
+
+## Licence
+
+The code is under the MIT licence. Chatterbox is MIT, and the bundled Resemblyzer speaker encoder is Apache-2.0.

voice_cloner_studio-2.0.0/pyproject.toml

@@ -0,0 +1,49 @@
+[build-system]
+requires = ["setuptools>=68", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "voice-cloner-studio"
+version = "2.0.0"
+description = "Natural voice cloning in 23 languages (incl. Hindi) with a browser GUI, CLI and Python API - powered by Chatterbox Multilingual."
+readme = "README.md"
+requires-python = ">=3.10,<3.14"
+license = "MIT"
+license-files = ["LICENSE", "src/voice_cloner_studio/assets/LICENSE-resemblyzer.txt"]
+authors = [{ name = "Mubashir" }]
+keywords = ["voice cloning", "tts", "text-to-speech", "chatterbox", "hindi", "gradio"]
+classifiers = [
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Operating System :: OS Independent",
+    "Topic :: Multimedia :: Sound/Audio :: Speech",
+    "Environment :: Web Environment",
+]
+dependencies = [
+    "chatterbox-tts>=0.1.7",
+    "soundfile>=0.12",
+    "numpy",
+    "librosa",
+    "pyloudnorm",
+    "gradio",
+]
+
+[project.optional-dependencies]
+denoise = ["noisereduce>=3.0"]
+test = ["pytest>=8", "gradio_client"]
+
+[project.scripts]
+voice-cloner-studio = "voice_cloner_studio.app:main"
+voice-clone = "voice_cloner_studio.cli:main"
+
+[tool.setuptools.packages.find]
+where = ["src"]
+
+[tool.setuptools.package-data]
+voice_cloner_studio = ["assets/*"]
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+filterwarnings = ["ignore::DeprecationWarning", "ignore::FutureWarning", "ignore::UserWarning"]

voice_cloner_studio-2.0.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

voice_cloner_studio-2.0.0/src/voice_cloner_studio/__init__.py

@@ -0,0 +1,10 @@
+"""Voice Cloner Studio - natural voice cloning (Chatterbox Multilingual) with a browser GUI.
+
+    from voice_cloner_studio import VoiceCloner
+    vc = VoiceCloner()
+    vc.clone("me", "my_voice.wav")
+    print(vc.speak("Hello world!", voice="me")["path"])
+"""
+__version__ = "2.0.0"
+
+from .engine import BUILTIN, LANGUAGES, STYLES, VoiceCloner  # noqa: E402,F401

voice_cloner_studio-2.0.0/src/voice_cloner_studio/__main__.py

@@ -0,0 +1,4 @@
+from voice_cloner_studio.app import main
+
+if __name__ == "__main__":
+    main()

voice_cloner_studio-2.0.0/src/voice_cloner_studio/app.py

@@ -0,0 +1,296 @@
+"""
+Voice Cloner Studio - browser GUI.
+
+Run:  voice-cloner-studio        (or: python -m voice_cloner_studio)
+"""
+import argparse
+import datetime as dt
+import functools
+import os
+import sys
+import traceback
+from pathlib import Path
+
+import gradio as gr
+import numpy as np
+
+from . import __version__
+from .engine import BUILTIN, LANGUAGES, STYLES, VoiceCloner, default_data_dir, split_text
+
+SIM_LEVELS = [(0.85, "🟢 Excellent - very hard to tell apart"),
+              (0.75, "🟢 Good - in the same range as two real recordings of this person"),
+              (0.65, "🟡 Fair - similar, but differences are noticeable"),
+              (0.00, "🔴 Weak - use a cleaner / longer reference clip")]
+
+TIPS = """
+### For the most natural, attractive clone
+1. **One clean clip of 10-20 s** beats many clips - the model listens to the first ~10 s.
+2. Quiet room, no music/echo, mic ~15 cm away, normal speaking volume.
+3. **Speak the way you want the output to sound** - warm reference → warm clone.
+4. Pick a **Style**: *Expressive* for ads/YouTube, *Calm narrator* for audiobooks.
+5. Speaking another language than the reference? Use *Other-language accent fix*.
+6. Set **Takes = 3** - the app keeps the take that sounds most like you.
+7. Hindi: type in Devanagari (नमस्ते). Urdu: type it in Devanagari or Roman script.
+"""
+
+CSS = "#title {text-align:center} #title h1 {font-size:2.2rem;margin-bottom:0}"
+
+engine: VoiceCloner = None
+
+
+def sim_label(score):
+    return next(label for th, label in SIM_LEVELS if score >= th)
+
+
+def friendly_errors(fn):
+    """Show the real reason in the GUI (not a bare "Error") and keep the traceback in error.log."""
+    @functools.wraps(fn)
+    def wrapper(*args, **kwargs):
+        try:
+            return fn(*args, **kwargs)
+        except gr.Error:
+            raise
+        except Exception as e:
+            log = engine.data_dir / "error.log"
+            with open(log, "a", encoding="utf-8") as f:
+                f.write(f"\n===== {dt.datetime.now():%Y-%m-%d %H:%M:%S} {fn.__name__} (v{__version__})\n"
+                        f"{traceback.format_exc()}")
+            traceback.print_exc()
+            msg = f"{type(e).__name__}: {e}"
+            if any(k in msg for k in ("ProxyError", "ConnectionError", "ConnectError", "Max retries",
+                                      "getaddrinfo", "403", "timed out", "LocalEntryNotFound")):
+                msg = ("Couldn't download the voice model (~2 GB). Check your internet / firewall and that "
+                       "huggingface.co is reachable, then try again. Details: " + msg)
+            raise gr.Error(f"{msg[:600]}  (full details in {log})")
+    return wrapper
+
+
+def voice_choices():
+    return [BUILTIN] + engine.list_voices()
+
+
+def library_rows():
+    rows = []
+    for v in engine.list_voices():
+        m = engine.voice_info(v)
+        rows.append([v, m.get("created", ""), m.get("clips", ""), m.get("seconds", ""), m.get("notes", "")])
+    return rows
+
+
+@friendly_errors
+def ui_clone(name, files, mic, denoise, notes, consent, progress=gr.Progress()):
+    if not consent:
+        raise gr.Error("Please confirm you have the speaker's consent before cloning a voice.")
+    sources = [f if isinstance(f, str) else f.name for f in (files or [])]
+    if mic:
+        sources.append(mic)
+    if not engine.safe_name(name):
+        raise gr.Error("Give the voice a name.")
+    if not sources:
+        raise gr.Error("Upload or record at least one reference clip.")
+    progress(0.2, desc="Loading model (first time downloads ~2 GB)")
+    _ = engine.model
+    progress(0.6, desc="Learning the voice")
+    name, rows, total = engine.clone(name, sources, denoise=denoise, notes=notes)
+
+    lines = [f"### ✅ Voice **{name}** created", "", "| Clip | Speech | Level | Clipping |", "|---|---|---|---|"]
+    lines += [f"| {n} | {d:.1f}s | {lv:.1f} dBFS | {c:.2f}% |" for n, d, lv, c in rows]
+    lines.append(f"\n**Total clean speech: {total:.1f}s**")
+    if total < 5:
+        lines.append("\n⚠️ Under 5 s - add a longer clip (10-20 s is ideal).")
+    elif total > 25:
+        lines.append("\nℹ️ Only the first ~15 s are used - put your best clip first.")
+    else:
+        lines.append("\n🟢 Good length.")
+    if any(r[3] > 0.5 for r in rows):
+        lines.append("\n⚠️ Some audio is clipping/distorted - re-record a bit quieter.")
+    choices = voice_choices()
+    return ("\n".join(lines), gr.update(choices=choices, value=name),
+            gr.update(choices=engine.list_voices(), value=name), library_rows())
+
+
+@friendly_errors
+def ui_speak(voice, text, language, style, exaggeration, cfg_weight, temperature, seed, takes,
+             progress=gr.Progress()):
+    if not (text or "").strip():
+        raise gr.Error("Type some text to speak.")
+    progress(0.02, desc="Loading model (first time downloads ~2 GB)")
+    r = engine.speak(text, voice=voice, language=language, exaggeration=exaggeration, cfg_weight=cfg_weight,
+                     temperature=temperature, seed=seed, takes=takes,
+                     on_progress=lambda f, d: progress(f, desc=d))
+    info = [f"Generated **{r['seconds']:.1f}s** · seed **{r['seed']}** · `{Path(r['path']).name}`"]
+    if r["similarity"] is not None:
+        info.append(f"\n### Voice similarity: **{r['similarity'] * 100:.1f}%**  \n{sim_label(r['similarity'])}")
+        if len(r["scores"]) > 1:
+            info.append("\nBest of {} takes (scores: {}).".format(
+                len(r["scores"]), ", ".join(f"{s * 100:.1f}%" for s in r["scores"])))
+        info.append("\n<sub>Independent speaker-verification score. Two real recordings of the same "
+                    "person typically score 75-90%; different people 45-70%.</sub>")
+    return r["path"], r["path"], "\n".join(info), history()
+
+
+def ui_style(style):
+    e, c, t = STYLES[style]
+    return e, c, t
+
+
+def ui_delete(name):
+    if name:
+        engine.delete_voice(name)
+    vs = engine.list_voices()
+    return (gr.update(choices=[BUILTIN] + vs, value=BUILTIN), gr.update(choices=vs, value=vs[0] if vs else None),
+            library_rows(), f"Deleted **{name}**." if name else "Nothing selected.")
+
+
+def ui_preview(name):
+    p = engine.reference_path(name) if name else None
+    return str(p) if p else None
+
+
+def history():
+    files = sorted(engine.output_dir.glob("*.wav"), key=lambda p: p.stat().st_mtime, reverse=True)
+    return [str(p) for p in files[:30]]
+
+
+def build_ui():
+    voices = engine.list_voices()
+    with gr.Blocks(title="Voice Cloner Studio") as demo:
+        gr.Markdown(f"# 🎙️ Voice Cloner Studio <small>v{__version__}</small>\n"
+                    f"Natural voice cloning in {len(LANGUAGES)} languages (incl. Hindi) · "
+                    f"engine: Chatterbox Multilingual · running on **{engine.device.upper()}**", elem_id="title")
+        with gr.Tabs():
+            with gr.Tab("1 · Clone a voice"):
+                with gr.Row():
+                    with gr.Column(scale=3):
+                        c_name = gr.Textbox(label="Voice name", placeholder="e.g. my_voice")
+                        c_files = gr.File(label="Reference clip(s) - one clean 10-20 s clip is ideal",
+                                          file_count="multiple", file_types=["audio"])
+                        c_mic = gr.Audio(label="…or record: read a paragraph naturally for 15 s",
+                                         sources=["microphone"], type="filepath")
+                        c_denoise = gr.Checkbox(label="Reduce background noise (only if your recording is noisy)",
+                                                value=False)
+                        c_notes = gr.Textbox(label="Notes (optional)")
+                        c_consent = gr.Checkbox(label="I own this voice or have the speaker's explicit permission "
+                                                      "to clone it, and won't use it to deceive or impersonate.")
+                        c_btn = gr.Button("🧬 Create voice", variant="primary", size="lg")
+                    with gr.Column(scale=2):
+                        c_report = gr.Markdown("Upload or record audio, then press **Create voice**.")
+                        gr.Markdown(TIPS)
+
+            with gr.Tab("2 · Speak"):
+                with gr.Row():
+                    with gr.Column(scale=3):
+                        with gr.Row():
+                            s_voice = gr.Dropdown([BUILTIN] + voices, value=voices[0] if voices else BUILTIN,
+                                                  label="Voice", scale=3)
+                            s_lang = gr.Dropdown(list(LANGUAGES), value="English", label="Language", scale=2)
+                        s_text = gr.Textbox(label="Text", lines=7,
+                                            value="Hello! This is my cloned voice. Sounds natural, doesn't it?",
+                                            placeholder="Type anything. Blank line = paragraph pause.")
+                        s_style = gr.Radio(list(STYLES), value="Natural", label="Style")
+                        with gr.Accordion("Fine-tune", open=False):
+                            s_exag = gr.Slider(0.25, 2.0, 0.5, step=0.05, label="Emotion / expressiveness")
+                            s_cfg = gr.Slider(0.0, 1.0, 0.5, step=0.05,
+                                              label="Pace & accent guidance (lower = slower, less accent carry-over)")
+                            s_temp = gr.Slider(0.05, 1.5, 0.8, step=0.05, label="Variation (temperature)")
+                            s_seed = gr.Number(value=-1, precision=0, label="Seed (-1 = random)")
+                        s_takes = gr.Slider(1, 5, 1, step=1, label="Takes - keep the one most like the voice")
+                        s_btn = gr.Button("🔊 Generate speech", variant="primary", size="lg")
+                    with gr.Column(scale=2):
+                        s_audio = gr.Audio(label="Result", type="filepath", interactive=False)
+                        s_file = gr.File(label="Download WAV")
+                        s_info = gr.Markdown()
+                        s_hist = gr.Files(label="Recent generations", value=history())
+
+            with gr.Tab("3 · Voice library"):
+                l_table = gr.Dataframe(headers=["Voice", "Created", "Clips", "Seconds", "Notes"],
+                                       value=library_rows(), interactive=False)
+                with gr.Row():
+                    l_voice = gr.Dropdown(voices, value=voices[0] if voices else None, label="Voice")
+                    l_play = gr.Audio(label="Reference used for this voice", interactive=False)
+                with gr.Row():
+                    l_refresh = gr.Button("🔄 Refresh")
+                    l_del = gr.Button("🗑️ Delete voice", variant="stop")
+                l_msg = gr.Markdown()
+
+        c_btn.click(ui_clone, [c_name, c_files, c_mic, c_denoise, c_notes, c_consent],
+                    [c_report, s_voice, l_voice, l_table], api_name="clone")
+        s_style.change(ui_style, s_style, [s_exag, s_cfg, s_temp], api_name="style")
+        s_btn.click(ui_speak, [s_voice, s_text, s_lang, s_style, s_exag, s_cfg, s_temp, s_seed, s_takes],
+                    [s_audio, s_file, s_info, s_hist], api_name="speak")
+        l_voice.change(ui_preview, l_voice, l_play, api_name="preview")
+        l_del.click(ui_delete, l_voice, [s_voice, l_voice, l_table, l_msg], api_name="delete")
+        l_refresh.click(lambda: (library_rows(), gr.update(choices=voice_choices()),
+                                 gr.update(choices=engine.list_voices())),
+                        None, [l_table, s_voice, l_voice], api_name="refresh")
+    return demo
+
+
+def launch(data_dir=None, host="127.0.0.1", port=7860, share=False, inbrowser=True, prevent_thread_lock=False,
+           cloner=None):
+    global engine
+    engine = cloner or VoiceCloner(data_dir)
+    demo = build_ui().queue()
+    # Gradio only serves files from temp/cwd unless told otherwise; voices and outputs live in data_dir.
+    demo.launch(server_name=host, server_port=port, share=share, inbrowser=inbrowser, show_error=True,
+                prevent_thread_lock=prevent_thread_lock, theme=gr.themes.Soft(primary_hue="violet"), css=CSS,
+                allowed_paths=[str(engine.voices_dir), str(engine.output_dir)])
+    return demo
+
+
+def self_check(data_dir=None):
+    import importlib
+    import platform
+    import urllib.request
+    ok = True
+
+    def item(name, fn):
+        nonlocal ok
+        try:
+            detail = fn()
+            print(f"  [OK]   {name}{': ' + str(detail) if detail not in (None, '') else ''}")
+        except Exception as e:
+            ok = False
+            print(f"  [FAIL] {name}: {type(e).__name__}: {e}")
+
+    def writable(d):
+        d.mkdir(parents=True, exist_ok=True)
+        (d / ".write_test").write_text("ok")
+        (d / ".write_test").unlink()
+        return d
+
+    d = Path(data_dir or default_data_dir())
+    print(f"Voice Cloner Studio v{__version__} - self check")
+    item("Python", lambda: f"{platform.python_version()} ({platform.architecture()[0]})")
+    import torch
+    item("PyTorch", lambda: f"{torch.__version__}, GPU: {torch.cuda.get_device_name(0) if torch.cuda.is_available() else 'none (CPU mode, slower)'}")
+    for mod in ["chatterbox", "gradio", "librosa", "soundfile", "pyloudnorm"]:
+        item(f"package {mod}", lambda m=mod: getattr(importlib.import_module(m), "__version__", ""))
+    item("text splitter", lambda: f"{len(split_text('Hello there. नमस्ते।', 'hi'))} chunk(s)")
+    item("similarity scorer", lambda: VoiceCloner(d)._embed(np.random.randn(16000).astype(np.float32) * .1, 16000).shape)
+    item("data folder writable", lambda: writable(d))
+    item("huggingface.co reachable (model download)",
+         lambda: urllib.request.urlopen("https://huggingface.co/api/models/ResembleAI/chatterbox", timeout=15).status)
+    print("\nAll good - start the app with: voice-cloner-studio" if ok else "\nSome checks failed - see above.")
+    return 0 if ok else 1
+
+
+def main(argv=None):
+    ap = argparse.ArgumentParser(prog="voice-cloner-studio", description="Voice Cloner Studio GUI")
+    ap.add_argument("--host", default="127.0.0.1")
+    ap.add_argument("--port", type=int, default=7860)
+    ap.add_argument("--share", action="store_true", help="also create a public gradio.live link")
+    ap.add_argument("--data-dir", help=f"where voices/outputs are stored (default: {default_data_dir()})")
+    ap.add_argument("--no-browser", action="store_true", help="don't open a browser tab")
+    ap.add_argument("--check", action="store_true", help="check this PC's setup and exit")
+    args = ap.parse_args(argv)
+    if args.check:
+        sys.exit(self_check(args.data_dir))
+    print(f"[voice-cloner] Voice Cloner Studio v{__version__}")
+    print(f"[voice-cloner] Data folder: {args.data_dir or default_data_dir()}")
+    print(f"[voice-cloner] Opening http://{args.host}:{args.port} - keep this window open while using the app.")
+    launch(args.data_dir, args.host, args.port, args.share, not args.no_browser)
+
+
+if __name__ == "__main__":
+    main()