import inspect import gradio as gr def generate(model, text, **kwargs): """Run a TTS model and show any failure in the UI""" if not text or not text.strip(): raise gr.Error("Enter some text to speak") try: return model.generate_speech(text, **kwargs) except Exception as e: raise gr.Error(f"Error: {e}") def transcribe(model, audio_path, language): """Run an STT model and return its text, or the error""" try: return model.transcribe( audio_path, language=None if language == "auto" else language ) except Exception as e: return f"Error: {e}" def voice_dropdown(voices, value=None, **kwargs): return gr.Dropdown( choices=voices, value=value or (voices[0] if voices else None), label="Voice", **kwargs, ) def language_and_voice(model): """A language dropdown that refills the voice dropdown when it changes""" languages = model.get_supported_languages() language = gr.Dropdown( choices=languages, value=languages[0] if languages else None, label="Language" ) # API calls are checked against the initial choices, so allow voices of the other languages too voice = voice_dropdown( model.get_available_voices(languages[0]) if languages else [], allow_custom_value=True, ) def update_voices(name): voices = model.get_available_voices(name) return gr.update(choices=voices, value=voices[0] if voices else None) language.change( update_voices, inputs=[language], outputs=[voice], api_visibility="private" ) return {"language": language, "voice": voice} def tts_tab( title, api_name, model, controls, placeholder="Enter text here...", prepare=None ): """One TTS tab: a text box, the model's own controls, a button and an audio player. controls() builds the extra inputs inside the tab and returns {generate_speech keyword: component}. prepare(text, kwargs), when given, may change those keywords before synthesis. """ with gr.Tab(title), gr.Row(): with gr.Column(): text_box = gr.Textbox( label="Text to speak", placeholder=placeholder, lines=5 ) inputs = controls() submit = gr.Button("Generate Speech") with gr.Column(): output = gr.Audio(label="Generated Speech") def synthesize(text, *values): kwargs = dict(zip(inputs, values)) return generate(model, text, **(prepare(text, kwargs) if prepare else kwargs)) # Gradio names API parameters from the signature, so give *values the control names and optional defaults synthesize.__signature__ = inspect.Signature( [inspect.Parameter("text", inspect.Parameter.POSITIONAL_OR_KEYWORD)] + [ inspect.Parameter( name, inspect.Parameter.POSITIONAL_OR_KEYWORD, default=None ) for name in inputs ] ) submit.click( synthesize, inputs=[text_box, *inputs.values()], outputs=[output], api_name=api_name, ) def stt_tab(title, api_name, model): """One STT tab: an audio input, a language picker from model.languages, and the transcription""" with gr.Tab(title), gr.Row(): with gr.Column(): audio = gr.Audio(label="Audio to transcribe", type="filepath") language = gr.Dropdown( choices=[("Auto-detect", "auto"), *model.languages.items()], value="auto", label="Language (optional)", ) submit = gr.Button("Transcribe") with gr.Column(): output = gr.Textbox(label="Transcription", lines=5) def transcribe_audio(audio, lang): return transcribe(model, audio, lang) submit.click( transcribe_audio, inputs=[audio, language], outputs=[output], api_name=api_name, ) def create_interface(tts_models, stt_models): """Build the gallery from the model instances the app loaded""" chatterbox = tts_models["ResembleAI/chatterbox"] kitten = tts_models["KittenML/KittenTTS"] kitten2 = tts_models["KittenML/kitten-tts-2"] whisper = stt_models["SYSTRAN/faster-whisper"] def add_reference_transcript(text, kwargs): # Reuse the loaded Faster Whisper; KittenTTS 2 would otherwise download its own Whisper if ( text and text.strip() and kwargs["audio_prompt"] and not (kwargs["reference_text"] or "").strip() ): try: kwargs["reference_text"] = whisper.transcribe(kwargs["audio_prompt"]) except Exception as e: print( f"Faster Whisper could not transcribe the reference, KittenTTS 2 will: {e}" ) return kwargs with gr.Blocks(title="TTS & STT Gallery") as demo: gr.Markdown("# TTS & STT Model Gallery") gr.Markdown("Explore different Text-to-Speech and Speech-to-Text models") with gr.Tabs(): with gr.Tab("Text-to-Speech"): gr.Markdown("## Text-to-Speech Models") with gr.Tabs(): tts_tab( "Chatterbox", "tts_chatterbox", chatterbox, lambda: { "language": gr.Dropdown( choices=chatterbox.get_supported_languages(), value="English", label="Language", ), "audio_prompt": gr.Audio( label="Voice reference (optional)", type="filepath" ), }, ) tts_tab( "KittenTTS", "tts_kittentts", kitten, lambda: { "voice": voice_dropdown(kitten.get_available_voices()), }, ) tts_tab( "KittenTTS 2", "tts_kittentts2", kitten2, lambda: { "voice": voice_dropdown( kitten2.get_available_voices(), kitten2.get_default_voice(), ), "preset": gr.Radio( choices=["stable", "expressive"], value="stable", label="Preset", ), "audio_prompt": gr.Audio( label="Voice reference to clone, 5-30 s (optional, replaces the voice above)", type="filepath", ), "reference_text": gr.Textbox( label="Reference transcript (optional, transcribed automatically when empty)" ), }, placeholder="Enter text here... Supports [joyful] emotion tags, events and (((emphasis)))", prepare=add_reference_transcript, ) for title, name in [ ("Piper", "piper-tts"), ("Kokoro", "hexgrad/kokoro"), ]: model = tts_models[name] tts_tab( title, f"tts_{title.lower()}", model, lambda model=model: language_and_voice(model), ) with gr.Tab("Speech-to-Text"): gr.Markdown("## Speech-to-Text Models") with gr.Tabs(): stt_tab("Faster Whisper", "stt_whisper", whisper) stt_tab( "Whistle", "stt_whistle", stt_models["Cactus-Compute/whistle"] ) return demo