Spaces:
Configuration error
Configuration error
| import os | |
| import random | |
| import numpy as np | |
| import torch | |
| import librosa | |
| import soundfile as sf | |
| import gradio as gr | |
| try: | |
| import spaces | |
| GPU_DECORATOR = spaces.GPU(duration=60) | |
| except ImportError: | |
| def GPU_DECORATOR(func): | |
| return func | |
| from chatterbox.tts import ChatterboxTTS | |
| from chatterbox.vc import ChatterboxVC | |
| DEVICE = "cuda" if torch.cuda.is_available() else "cpu" | |
| print(f"Initializing Chatterbox on {DEVICE}...") | |
| tts_model = ChatterboxTTS.from_pretrained("cpu") | |
| vc_model = ChatterboxVC.from_pretrained("cpu") | |
| if hasattr(tts_model, "t3") and hasattr(tts_model.t3, "tfmr"): | |
| if hasattr(tts_model.t3.tfmr, "config"): | |
| tts_model.t3.tfmr.config._attn_implementation = "eager" | |
| def sanitize_reference_audio(audio_path, max_duration=20.0): | |
| if not audio_path: | |
| return None | |
| wav, sr = librosa.load(audio_path, sr=24000, mono=True) | |
| max_samples = int(max_duration * sr) | |
| if len(wav) > max_samples: | |
| wav = wav[:max_samples] | |
| max_val = np.max(np.abs(wav)) | |
| if max_val > 0: | |
| wav = (wav / max_val) * 0.95 | |
| clean_path = "/tmp/clean_ref_prompt.wav" | |
| sf.write(clean_path, wav, sr, subtype="PCM_16") | |
| return clean_path | |
| def generate_tts(text, audio_prompt_path, exaggeration, temperature, seed_num, cfgw, min_p, top_p, repetition_penalty): | |
| text = text.strip() | |
| if not text: | |
| raise gr.Error("Please enter text to synthesize.") | |
| if len(text) > 450: | |
| text = text[:450] | |
| target_dev = "cuda" if torch.cuda.is_available() else "cpu" | |
| tts_model.device = target_dev | |
| tts_model.t3.to(target_dev) | |
| tts_model.s3gen.to(target_dev) | |
| tts_model.ve.to(target_dev) | |
| if tts_model.conds is not None: | |
| tts_model.conds.to(target_dev) | |
| if int(seed_num) != 0: | |
| torch.manual_seed(int(seed_num)) | |
| if torch.cuda.is_available(): | |
| torch.cuda.manual_seed_all(int(seed_num)) | |
| random.seed(int(seed_num)) | |
| np.random.seed(int(seed_num)) | |
| clean_prompt = sanitize_reference_audio(audio_prompt_path) | |
| with torch.inference_mode(): | |
| wav = tts_model.generate( | |
| text, | |
| audio_prompt_path=clean_prompt, | |
| exaggeration=float(exaggeration), | |
| temperature=float(temperature), | |
| cfg_weight=float(cfgw), | |
| min_p=float(min_p), | |
| top_p=float(top_p), | |
| repetition_penalty=float(repetition_penalty), | |
| ) | |
| return (tts_model.sr, wav.squeeze(0).cpu().numpy()) | |
| def convert_voice(input_audio, target_voice_audio): | |
| if input_audio is None or target_voice_audio is None: | |
| raise gr.Error("Please upload both source audio and target voice audio.") | |
| target_dev = "cuda" if torch.cuda.is_available() else "cpu" | |
| vc_model.device = target_dev | |
| vc_model.s3gen.to(target_dev) | |
| clean_src = sanitize_reference_audio(input_audio, max_duration=30.0) | |
| clean_tgt = sanitize_reference_audio(target_voice_audio, max_duration=20.0) | |
| with torch.inference_mode(): | |
| wav = vc_model.generate( | |
| audio=clean_src, | |
| target_voice_path=clean_tgt, | |
| ) | |
| return (vc_model.sr, wav.squeeze(0).cpu().numpy()) | |
| with gr.Blocks(title="Chatterbox Web Studio") as demo: | |
| gr.Markdown("# 🎙️ Chatterbox Standalone Web Studio") | |
| with gr.Tabs(): | |
| with gr.TabItem("🗣️ Text-to-Speech (TTS)"): | |
| with gr.Row(): | |
| with gr.Column(): | |
| text = gr.Textbox(label="Text to Synthesize", lines=5, value="Welcome to Chatterbox Web Studio!") | |
| ref_wav = gr.Audio(sources=["upload", "microphone"], type="filepath", label="Reference Voice Audio (Optional - 10 to 20s WAV)", value=None) | |
| with gr.Row(): | |
| exaggeration = gr.Slider(0.25, 2.0, step=0.05, value=0.5, label="Exaggeration") | |
| cfg_weight = gr.Slider(0.0, 1.0, step=0.05, value=0.5, label="CFG / Pace Control") | |
| with gr.Accordion("⚙️ Advanced Sampler Settings", open=False): | |
| seed_num = gr.Number(value=0, label="Random Seed", precision=0) | |
| temp = gr.Slider(0.05, 2.0, step=0.05, value=0.8, label="Temperature") | |
| min_p = gr.Slider(0.00, 1.00, step=0.01, value=0.05, label="min_p") | |
| top_p = gr.Slider(0.00, 1.00, step=0.01, value=1.00, label="top_p") | |
| rep_penalty = gr.Slider(1.00, 2.00, step=0.05, value=1.20, label="Repetition Penalty") | |
| tts_btn = gr.Button("Generate Speech ⚡", variant="primary") | |
| with gr.Column(): | |
| tts_output = gr.Audio(label="Generated Output", interactive=False) | |
| tts_btn.click(fn=generate_tts, inputs=[text, ref_wav, exaggeration, temp, seed_num, cfg_weight, min_p, top_p, rep_penalty], outputs=tts_output) | |
| with gr.TabItem("🔄 Voice Conversion (VC)"): | |
| with gr.Row(): | |
| with gr.Column(): | |
| src_audio = gr.Audio(sources=["upload", "microphone"], type="filepath", label="Source Audio") | |
| tgt_audio = gr.Audio(sources=["upload", "microphone"], type="filepath", label="Target Voice Audio") | |
| vc_btn = gr.Button("Convert Voice 🔄", variant="primary") | |
| with gr.Column(): | |
| vc_output = gr.Audio(label="Converted Audio Output", interactive=False) | |
| vc_btn.click(fn=convert_voice, inputs=[src_audio, tgt_audio], outputs=vc_output) | |
| if __name__ == "__main__": | |
| demo.queue().launch() |