Spaces:
Sleeping
Sleeping
| import gradio as gr | |
| from TTS.api import TTS | |
| import tempfile | |
| import docx | |
| # Voice models dictionary with metadata on whether they support multi-speaker | |
| VOICE_MODELS = { | |
| "LJSpeech (Standard Female)": { | |
| "model_name": "tts_models/en/ljspeech/vits", | |
| "multi_speaker": False | |
| }, | |
| "VCTK (Multi-speaker English)": { | |
| "model_name": "tts_models/en/vctk/vits", | |
| "multi_speaker": True | |
| } | |
| } | |
| # Embedded short speaker metadata (from your CSV) | |
| SPEAKER_METADATA = { | |
| "225": {"age": 23, "gender": "F", "accent": "English"}, | |
| "226": {"age": 22, "gender": "M", "accent": "English"}, | |
| "227": {"age": 38, "gender": "M", "accent": "English"}, | |
| "228": {"age": 22, "gender": "F", "accent": "English"}, | |
| "229": {"age": 23, "gender": "F", "accent": "English"}, | |
| "230": {"age": 22, "gender": "F", "accent": "English"}, | |
| "231": {"age": 23, "gender": "F", "accent": "English"}, | |
| "232": {"age": 23, "gender": "M", "accent": "English"}, | |
| "233": {"age": 23, "gender": "F", "accent": "English"}, | |
| "234": {"age": 22, "gender": "F", "accent": "Scottish"} | |
| # Add more as needed | |
| } | |
| # Pre-format speaker dropdown choices | |
| SPEAKER_CHOICES = [ | |
| (sid, f"p{sid} ({data['gender']}, {data['accent']}, {data['age']} yrs)") | |
| for sid, data in SPEAKER_METADATA.items() | |
| ] | |
| # Model cache | |
| MODEL_CACHE = {} | |
| def load_tts_model(model_key): | |
| if model_key in MODEL_CACHE: | |
| return MODEL_CACHE[model_key] | |
| model_info = VOICE_MODELS[model_key] | |
| tts = TTS(model_name=model_info["model_name"], gpu=False) | |
| MODEL_CACHE[model_key] = tts | |
| return tts | |
| def extract_text_from_docx(file): | |
| doc = docx.Document(file) | |
| return "\n".join([para.text for para in doc.paragraphs if para.text.strip()]) | |
| def generate_audio(voice_key, speaker_id, docx_file): | |
| text = extract_text_from_docx(docx_file) | |
| tts = load_tts_model(voice_key) | |
| kwargs = {} | |
| if VOICE_MODELS[voice_key]["multi_speaker"]: | |
| kwargs["speaker"] = speaker_id | |
| output_path = tempfile.mktemp(suffix=".wav") | |
| tts.tts_to_file(text=text, file_path=output_path, **kwargs) | |
| return output_path | |
| def update_speaker_visibility(voice_key): | |
| visible = VOICE_MODELS[voice_key]["multi_speaker"] | |
| return gr.update(visible=visible) | |
| with gr.Blocks() as demo: | |
| gr.Markdown("## DOCX to Speech with Speaker Selection") | |
| with gr.Row(): | |
| voice_dropdown = gr.Dropdown( | |
| choices=list(VOICE_MODELS.keys()), | |
| value="LJSpeech (Standard Female)", | |
| label="Select Voice" | |
| ) | |
| speaker_dropdown = gr.Dropdown( | |
| choices=SPEAKER_CHOICES, | |
| label="Select Speaker", | |
| visible=False | |
| ) | |
| docx_input = gr.File(label="Upload .docx File", file_types=[".docx"]) | |
| generate_btn = gr.Button("Generate Audio") | |
| audio_output = gr.Audio(label="Output Audio") | |
| voice_dropdown.change(fn=update_speaker_visibility, inputs=voice_dropdown, outputs=speaker_dropdown) | |
| generate_btn.click( | |
| fn=generate_audio, | |
| inputs=[voice_dropdown, speaker_dropdown, docx_input], | |
| outputs=audio_output | |
| ) | |
| demo.launch() | |