app.py

import gradio as gr
import numpy as np
import torch
import soundfile as sf
from kokoro import KPipeline
import re
import traceback

# Helper: Format seconds into SRT timestamp (hh:mm:ss,ms)
def format_time(seconds):
    hours = int(seconds // 3600)
    minutes = int((seconds % 3600) // 60)
    secs = seconds % 60
    # Ensure milliseconds are comma separated
    return f"{hours:02d}:{minutes:02d}:{secs:06.3f}".replace('.', ',')

def generate_audio(text, voice, speed, lang_code, split_pattern, debug):
    debug_logs = []
    debug_logs.append("Starting Kokoro TTS generation...")
    
    try:
        debug_logs.append(f"Initializing Kokoro pipeline with lang_code: '{lang_code}' (CPU mode assumed)")
        # Initialize the pipeline; by default, it will run on CPU if no GPU is available.
        pipeline = KPipeline(lang_code=lang_code)
        debug_logs.append("Pipeline initialized successfully.")
    except Exception as e:
        error_msg = f"Error initializing pipeline: {str(e)}"
        debug_logs.append(error_msg)
        return None, "", "\n".join(debug_logs)
    
    # Prepare lists for audio segments, SRT entries, and segment-level debug info.
    audio_segments = []
    srt_entries = []
    current_time = 0.0  # cumulative time for SRT timestamps
    segment_index = 1
    segment_debug_info = []
    
    try:
        debug_logs.append("Generating audio segments from input text...")
        # Invoke the pipeline to process the text.
        # The split_pattern parameter (regex) allows you to define how text is segmented.
        generator = pipeline(
            text,
            voice=voice,
            speed=speed,
            split_pattern=split_pattern
        )
        
        for i, (gs, ps, audio) in enumerate(generator):
            duration = len(audio) / 24000.0  # assuming a sample rate of 24000 Hz
            start_timestamp = current_time
            end_timestamp = current_time + duration
            # Create an SRT entry for the segment.
            srt_entry = f"{segment_index}\n{format_time(start_timestamp)} --> {format_time(end_timestamp)}\n{gs}\n"
            srt_entries.append(srt_entry)
            current_time = end_timestamp
            
            # Record segment details for debugging.
            segment_debug_info.append(f"Segment {segment_index}: Duration = {duration:.3f}s, Graphemes = {gs}, Phonemes = {ps}")
            audio_segments.append(audio)
            segment_index += 1
        
        debug_logs.append("Audio segments generated successfully.")
    except Exception as e:
        error_msg = f"Error during audio generation: {str(e)}\n{traceback.format_exc()}"
        debug_logs.append(error_msg)
        return None, "", "\n".join(debug_logs)
    
    # Concatenate all the generated segments into a single audio array.
    if audio_segments:
        full_audio = np.concatenate(audio_segments)
    else:
        debug_logs.append("No audio segments were generated.")
        return None, "", "\n".join(debug_logs)
    
    # Combine all SRT entries into one string.
    srt_content = "\n".join(srt_entries)
    
    # Combine all debug logs (with optional segment details).
    if debug:
        debug_info = "\n".join(debug_logs + segment_debug_info)
    else:
        debug_info = "\n".join(debug_logs)
    
    # Return a tuple: audio (with sample rate), the SRT text, and the debug log.
    return (24000, full_audio), srt_content, debug_info

# Build the Gradio interface.
iface = gr.Interface(
    fn=generate_audio,
    inputs=[
        gr.Textbox(label="Input Text", lines=10, placeholder="Enter the text to be synthesized here..."),
        gr.Textbox(label="Voice (e.g., af_heart)", value="af_heart"),
        gr.Slider(label="Speed", minimum=0.5, maximum=2.0, step=0.1, value=1.0),
        gr.Textbox(label="Language Code", value="a", 
                   placeholder="Enter language code ('a' for American English, 'b' for British, etc.)"),
        gr.Textbox(label="Split Pattern (Regex)", value=r'\n+',
                   placeholder="Regex to split the input text (e.g., '\\n+')"),
        gr.Checkbox(label="Enable Debug Mode", value=True)
    ],
    outputs=[
        gr.Audio(label="Generated Audio", type="numpy"),
        gr.Textbox(label="Generated SRT"),
        gr.Textbox(label="Debug Information", lines=15)
    ],
    title="Kokoro TTS Gradio App (CPU Mode)",
    description=("This app uses the Kokoro TTS model to generate audio from text. "
                 "You can tweak parameters such as voice, speed, language code, and the text split pattern. "
                 "When debug mode is enabled, detailed processing steps (including grapheme and phoneme outputs) are displayed.")
)

if __name__ == "__main__":
    iface.launch()