app.py

help me analyse this code, it is for a tts hugginface space
import spaces
import gradio as gr
import edge_tts
import asyncio
import tempfile
import os
import re
from pathlib import Path
from pydub import AudioSegment

def get_silence(duration_ms=1000):
    # Create silent audio segment with specified parameters
    silent_audio = AudioSegment.silent(
        duration=duration_ms,
        frame_rate=24000  # 24kHz sampling rate
    )

    # Set audio parameters
    silent_audio = silent_audio.set_channels(1)  # Mono
    silent_audio = silent_audio.set_sample_width(4)  # 32-bit (4 bytes per sample)

    with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmp_file:
        # Export with specific bitrate and codec parameters
        silent_audio.export(
            tmp_file.name,
            format="mp3",
            bitrate="48k",
            parameters=[
                "-ac", "1",  # Mono
                "-ar", "24000",  # Sample rate
                "-sample_fmt", "s32",  # 32-bit samples
                "-codec:a", "libmp3lame"  # MP3 codec
            ]
        )
        return tmp_file.name

# Get all available voices
async def get_voices():
    voices = await edge_tts.list_voices()
    return {f"{v['ShortName']} - {v['Locale']} ({v['Gender']})": v['ShortName'] for v in voices}

async def process_transcript_line(line, voice, rate, pitch):
    """Processes a single transcript line to extract time, voice commands, and generate audio."""
    match = re.match(r'(\d+):(\d+)(?:\.(\d+))?\s+(.*)', line)
    if match:
        minutes, seconds, milliseconds_str, text_with_commands = match.groups()
        start_time_ms = int(minutes) * 60000 + int(seconds) * 1000 + (int(milliseconds_str) * 10 if milliseconds_str else 0)
        if not text_with_commands.strip():
            return start_time_ms, None

        current_voice = voice
        current_rate = rate
        current_pitch = pitch
        processed_text = text_with_commands

        voice1 = "en-AU-WilliamNeural - en-AU (Male)"
        voice1F ="en-GB-SoniaNeural - en-GB (Female)"
        voice2 = "en-GB-RyanNeural - en-GB (Male)"
        voice2F = "en-US-JennyNeural - en-US (Female)"
        voice3 ="en-US-BrianMultilingualNeural - en-US (Male)"  #good for reading
        voice3F = "en-HK-YanNeural - en-HK (Female)"
        voice4 = "en-GB-ThomasNeural - en-GB (Male)"
        voice4F ="en-US-EmmaNeural - en-US (Female)"
        voice5 = "en-GB-RyanNeural - en-GB (Male)" #Old Man
        voice6 = "en-GB-MaisieNeural - en-GB (Female)"  #Child

        if text_with_commands.startswith("1F"):
            current_voice = voice1F.split(" - ")[0]
            current_pitch = 25
            processed_text = text_with_commands[2:].strip()
        elif text_with_commands.startswith("2F"):
            current_voice = voice2F.split(" - ")[0]
            processed_text = text_with_commands[2:].strip()
        elif text_with_commands.startswith("3F"):
            current_voice = voice3F.split(" - ")[0]
            processed_text = text_with_commands[2:].strip()
        elif text_with_commands.startswith("4F"):
            current_voice = voice4F.split(" - ")[0]
            processed_text = text_with_commands[2:].strip()
        elif text_with_commands.startswith("1M"):
            current_voice = voice1.split(" - ")[0]
            processed_text = text_with_commands[2:].strip()
        elif text_with_commands.startswith("2M"):
            current_voice = voice2.split(" - ")[0]
            processed_text = text_with_commands[2:].strip()
        elif text_with_commands.startswith("3M"):
            current_voice = voice3.split(" - ")[0]
            processed_text = text_with_commands[2:].strip()
        elif text_with_commands.startswith("4M"):
            current_voice = voice4.split(" - ")[0]
            processed_text = text_with_commands[2:].strip()
        elif text_with_commands.startswith("1O"):  # Old man voice
            current_voice = voice5.split(" - ")[0]
            current_pitch = -20
            current_rate = -10
            processed_text = text_with_commands[2:].strip()
        elif text_with_commands.startswith("1C"):  #Child voice
            current_voice = voice6.split(" - ")[0]
            processed_text = text_with_commands[2:].strip()

        rate_str = f"{current_rate:+d}%"
        pitch_str = f"{current_pitch:+d}Hz"
        communicate = edge_tts.Communicate(processed_text, current_voice, rate=rate_str, pitch=pitch_str)
        with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmp_file:
            audio_path = tmp_file.name
            await communicate.save(audio_path)
        return start_time_ms, audio_path
    return None, None

async def transcript_to_speech(transcript_text, voice, rate, pitch):
    if not transcript_text.strip():
        return None, gr.Warning("Please enter transcript text.")
    if not voice:
        return None, gr.Warning("Please select a voice.")

    lines = transcript_text.strip().split('\n')
    audio_segments_with_time = []
    max_end_time_ms = 0

    for line in lines:
        start_time, audio_path = await process_transcript_line(line, voice, rate, pitch)
        if start_time is not None and audio_path:
            audio = AudioSegment.from_mp3(audio_path)
            audio_segments_with_time.append({'start': start_time, 'audio': audio, 'path': audio_path})
            max_end_time_ms = max(max_end_time_ms, start_time + len(audio))
        elif audio_path:
            os.remove(audio_path) # Clean up even if no timestamp

    if not audio_segments_with_time:
        return None, "No valid transcript lines found."

    # Create initial silence audio
    final_audio = AudioSegment.silent(duration=max_end_time_ms, frame_rate=24000)

    for segment in audio_segments_with_time:
        final_audio = final_audio.overlay(segment['audio'], position=segment['start'])
        os.remove(segment['path']) # Clean up individual audio files

    combined_audio_path = tempfile.mktemp(suffix=".mp3")
    final_audio.export(combined_audio_path, format="mp3")
    return combined_audio_path, None

@spaces.GPU
def tts_interface(transcript, voice, rate, pitch):
    audio, warning = asyncio.run(transcript_to_speech(transcript, voice, rate, pitch))
    return audio, warning

async def create_demo():
    voices = await get_voices()
    default_voice = "en-US-AndrewMultilingualNeural - en-US (Male)"
    description = """
    Process YouTube transcript text with timestamps to generate synchronized audio.
    Each line should be in the format: `minutes:seconds[.milliseconds] text`.
    Voice prefixes (e.g., 1F, 1C) can be used at the beginning of a line to switch voices.
    Example:
    ```
    0:00 This
    0:14 is the story of little Red Riding Hood
    0:38 1F Grandma isn’t feeling very well.
    0:48 1C Yes, said Little Red Riding Hood.
    ```
    """
    demo = gr.Interface(
        fn=tts_interface,
        inputs=[
            gr.Textbox(label="YouTube Transcript", lines=10, placeholder="0:00 This\n0:14 is the story...\n0:38 1F Grandma..."),
            gr.Dropdown(choices=[""] + list(voices.keys()), label="Select Voice", value=default_voice),
            gr.Slider(minimum=-50, maximum=50, value=0, label="Speech Rate Adjustment (%)", step=1),
            gr.Slider(minimum=-50, maximum=50, value=0, label="Pitch Adjustment (Hz)", step=1)
        ],
        outputs=[
            gr.Audio(label="Generated Audio", type="filepath"),
            gr.Markdown(label="Warning", visible=False)
        ],
        title="TTS for YouTube Transcripts with Voice Switching",
        description=description,
        analytics_enabled=False,
        allow_flagging=False
    )
    return demo

if __name__ == "__main__":
    demo = asyncio.run(create_demo())
    demo.launch()