# app.py
import streamlit as st
import time
import re
import os
import tempfile
import pypdf
from pydub import AudioSegment, effects
import difflib
from utils import (
generate_script,
generate_audio_mp3,
truncate_text,
extract_text_from_url,
transcribe_youtube_video,
research_topic,
mix_with_bg_music,
DialogueItem
)
from prompts import SYSTEM_PROMPT
# NEW: For Q&A
from qa import transcribe_audio_deepgram, handle_qa_exchange
MAX_QA_QUESTIONS = 5 # up to 5 voice/text questions
def parse_user_edited_transcript(edited_text: str, host_name: str, guest_name: str):
pattern = r"\*\*(.+?)\*\*:\s*(.+)"
matches = re.findall(pattern, edited_text)
items = []
if not matches:
raw_name = host_name or "Jane"
text_line = edited_text.strip()
speaker = "Jane"
if raw_name.lower() == guest_name.lower():
speaker = "John"
item = DialogueItem(
speaker=speaker,
display_speaker=raw_name,
text=text_line
)
items.append(item)
return items
for (raw_name, text_line) in matches:
if raw_name.lower() == host_name.lower():
speaker = "Jane"
elif raw_name.lower() == guest_name.lower():
speaker = "John"
else:
speaker = "Jane"
item = DialogueItem(
speaker=speaker,
display_speaker=raw_name,
text=text_line
)
items.append(item)
return items
def regenerate_audio_from_dialogue(dialogue_items, custom_bg_music_path=None):
audio_segments = []
transcript = ""
crossfade_duration = 50 # ms
for item in dialogue_items:
audio_file = generate_audio_mp3(item.text, item.speaker)
seg = AudioSegment.from_file(audio_file, format="mp3")
audio_segments.append(seg)
transcript += f"**{item.display_speaker}**: {item.text}\n\n"
os.remove(audio_file)
if not audio_segments:
return None, "No audio segments were generated."
combined_spoken = audio_segments[0]
for seg in audio_segments[1:]:
combined_spoken = combined_spoken.append(seg, crossfade=crossfade_duration)
final_mix = mix_with_bg_music(combined_spoken, custom_bg_music_path)
with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as temp_audio:
final_mix.export(temp_audio.name, format="mp3")
final_mp3_path = temp_audio.name
with open(final_mp3_path, "rb") as f:
audio_bytes = f.read()
os.remove(final_mp3_path)
return audio_bytes, transcript
def generate_podcast(
file,
url,
video_url,
research_topic_input,
tone,
length_minutes,
host_name,
host_desc,
guest_name,
guest_desc,
user_specs,
sponsor_content,
sponsor_style,
custom_bg_music_path
):
sources = [bool(file), bool(url), bool(video_url), bool(research_topic_input)]
if sum(sources) > 1:
return None, "Provide only one input (PDF, URL, YouTube, or Topic)."
if not any(sources):
return None, "Please provide at least one source."
text = ""
if file:
try:
if not file.name.lower().endswith('.pdf'):
return None, "Please upload a PDF file."
reader = pypdf.PdfReader(file)
text = " ".join(page.extract_text() for page in reader.pages if page.extract_text())
except Exception as e:
return None, f"Error reading PDF: {str(e)}"
elif url:
try:
text = extract_text_from_url(url)
if not text:
return None, "Failed to extract text from URL."
except Exception as e:
return None, f"Error extracting text from URL: {str(e)}"
elif video_url:
try:
text = transcribe_youtube_video(video_url)
if not text:
return None, "Failed to transcribe YouTube video."
except Exception as e:
return None, f"Error transcribing YouTube video: {str(e)}"
elif research_topic_input:
try:
text = research_topic(research_topic_input)
if not text:
return None, f"Sorry, no information found on '{research_topic_input}'."
except Exception as e:
return None, f"Error researching topic: {str(e)}"
from utils import truncate_text
text = truncate_text(text)
extra_instructions = []
if host_name or guest_name:
host_line = f"Host: {host_name or 'Jane'} - {host_desc or 'a curious host'}."
guest_line = f"Guest: {guest_name or 'John'} - {guest_desc or 'an expert'}."
extra_instructions.append(f"{host_line}\n{guest_line}")
if user_specs.strip():
extra_instructions.append(f"Additional User Instructions: {user_specs}")
if sponsor_content.strip():
extra_instructions.append(
f"Sponsor Content Provided (should be under ~30 seconds):\n{sponsor_content}"
)
from prompts import SYSTEM_PROMPT
combined_instructions = "\n\n".join(extra_instructions).strip()
full_prompt = SYSTEM_PROMPT
if combined_instructions:
full_prompt += f"\n\n# Additional Instructions\n{combined_instructions}\n"
from utils import generate_script, generate_audio_mp3, mix_with_bg_music
try:
script = generate_script(
full_prompt,
text,
tone,
f"{length_minutes} Mins",
host_name=host_name or "Jane",
guest_name=guest_name or "John",
sponsor_style=sponsor_style,
sponsor_provided=bool(sponsor_content.strip())
)
except Exception as e:
return None, f"Error generating script: {str(e)}"
audio_segments = []
transcript = ""
crossfade_duration = 50
try:
for item in script.dialogue:
audio_file = generate_audio_mp3(item.text, item.speaker)
seg = AudioSegment.from_file(audio_file, format="mp3")
audio_segments.append(seg)
transcript += f"**{item.display_speaker}**: {item.text}\n\n"
os.remove(audio_file)
if not audio_segments:
return None, "No audio segments generated."
combined_spoken = audio_segments[0]
for seg in audio_segments[1:]:
combined_spoken = combined_spoken.append(seg, crossfade=crossfade_duration)
final_mix = mix_with_bg_music(combined_spoken, custom_bg_music_path)
with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as temp_audio:
final_mix.export(temp_audio.name, format="mp3")
final_mp3_path = temp_audio.name
with open(final_mp3_path, "rb") as f:
audio_bytes = f.read()
os.remove(final_mp3_path)
return audio_bytes, transcript
except Exception as e:
return None, f"Error generating audio: {str(e)}"
def highlight_differences(original: str, edited: str) -> str:
matcher = difflib.SequenceMatcher(None, original.split(), edited.split())
highlighted = []
for opcode, i1, i2, j1, j2 in matcher.get_opcodes():
if opcode == 'equal':
highlighted.extend(original.split()[i1:i2])
elif opcode in ('replace', 'insert'):
added_words = edited.split()[j1:j2]
highlighted.extend([f'{word}' for word in added_words])
elif opcode == 'delete':
pass
return ' '.join(highlighted)
def main():
st.set_page_config(
page_title="MyPod v2: AI-Powered Podcast Magic",
layout="centered"
)
# Inject custom CSS for styling adjustments
st.markdown("""
""", unsafe_allow_html=True)
logo_col, title_col = st.columns([1, 10])
with logo_col:
st.image("logomypod.jpg", width=70) # Increased size by ~15%
with title_col:
st.markdown("## MyPod v2: AI-Powered Podcast Magic")
# Reinstated welcome section
st.markdown("""
Welcome to **MyPod**, your go-to AI-powered podcast generator! 🎉
MyPod transforms your documents, webpages, YouTube videos, or research topics into a more human-sounding, conversational podcast.
Select a tone and a duration range. The output script will be on-topic, concise, and respect your chosen length.
""")
# "How to Use" as an expander with enumerated list and larger text
with st.expander("How to Use"):
st.markdown("""
Provide one source: PDF Files, Website URL, YouTube videos, or a Topic to Research.
Choose the tone and the target duration.
Add custom names and descriptions for the speakers if you wish.
Add sponsored content as a separate break or blended into the script.
Click 'Generate Podcast' to produce your podcast.
Post generation you can edit the transcript and re-generate the audio with your edits if needed.
Ask Follow-up Questions via text or voice and get immediate answers.
""", unsafe_allow_html=True)
# Retained text below "How to Use"
st.markdown("""
**Research a Topic:** If it's too niche or specific, you might not get the desired outcome.
**Token Limit:** Up to ~2,048 tokens are supported. Long inputs may be truncated.
**Note:** YouTube videos will only work if they have captions built in.
⏳**Please be patient while your podcast is being generated.** This process involves content analysis, script creation, and high-quality audio synthesis, which may take a few minutes.
🔥 **Ready to create your personalized podcast?** Give it a try now and let the magic happen! 🔥
""")
# Original placement of input options
col1, col2 = st.columns(2)
with col1:
file = st.file_uploader("Upload File (.pdf only)", type=["pdf"])
url = st.text_input("Or Enter Website URL")
video_url = st.text_input("Or Enter YouTube Link (Captioned videos)")
with col2:
research_topic_input = st.text_input("Or Research a Topic")
tone = st.radio("Tone", ["Casual", "Formal", "Humorous", "Youthful"], index=0)
length_minutes = st.slider("Podcast Length (in minutes)", 1, 60, 3)
st.markdown("### Customize Your Podcast (New Features)")
with st.expander("Set Host & Guest Names/Descriptions (Optional)"):
host_name = st.text_input("Female Host Name (leave blank for 'Jane')")
host_desc = st.text_input("Female Host Description (Optional)")
guest_name = st.text_input("Male Guest Name (leave blank for 'John')")
guest_desc = st.text_input("Male Guest Description (Optional)")
user_specs = st.text_area("Any special instructions or prompts for the script? (Optional)", "")
sponsor_content = st.text_area("Sponsored Content / Ad (Optional)", "")
# Removed bold heading for Sponsor Integration Style
sponsor_style = st.selectbox(
"Sponsor Integration Style",
["Separate Break", "Blended"]
)
custom_bg_music_file = st.file_uploader("Upload Custom Background Music (Optional)", type=["mp3", "wav"])
custom_bg_music_path = None
if custom_bg_music_file:
with tempfile.NamedTemporaryFile(delete=False, suffix=os.path.splitext(custom_bg_music_file.name)[1]) as tmp:
tmp.write(custom_bg_music_file.read())
custom_bg_music_path = tmp.name
if "audio_bytes" not in st.session_state:
st.session_state["audio_bytes"] = None
if "transcript" not in st.session_state:
st.session_state["transcript"] = None
if "transcript_original" not in st.session_state:
st.session_state["transcript_original"] = None
# For Q&A
if "qa_count" not in st.session_state:
st.session_state["qa_count"] = 0
if "conversation_history" not in st.session_state:
st.session_state["conversation_history"] = ""
generate_button = st.button("Generate Podcast")
if generate_button:
progress_bar = st.progress(0)
progress_text = st.empty()
progress_messages = [
"🔍 Analyzing your input...",
"📝 Crafting the perfect script...",
"🎙️ Generating high-quality audio...",
"🎶 Adding the finishing touches..."
]
progress_text.write(progress_messages[0])
progress_bar.progress(0)
time.sleep(1.0)
progress_text.write(progress_messages[1])
progress_bar.progress(25)
time.sleep(1.0)
progress_text.write(progress_messages[2])
progress_bar.progress(50)
time.sleep(1.0)
progress_text.write(progress_messages[3])
progress_bar.progress(75)
time.sleep(1.0)
audio_bytes, transcript = generate_podcast(
file,
url,
video_url,
research_topic_input,
tone,
length_minutes,
host_name,
host_desc,
guest_name,
guest_desc,
user_specs,
sponsor_content,
sponsor_style,
custom_bg_music_path
)
progress_bar.progress(100)
progress_text.write("✅ Done!")
if audio_bytes is None:
st.error(transcript)
st.session_state["audio_bytes"] = None
st.session_state["transcript"] = None
st.session_state["transcript_original"] = None
else:
st.success("Podcast generated successfully!")
st.session_state["audio_bytes"] = audio_bytes
st.session_state["transcript"] = transcript
st.session_state["transcript_original"] = transcript
# Reset Q&A
st.session_state["qa_count"] = 0
st.session_state["conversation_history"] = ""
if st.session_state["audio_bytes"]:
st.audio(st.session_state["audio_bytes"], format='audio/mp3')
st.download_button(
label="Download Podcast (MP3)",
data=st.session_state["audio_bytes"],
file_name="my_podcast.mp3",
mime="audio/mpeg"
)
st.markdown("### Generated Transcript (Editable)")
edited_text = st.text_area(
"Feel free to tweak lines, fix errors, or reword anything.",
value=st.session_state["transcript"],
height=300
)
if st.session_state["transcript_original"]:
highlighted_transcript = highlight_differences(
st.session_state["transcript_original"],
edited_text
)
st.markdown("### **Edited Transcript Highlights**", unsafe_allow_html=True)
st.markdown(highlighted_transcript, unsafe_allow_html=True)
if st.button("Regenerate Audio From Edited Text"):
regen_bar = st.progress(0)
regen_text = st.empty()
regen_text.write("🔄 Regenerating your podcast with the edits...")
regen_bar.progress(25)
time.sleep(1.0)
regen_text.write("🔧 Adjusting the script based on your changes...")
regen_bar.progress(50)
time.sleep(1.0)
dialogue_items = parse_user_edited_transcript(
edited_text,
host_name or "Jane",
guest_name or "John"
)
new_audio_bytes, new_transcript = regenerate_audio_from_dialogue(dialogue_items, custom_bg_music_path)
regen_bar.progress(75)
time.sleep(1.0)
if new_audio_bytes is None:
regen_bar.progress(100)
st.error(new_transcript)
else:
regen_bar.progress(100)
regen_text.write("✅ Regeneration complete!")
st.success("Regenerated audio below:")
st.session_state["audio_bytes"] = new_audio_bytes
st.session_state["transcript"] = new_transcript
st.session_state["transcript_original"] = new_transcript
st.audio(new_audio_bytes, format='audio/mp3')
st.download_button(
label="Download Edited Podcast (MP3)",
data=new_audio_bytes,
file_name="my_podcast_edited.mp3",
mime="audio/mpeg"
)
st.markdown("### Updated Transcript")
st.markdown(new_transcript)
# -----------------------
# POST-PODCAST Q&A Logic
# -----------------------
st.markdown("## Post-Podcast Q&A")
used_questions = st.session_state["qa_count"]
remaining = MAX_QA_QUESTIONS - used_questions
if remaining > 0:
st.write(f"You can ask up to {remaining} more question(s).")
typed_q = st.text_input("Type your follow-up question:")
audio_q = st.audio_input("Or record an audio question (WAV)")
if st.button("Submit Q&A"):
if used_questions >= MAX_QA_QUESTIONS:
st.warning("You have reached the Q&A limit.")
else:
question_text = typed_q.strip()
if audio_q is not None:
suffix = ".wav"
with tempfile.NamedTemporaryFile(delete=False, suffix=suffix) as tmp:
tmp.write(audio_q.read())
local_audio_path = tmp.name
st.write("Transcribing your audio question...")
audio_transcript = transcribe_audio_deepgram(local_audio_path)
if audio_transcript:
question_text = audio_transcript
if not question_text:
st.warning("No question found (text or audio).")
else:
st.write("Generating an answer...")
ans_audio, ans_text = handle_qa_exchange(question_text)
if ans_audio:
st.audio(ans_audio, format="audio/mp3")
st.markdown(f"**John**: {ans_text}")
st.session_state["qa_count"] += 1
else:
st.warning("No response could be generated.")
else:
st.write("You have used all 5 Q&A opportunities.")
# Footer with updated text
st.markdown("", unsafe_allow_html=True)
if __name__ == "__main__":
main()