From f0798dfb6e61eb98342040d607181cc57fd89b59 Mon Sep 17 00:00:00 2001
From: daavoo <daviddelaiglesiacastro@gmail.com>
Date: Fri, 3 Jan 2025 12:38:53 +0100
Subject: [PATCH] Add random-lenght silences.

Insert silences of lenght between 0.1 and 1.1 seconds between each speaker segment.
---
 demo/app.py                      | 6 ++++--
 src/document_to_podcast/cli.py   | 4 ++--
 src/document_to_podcast/utils.py | 9 +++++++++
 3 files changed, 15 insertions(+), 4 deletions(-)
 create mode 100644 src/document_to_podcast/utils.py

diff --git a/demo/app.py b/demo/app.py
index f3f904c..a360a13 100644
--- a/demo/app.py
+++ b/demo/app.py
@@ -1,7 +1,6 @@
 import re
 from pathlib import Path
 
-import numpy as np
 import soundfile as sf
 import streamlit as st
 
@@ -13,6 +12,7 @@
 from document_to_podcast.config import DEFAULT_PROMPT, DEFAULT_SPEAKERS, Speaker
 from document_to_podcast.inference.text_to_speech import text_to_speech
 from document_to_podcast.inference.text_to_text import text_to_text_stream
+from document_to_podcast.utils import stack_audio_segments
 
 
 @st.cache_resource
@@ -160,7 +160,9 @@ def gen_button_clicked():
 
     if st.session_state[gen_button]:
         if st.button("Save Podcast to audio file"):
-            st.session_state.audio = np.concatenate(st.session_state.audio)
+            st.session_state.audio = stack_audio_segments(
+                st.session_state.audio, speech_model.audio_codec.sr
+            )
             sf.write(
                 "podcast.wav",
                 st.session_state.audio,
diff --git a/src/document_to_podcast/cli.py b/src/document_to_podcast/cli.py
index f71e1cb..c056570 100644
--- a/src/document_to_podcast/cli.py
+++ b/src/document_to_podcast/cli.py
@@ -1,7 +1,6 @@
 import re
 from pathlib import Path
 
-import numpy as np
 import soundfile as sf
 import yaml
 from fire import Fire
@@ -22,6 +21,7 @@
 from document_to_podcast.inference.text_to_text import text_to_text_stream
 from document_to_podcast.inference.text_to_speech import text_to_speech
 from document_to_podcast.preprocessing import DATA_CLEANERS, DATA_LOADERS
+from document_to_podcast.utils import stack_audio_segments
 
 
 @logger.catch(reraise=True)
@@ -158,7 +158,7 @@ def document_to_podcast(
     logger.info("Saving Podcast...")
     sf.write(
         str(output_folder / "podcast.wav"),
-        np.concatenate(podcast_audio),
+        stack_audio_segments(podcast_audio),
         samplerate=sample_rate,
     )
     (output_folder / "podcast.txt").write_text(podcast_script)
diff --git a/src/document_to_podcast/utils.py b/src/document_to_podcast/utils.py
new file mode 100644
index 0000000..2695c7b
--- /dev/null
+++ b/src/document_to_podcast/utils.py
@@ -0,0 +1,9 @@
+import numpy as np
+
+
+def stack_audio_segments(audio_segments, sample_rate):
+    stacked = []
+    for segment in audio_segments:
+        stacked.append(segment)
+        stacked.append(np.zeros(int((0.1 + np.random.rand()) * sample_rate)))
+    return np.concatenate(stacked)